A BIOMECHANICAL ANALYSIS OF SHOULDER LOADING AND EFFORT DURING LOAD TRANSFER TASKS
by
Clark Rutherford Dickerson
A dissertation submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy (Biomedical Engineering) in The University of Michigan 2005
Doctoral Committee: Professor Don B. Chaffin, Co-Chairman Associate Professor Richard E. Hughes, Co-Chairman Associate Professor Bernard J. Martin Research Professor James A. Ashton-Miller
©
Clark Rutherford Dickerson 2005 All Rights Reserved
To My Wife Marnie Mae Without you, this journey may have been shorter. But it would have seemed so much longer. Thank you for everything.
ii
ACKNOWLEDGMENTS
I have worked with many remarkable people who deserve recognition and thanks. Firstly, I cannot overstate the influence of Don Chaffin on my progress and the quality of my work. With timely suggestions and constructive criticisms, he has enabled me to gain a perspective of the importance of musculoskeletal modeling in ergonomics, and the opportunity to study biomechanics in an excellent research environment. Richard Hughes has been instrumental in my development as a critical thinker. Beginning with qualification exams and continuing through the interpretation of the shoulder model, he has been unequivicolly supportive, directive, and respectful of my ideas and technical abilities. His enthusiasm and knowledge were indispensable in enabling me to maintain perspective and a rigorous approach to research. Bernard Martin and James Ashton-Miller were also very supportive in the completion of this work, both by offering specific suggestions as well as through discussions of the novelty and motivations of the research project. Their unique perspectives and feedback forced me to think about my reseach in critical ways. Fellow researchers and students have contributed greatly to my academic development, specifically, Chuck Woolley, Jim Foulke, Matt Reed, Woojin Park, Kyunghan Kim, Matthew Parkinson, Kevin Rider, and David Wagner. I wish success to Suzanne Hoffman and Deepti Sood who will continue HUMOSIM shoulder research. Finally, I would like to thank family and friends, from Maryland to Mississauga and everywhere in between, for their support. Their encouragement, patience, and positive attitudes have given me inspiration during an intense phase in my life. This work would be impossible without all of you.
iii
TABLE OF CONTENTS DEDICATION................................................................................................................... ii ACKNOWLEDGMENTS ............................................................................................... iii LIST OF FIGURES ......................................................................................................... vi LIST OF TABLES ......................................................................................................... viii LIST OF APPENDICES ................................................................................................. ix CHAPTER I. INTRODUCTION................................................................................................1 1.1 1.2 1.3 1.4 1.5 1.6 II.
Applied Problem .........................................................................................1 Theoretical Problem....................................................................................4 Thesis Statement .........................................................................................6 Research Objectives....................................................................................6 Thesis Organization ....................................................................................7 References...................................................................................................9
QUANTIFICATION OF SHOULDER LOADING AND ITS RELATIONSHIP TO THE PERCEPTION OF EFFORT – I. SHOULDER TORQUES ..........................................................................12 2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8
Abstract .....................................................................................................12 Introduction...............................................................................................13 Methods.....................................................................................................15 A Biomechanical Torque and Effort Model of the Shoulder....................22 Results.......................................................................................................30 Discussion .................................................................................................34 Conclusions...............................................................................................40 References.................................................................................................42
III. A BIOMECHANICAL SHOULDER MODEL FOR ERGONOMIC ANALYSIS – I. MODEL DESCRIPTION.......................45 3.1 3.2 3.3 3.4
Introduction...............................................................................................45 Biomechanical Model Description ...........................................................47 Model Outputs ..........................................................................................75 Discussion .................................................................................................79
iv
3.5 References.................................................................................................90 IV. A BIOMECHANICAL SHOULDER MODEL FOR ERGONOMIC ANALYSIS – II. EMPIRICAL EVALUATION.............93 4.1 4.2 4.3 4.4 4.5 4.6 4.7 4.8 V.
Abstract .....................................................................................................93 Introduction...............................................................................................94 Methods.....................................................................................................96 Data Analysis ..........................................................................................104 Results.....................................................................................................113 Discussion ...............................................................................................122 Conclusions.............................................................................................133 References...............................................................................................135
QUANTIFICATION OF SHOULDER LOADING AND ITS RELATIONSHIP TO THE PERCEPTION OF EFFORT – II. SHOULDER MUSCLE FORCE PREDICTIONS .............................138 5.1 5.2 5.3 5.4 5.5 5.5 5.6
Abstract ...................................................................................................138 Introduction.............................................................................................139 Methods...................................................................................................141 Results.....................................................................................................148 Discussion ...............................................................................................153 Conclusions.............................................................................................157 References...............................................................................................161
VI. CONCLUSIONS ..........................................................................................163 6.1 6.2 6.3 6.4 6.4
Overview.................................................................................................163 Principal Research Contributions ...........................................................164 Future Research Directions.....................................................................168 Summary .................................................................................................174 References...............................................................................................175
APPENDICES ................................................................................................................177
v
LIST OF FIGURES
FIGURE CHAPTER II 2.1. Strength Testing Postures for Hand Load and Effort Rating Calibrations .........16 2.2
Psychophysical Effort Rating Scale....................................................................17
2.3
Marker Positioning During Experimental Trials ................................................18
2.4
Representation of an Experimental Loaded Reach Task. ...................................20
2.5
Spatial Distribution of Target Locations.............................................................21
2.6
Local Coordinate Systems ..................................................................................24
2.7
Normalized Resultant Torque Profile for a Representative Reaching Task. ......31
2.8
A Theoretical Perspective of Effort Perception in the Shoulder.........................40
CHAPTER III 3.1
Data flow through the components of the shoulder model .................................48
3.2
Local coordiante systems for the segments of the shoulder model ....................51
3.3
Spherical Muscle Wrapping................................................................................58
3.4
View of spherical muscle wrapping in the geometric model..............................59
3.5 Graphical representation of cylindrical muscle wrapping ..................................61 3.6
Cylindrical wrapping of the serratus anterior .....................................................62
3.7
Segmental Coordinate Systems...........................................................................66
3.8 Directional Shoulder Stability Ratios .................................................................72 3.9
Graphical Representation of the Right Shoulder Mechanism.............................76
3.10 Temporal Representation of Global External Dynamic Shoulder Torques ........77 3.11 Example muscle force predictions generated by the model for a forward reach with a 2.4 kg mass in the hand ..................................................................78 CHAPTER IV 4.1
Data flow through shoulder model components .................................................97
4.2
Several views of an experimental reach for a given condition are shown........100
vi
4.3
Tracked motion sensors for the experimental task ...........................................102
4.4
Trial Starting Position .......................................................................................106
4.5
Time-torque curve for a sample reach ..............................................................108
4.6
Muscle force prediction throughout a trial, for one muscle..............................114
4.7
Middle deltoid EMG activity during an experimental trial ..............................115
4.8
Muscle force predictions plotted against electromyographical data for the 9 muscles and muscle parts for all trials for one subject ..................................117
4.9
Effect of varying the stability multiplier, ϑ, on muscle force predictions........121
CHAPTER V 5.1 Rating Exertion Scale .......................................................................................144 5.2
Relative performance of effort prediction models ............................................151
5.3
An Integrative Model of Effort Perception.......................................................159
5.4
The Impact of Effort Regression Model Construction on Model Variance Explanation .......................................................................................................160
CHAPTER VI 6.1
Visual Interface Allowing Comparison of Posture and Muscle Force .............168
vii
LIST OF TABLES
TABLE CHAPTER II 2.1. Subject Summary Characteristics for Torque Effort Study ..........................15 2.2. Arm segment local coordinate system definitions used in the biomechanical model of the arm, in the defined neutral posture..................25 2.3. Parameters used in the statistical effort prediction models...........................29 CHAPTER III 3.1 3.2
Arm Segment Local Coordinate System Definitions....................................66 Construction of the Optimization Constraint Equations...............................74
CHAPTER IV 4.1 4.2 4.3 4.4 4.5 4.6 4.7
Summary subject characteristics...................................................................97 Experimental variation of reach task parameters..........................................99 Muscle sites monitored with EMG electrodes ............................................103 Concordance matrix, per muscle.................................................................110 Matrix for perfect concordance...................................................................110 Static correlation coefficients (r) for arm elevators, combined azimuths...116 Concordance matrix averaged over subjects...............................................118
CHAPTER V 5.1. 5.2 5.3 5.4 5.5
Sumary subject characteristics....................................................................141 Experimental variation of reach task parameters........................................143 Regression model component descriptions.................................................147 Specific task effects considered in multiple regression ..............................147 Specific EMG and Muscle Force Prediction Model Performance..............152
viii
LIST OF APPENDICES
APPENDIX A.
Individual subject muscle prediction correlations stratified by azimuth.....................................................................................................178
B.
EMG channel variability by subject across all recorded trials ................181
C.
Distribution of EMG levels during the static hold phase by subject and muscle in percent MVC. ...................................................................185
D.
Muscle force and electromyographic comparison by subject and muscle – grouped for all azimuths. ..........................................................193
E.
Dynamic concordance matrices, by subject.............................................201
F.
List of muscle forces predicted by the biomechanical shoulder model ...206
ix
CHAPTER I INTRODUCTION
1.1
Applied Problem Shoulder pain, injury and discomfort are public health and economic issues
worldwide. Due to the frequency of shoulder injuries, which are often attributed to repetitive strain in the workplace, efforts have been undertaken to evaluate the workrelatedness of shoulder musculoskeletal disorders (NIOSH, 1997). Incorporating indirect costs, including sick leave and workman’s compensation, estimated U.S. costs associated with occupational shoulder injuries are approximately $1-2 billion annually (Reynolds, 1999). This financial cost, which is compounded by the suffering of afflicted workers, prompts investigation of potential underlying causes of shoulder injuries. Load levels in shoulder tissues have been cited as a risk factor for development of musculoskeletal disorders (Herberts, 1984). This encourages the development of a representation of the loading profiles in the shoulder, both in work and daily living. Several workplace ergonomic evaluation tools exist to assess and limit exposure to physical stressors. However, these tools do not incorporate all key elements needed to effectively evaluate occupational tasks. Often, shoulder analysis relies on postural estimates made by direct observation of a worker performing a job. Some examples of these techniques include the Rapid Upper Limb Assessment [RULA] (McAtamney et al.,
1
1993), ranking of extreme postures (Latko et al, 1997), and application of planar models (Dul, 1988; Kuzkay et al, 1998). These models provide valuable information about shoulder exposure during work tasks, but they do not address the dynamic aspects of work. Further, they do not relate loading metrics to the effort that is perceived by a worker. Another limitation is that many do not allow detailed prospective analysis of work tasks and stations that are yet to be implemented. Computerized biomechanical models offer an alternative to heuristic, subjective analysis techniques. Three dimensional (3D) static models have been used to estimate shoulder loading, including the University of Michigan’s 3DSSPP (Chaffin, 1997). In addition, several research groups have investigated shoulder muscle force prediction modeling. Two major different biomechanical analysis approaches have been used. Optimization modeling (the “Gothenburg” and “Dutch” models) (Gothenburg: Hogfors et al., 1987,1991,1995; Karlsson et al., 1992; Makhsous et al., 1999; Dutch: van der Helm et al., 1991,1992,1994a,1994b,1995,1996; de Groot et al., 1997,1998,1999a,1999b; Pronk et al., 1989,1991; Veeger et al 1991a,1991b), and electromyography-based (EMG) estimated force models (the “Copenhagen” and “Utah” models) (Copenhagen: Laursen, 1998, 2000; Utah: Meek and Wood, 1989a,1989b) have been used to predict shoulder muscle forces for an array of loading situations. The former models implement optimization schemes and mathematical relationships to predict muscle forces, while the latter are based on relationships between recorded EMG values and muscle force. Adapting these models may bridge gaps between subjective and objective job analysis. In order for perception to relate to external loading (such as resultant shoulder moments), there must be a psychophysical mechanism by which the physical loading
2
produced by the performed task is converted into a quantifiable muscular effort perception (exertion rating). Modeling of this mechanism requires the use of a mathematical transformation. A statistical regression method that includes confounding task and personal factors is developed in this dissertation. One potential application of mathematical shoulder models is within digital human modeling (DHM) software. DHM is an emergent technology that offers promise in effective prospective human interface design. DHM allows ergonomists to anticipate aspects of the human/machine interface, enabling them to evaluate how a simulated human, or avatar, would interact with an environment. The ability to discern the effort associated with performing a physical task in the DHM environment has not yet been addressed. With biomechanical shoulder models incorporated into the DHM framework, it would be possible to provide feedback to the designer about the ability of a simulated worker to perform a task or motion as well as an indication of the psychophysical difficulty associated with the task. Compatibility with DHM software requires tools that are robust, flexible, and strictly validated. One challenge of the dissertation was to address the integration of the research with motion prediction algorithms developed in the HUMOSIM laboratory. While it is useful to investigate tasks that are performed in the laboratory, as well as existing industrial tasks, the greatest benefit that these models can provide is during the proactive planning stages of workplace and product design. In order to leverage the unique capabilities of the HUMOSIM database and motion simulation methods, specific criteria must be met.
Currently, HUMOSIM motion predictions can provide the inputs
3
for the developed shoulder model. For the shoulder effort module to be effective for design purposes, it was developed to meet the following requirements: •
Uses the same inputs as the HUMOSIM motion prediction module
•
Usable by practicing engineers without extensive biomechanical training
•
Estimates physical loading in general and in tissue-specific forms
•
Calculates an index of shoulder effort required by the task for rank ordering tasks
•
Able to compare and contrast several simulations to aid in job design
1.2
Theoretical Problem This dissertation investigates two issues concerning the evaluation of jobs with
respect to the shoulder: 1) the characterization of general and specific tissue loading in the shoulder, and 2) the perception of effort in the shoulder as it relates to general and specific shoulder loading.
1.2.1
Characterizing Loading in the Shoulder Many workplace ergonomic analyses depend on static evaluation of external
physical loading and postures. These techniques often calculate an index of general shoulder joint loading, most typically in the form of torque values around specified anatomical axes. A preponderance of common work activities, however, includes dynamics and thus may impart somewhat different mechanical consequences on the human shoulder when compared to static postures. To address this inconsistency, a dynamic assessment of shoulder loading, one that incorporates both kinematic and kinetic components was formulated. This model provides potential insights unattainable with
4
static models in evaluating general shoulder torque loading, including the evaluation of the significance of inertial terms for a given task. Beyond describing general loading in the shoulder, it is highly desirable to gain further insight into the loading of individual tissues when performing tasks. The motivation for this is readily apparent as individual tissues are the sites of musculoskeletal disorders and associated pathologies. The evaluation of individual tissue loading, specifically muscle loading, introduces a complex theoretical problem. In the shoulder, as in other biomechanical subsystems, there are far more muscles present in the system than there are mechanical equilibrium equations available. This results in an underdetermined system with an infinite number of solutions possible. Though numerical optimization techniques have been applied to solve this indeterminacy in the shoulder, the present formulations of these optimization-based models are either unavailable or impractical for the practicing ergonomist. Although the Gothenburg Shoulder Model formed much of the conceptual foundation for the model presented in this thesis, several alterations were incorporated into the final muscle force prediction model created as a part of this work. These included: 1) adaptation of the “shoulder rhythm” convention, 2) methodologies for locating joint center locations in vivo, 3) an empirical shoulder joint contact force constraint, 4) an interface with a dynamic shoulder torque prediction model, and 5) a link with a novel model that estimates perceived effort. In addition, a laboratory study was performed to evaluate the model performance for a set of industrially relevant tasks.
5
1.2.2 Perception of Effort in the Shoulder A second area of intellectual interest in this work was the consistency of the relationship between physical exposure to loading and the psychophysical perception of that loading. Though there is a supposed relationship between physical loading and perceived effort, the nature of this relationship is incompletely understood. This work addresses which facets of physical loading predominate the perception of effort, specifically the sensitivity to general joint loading, which is parameterized by joint torques; and specific tissue loading, as is calculated by the muscle force prediction model. Further, it establishes a mathematical prediction method for the perceived effort associated with a task.
1.3
Thesis Statement This dissertation has a central global thesis: Perceived effort in the shoulder is
quantitatively related to loading metrics generated by biomechanical models that consider the external loads and motions associated with a variety of reaching tasks.
1.4
Research Objectives This dissertation consists of five complementary research objectives that address
this thesis. All apply to the reaching exertions described throughout the text: 1) Develop a model to describe dynamic external shoulder torque loading. 2) Develop a model to represent the internal geometry of the musculoskeletal components of the shoulder with visual and numeric representations.
6
3) Develop and empirically evaluate an optimization-based model to predict forces and stresses in the muscles of the shoulder. 4) Establish the nature of the relationship between physical loading and psychophysical effort perception in the shoulder for light manual material handling tasks. 5) Combine the results of objectives 1-4 into a tool that incorporates the functionality of the developed shoulder models. Model evaluation was achieved with an exploratory set of data from a set of load transfer exertions performed by eight subjects. This initial evaluation established both an index of model performance, as well as identified areas for potential future research.
1.5
Thesis Organization This dissertation is written as a collection of four manuscripts, placed between
introductory and concluding chapters. All of the experimental work is dedicated to the study of one-handed load transfer tasks. Chapter II describes the development and application of two mathematical models: a quantitative representation of shoulder loading, in the form of dynamic external shoulder joint torques; and a statistically-driven method of comparing those torque values to recorded effort perception data. Chapter III explains the methodology of developing a shoulder muscle force analysis program for the analysis of workplace man/machine interfaces. The paper describes in detail the components of the muscle force prediction program, as well as its underlying physiological, mechanical, and mathematical bases. Chapter IV addresses the evaluation of the musculoskeletal shoulder model through a set of experiments which measured the
7
muscle activity present during task performance through the use of surface electromyography. Included is a detailed description of the experimental methodology used in the validation study, as well as an evaluation of the performance of the model. Chapter V addresses the importance of tissue-specific loading knowledge in the psychophysical interpretation of effort while performing a task. An extension of Chapter II in some respects, this chapter establishes which mechanical quantities are most closely associated with the perception of effort in the shoulder. Finally, Chapter VI summarizes the major findings and contributions of the dissertation, discusses the implications of these accomplishments, and concludes with recommendations for further pursuit of the improvement of ergonomic analyses of tissue loading and effort perception in the shoulder region.
8
1.6
References
Chaffin, D.B. Development of Computerized Human Static Strength Simulation Model for Job Design, Human Factors and Ergonomics in Manufacturing 7(4):305-322 (1997) de Groot JH, "The variability of shoulder motions recorded by means of palpation", Clin Biomech 12:461-472 (1997) de Groot JH, Valstar ER, and Anvert HJ, "Velocity effects on the scapulo-humeral rhythm", Clin Biomech 13:593-602 (1998) de Groot JH, van Woensel W, and Van der Helm, FCT, "Effect of different arm loads on the position of the scapula in abduction postures", Clin Biomech 14:309-314 (1999) de Groot JH, "The scapulo-humeral rhythm: effects of 2-D roentgen projection", Clin Biomech 14:63-68 (1999) Dul J, “A biomechanical model to quantify shoulder load at the workplace”, Clin Biomech 3:124-128 (1988) Herberts P, Kadefors R, Hogfors C, and Sigholm G, “Shoulder pain and heavy manual labour”, Clin Orthopaedics 191:161-178 (1984) Hogfors C, Sigholm G, and Herberts P, "Biomechanical model of the human shoulder – 1 elements", J Biomech 20:157-166 (1987) Hogfors C, Peterson B, Sigholm G, and Herberts P, "Biomechanical model of the human shoulder joint – 2 the shoulder rhythm", J Biomech 24:699-709 (1991) Hogfors C, Karlsson D, and Peterson B, "Structure and internal consistency of a shoulder model", J Biomech 28:767-777 (1995) Karlsson D and Peterson B, "Towards a model for force predictions in the human shoulder", J Biomech 25:189-199 (1992) Kuzkay C, Boston JR, Rudy TE and Lieber SJ, “Joint moments and balance during lifting in amputees and controls,” North American Congress on Biomechanics, Waterloo, Ontario, Canada (1998) Latko WA, Armstrong TJ, Foulke JA, Herrin GD, “Development and evaluation of an observational method for assessing repetition in hand tasks,” AIHA, 58:278-285 (1997) Laursen B, Jensen BR, Nemeth G, and Sjogaard G, “A model predicting individual shoulder muscle forces based on relationship between electromyographic and 3D external forces in static position, J. Biomech 31:731-739 (1998)
9
Laursen B and Jensen BR, “Shoulder muscle activity in young and older people during a computer mouse task,” Clin Biomech 15:S20-S33 (2000) Makhsous M, Hogfors C, Siemien'ski A, and Peterson B, "Total shoulder and relative muscle strength in the scapular plane", J Biomech 32:1213-1220 (1999) McAtamney L, Corlett EN, “RULA: A survey method for the investigation of workrelated upper limb disorders,” Applied Ergonomics 24(2):91-99 (1993) Musculoskeletal disorders (MSDs) and workplace factors: a critical review of epidemiologic evidence for work-related musculoskeletal disorders of the neck, upper extremity, and low back, chapter 3: shoulder musculoskeletal disorders: evidence for work-relatedness. Second Printing, DHHS (NIOSH) Publication No. 97-141 (1997) Pronk GM, "A kinematic model of the shoulder girdle - a resume", J Med Eng & Tech 13:119-123 (1989) Pronk GM and Van der Helm FCT, "The Palpator - an instrument for measuring the positions of bones In 3 dimensions", J Med Eng & Tech 15:15-20 (1991) Reynolds L, “Zeroing in on ergonomics costs and solutions,” HR Today 7 (1999) Van der Helm, FCT and Veenbaas R, "Modeling the mechanical effect of muscles with large attachment sites - application to the shoulder mechanism", J Biomech 24:11511163 (1991) Van der Helm FCT, Veeger HEJ, Pronk GM, Vanderwoude LHV, and Rozendal RH, "Geometry parameters for musculoskeletal modeling of the shoulder system", J Biomech 25:129-144 (1992) Van der Helm FCT, "Analysis of the kinematic and dynamic behavior of the shoulder mechanism", J Biomech 27:527-550 (1994) Van der Helm FCT, "A finite-element musculoskeletal model of the shoulder mechanism", J Biomech 27:551-569 (1994) Van der Helm FCT and Pronk GM, "3-dimensional recording and description of motions of the shoulder mechanism", J Biomech Eng -Transactions Of The Asme 117:27-40 (1995) Van der Helm FCT and Veeger HEJ, "Quasi-Static Analysis Of Muscle Forces In The Shoulder Mechanism During Wheelchair Propulsion", J Biomech 29:39-52 (1996) Veeger HEJ, Van der Helm FCT, Van der Woude LHV, Pronk GM., and Rozendal RH, "Inertia And Muscle-Contraction Parameters For Musculoskeletal Modeling Of The Shoulder Mechanism", J Biomech 24:615 (1991)
10
Veeger HEJ, Van der Woude LHV, and Rozendal RH, "Load on the upper extremity in manual wheelchair propulsion", J Electromyography And Kinesiology 1:270-280, (1991) Wood JE, Meek SG, Jacobsen SC, “Quantitation of human shoulder anatomy for prosthetic control, part I”, J Biomech 22:273-293 (1989) Wood JE, Meek SG, Jacobsen SC, “Quantitation of human shoulder anatomy for prosthetic control, part II”, J Biomech 22:309-325 (1989)
11
CHAPTER II QUANTIFICATION OF SHOULDER LOADING AND ITS RELATIONSHIP TO THE PERCEPTION OF SHOULDER EFFORT – I. SHOULDER TORQUES
2.1
Abstract This study explores the relationship between shoulder externally induced torques
and perceived exertion levels for sub-maximal manual materials handling tasks. The motions of eight subjects were recorded while performing light load transfer tasks. Following each transfer exertion, subjects rated their perceived shoulder effort for the task. The recorded motion profiles were processed using a biomechanical upper extremity model in order to estimate external shoulder torques. Mean, maximum and integrated shoulder torque values were correlated with task perceived effort scores. Individual subject torque profiles were significantly positively correlated with perceived effort scores (r2 = 0.45 - 0.77). A general predictive model of perceived effort based on mean shoulder torques was generated across all subjects, which showed a decreased prediction accuracy (r2 = 0.50). The accuracy of this predictive model was improved through the inclusion of additional subject and task-specific characteristics as main effects. Specifically, the azimuth direction of the target location, target elevation, subject stature, subject strength, and distance to target location resulted in a coefficient of determination of 0.70. This suggests that a combination of shoulder torque loading and
12
other task and individual characteristics may contribute to the perception of shoulder effort.
2.2
Introduction The analysis of the biomechanics of the upper extremities is at an early stage
when compared to analysis of gait or low back biomechanics (Rau, 2000). However, the prevalence of upper extremity use and resulting trauma in common occupational tasks, such as reaching, grasping, tool use, machine operation, and many other fine motor tasks, demands that the biomechanical effects of such exertions be modeled for the upper limb joints, particularly the shoulder. The use of psychophysical rating scales in ergonomic analysis is common. Scales have been used to evaluate postural discomfort (Genaidy et al., 1995), discomfort (Ulin et al., 1993;), pain (Harms-Ringdahl et al., 1983), effort (Burgess et al., 1995) and exertion levels (Pincivero et al., 2003; Snook, 1978; Borg 1974 & 1990) associated with performing movements or tasks. One attractive aspect of psychophysical metrics is that they represent a psychological integration of the physical effort related to performing a task (Kim et al., 2004). Relationships between the perception of effort and physical workload intensity & duration for constrained tasks have been found (Pincivero, 2000; Capodaglio et al, 1996). These relationships have also previously been shown to be sensitive to interpersonal variation, specifically gender (O’Connor et al, 2002), stature (Burges and Jones, 1997), age (Allman and Rice, 2003), and bodyweight (Baker et al 2001). Previous studies have tended to evaluate the perception of effort with highly constrained tasks, such as knee extension and flexion. In occupational settings, however,
13
tasks involve a complex combination of joints, muscles and other tissues which have been invoked in a synchronous fashion to produce movements and force. Recently, perceived effort ratings for the shoulder and torso regions have been related to workplace geometry and subject-specific characteristics for more complex reaching tasks (Kim et al., 2004). These studies, however, do not consider biomechanical metrics of the task as inputs to the perception of efforts reported by subjects. Hence, the effect of objective physical stressors on perceived effort for more complex upper extremity movement patterns would be beneficial to task analysis. The results of this study will assist in designing future workstations, as it is desirable to not only insure safety (in the form of minimizing physical exposure), but also operator comfort, of which perceived effort is a descriptor.
2.2.1
Research Objectives This investigation had two primary objectives: 1) To quantify the dynamic torque
loading present at the shoulder joint during a simulated materials manual handling task, and 2) To examine the relationship between the subjective perceived effort level and shoulder torque loading that occur while performing such tasks. A biomechanical model was first created to analyze dynamic torque loading at the shoulder during various tasks. The results of this model were then compared statistically with reported subjective exertion levels for a set of goal-directed reaching and object moving tasks performed by several subjects in order to establish the nature of the empirical relationship between these quantities.
14
2.3
Methods
2.3.1
Subjects Eight subjects (4 females and 4 males) participated in the exploratory experiments
as paid volunteers. All subjects were right-handed and free from any known musculoskeletal or neurological disorders. All subjects were performing experiments within a protocol approved by the University of Michigan Institutional Review Board. Data for these subjects is summarized in Table 2.1.
Table 2.1. Gender Male Female
2.3.2
Subject Summary Characteristics for Torque Effort Study
Mean S.D. Mean S.D.
N 4 4
Age (Years) 24.3 3.3 24.3 5.5
Stature (cm) 182.0 9.2 163.5 6.4
Weight (kg) 78.2 11.7 57.9 0.2
Strength (Nm) 52.32 11.0 30.0 0.9
Anthropometric Measurements Several anthropometric measurements were taken on each subject. The measured
quantities were stature, bodyweight, and upper arm segment lengths and widths. Segmental circumferential measurements were taken according to Yeadon and Morlock (1989).
2.3.3
Determination of Maximum Voluntary Contraction (MVC) strength Following anthropometric measurement, the maximum voluntary isometric
shoulder strength of each subject was measured for three standardized conditions (Figure 2.1): sagittal (flexion), mid-sagittal, and lateral (abduction). The exertion was repeated twice for each plane and force data were measured using a Strength Test Monitor
15
(Measurement Systems, Inc). A minimum of thirty seconds of rest was provided between recorded trials. The maximum of the two strength values was considered to represent a 100% MVC level, provided that the trial values were within 20% of one another. These data were used to calibrate the hand loads manipulated during the experiment
Figure 2.1
2.3.4
Strength Testing Postures for Hand Load and Effort Rating Calibrations. A) Abduction; B) Flexion/Abduction; C) Flexion.
Psychophysical Effort Perception Calibration Procedure Prior to experimental trials, subjects were trained to calibrate their effort
perception response as a percentage of their maximal shoulder strength. Following the determination of the MVC, Subjects were then directed to produce exertion levels that represented specified fractions of their MVC values. Visual feedback via a digital readout of the force applied to the load cell was provided initially to help in learning the association between the physical sensation and the exertion levels. Following this initial stage, the subjects were asked to reproduce requested exertions of 20%, 50%, and 80% MVC without the aid of visual feedback. They continued practicing the tasks until they could consistently produce a desired exertion level to within ± 5%.
16
2.3.5
Psychophysical Rating Scale Used Immediately following each task, the subject was asked to evaluate the shoulder
specific effort they perceived while performing the task, with respect to the calibrated exertions performed. This effort rating was ranked using a continuous modified Borg CR-10 Exertion Scale (Borg, 1982), presented in a previous study (Kim et al, 2004). The exertion scale was modified to a continuous scale through the use of a linear potentiometer placed in front of the subject. The scale and corresponding verbal anchors (Figure 2.2) was printed in large type and displayed on a wall facing the subject for reference purposes. The perceived effort response, defined by the position of the potentiometer cursor on the scale, was automatically recorded after each trial and appended to the kinematic data recorded.
Figure 2.2
Psychophysical Effort Rating Scale. Each level corresponds to 10% of the reference MVC level determined through strength testing.
17
2.3.6
Motion Capture Methodology Optical and electromagnetic markers were placed on palpable bony landmarks in
using a reduced marker set configuration (Figure 2.3). Optical markers were placed symmetrically on the following upper body landmarks: RAC, LAC (Right and Left Acromion), RHE, LHE (Right and Left Humeral Epicondoyle), RHA, LHA (Right and Left Hand), and are shown as spheres in Figure 2.3. In addition, four electromagnetic sensors were placed on the back of the hands, as well as at the L5/S1 juncture (black box in rear view) and the suprasternale (black box in front view). The three-dimensional (3D) locations of these markers were tracked during the experiment at a rate of 25 Hz.
Figure 2.3
Marker Positioning During Experimental Trials. Optical (light spheres) and electromagnetic (dark cubes) motion tracking markers placed on body landmarks. These external markers were used to estimate internal joint centers and link lengths for biomechanical torque modelling.
18
2.3.7
Experimental Task Description An articulating arm controlled by a computer was used in order to place a push-
button target to be reached (a 25-mm diameter button switch) at any position in a 3-D simulated work place. The subjects, seated in an industrial-style chair, were required to reach and press the switch with the right index finger (Figure 2.4). The task consisted of reaching to targets during both hand-loaded and unloaded conditions. First, the target was positioned in space relative to the seated subject in one of the pre-defined directions. Once the target was visually located, an auditory signal was given by the computer, to initiate the reach movement from a “home” location. The motion was completed by depressing the button target. Following a 3-second hold at the target position, a second auditory signal indicated that the reach was over and the subject should return to the starting position. After completing the task, the subject rated their perceived shoulder effort on the scale using the potentiometer.
19
Figure 2.4
Representation of an Experimental Loaded Reaching Task. The reach is directed to target in the sagittal plane at a high elevation Visible in the figure are the starting position (dark disk on platform), the potentiometer used to indicate their perceived effort score (rectangle on platform), and the target (a push button attached to an articulated arm).
The locations of the targets were varied in order to generate motions that produced a wide range of shoulder torque values. Figure 2.5A shows the radial azimuth target orientations used in the experiment (-30°, 0°, 45°, 90°, and 120°) from an overhead view. A perspective view of the targets is shown in Figure 2.5B. The 0° azimuth is directly in front of the subject (in the sagittal plane). Three target elevations from the hip were also tested: Low (At Hip Level, at a 0° [horizontal] angle from the hip), Mid (at 35° above horizontal from hip), and High (above shoulder, at 70° above the hip). The hip level reaches were not performed along the –30 and 0 azimuths, as they were obstructed by the work bench (see Figure 2.4). Two reach distances, derived from the subjects’ 20
maximum reach envelopes were also tested, representing both near (60% maximum reach distance) and far (85% maximum reach distance) reaches. The weight in the handload during the motion was set at 0%, 25%, and 50% static shoulder strength obtained during strength testing. The load was shaped to be easily gripped during the reaching motions.
1000
Y Axis
Z Axis
1000
A
500
-30
B Global Origin
0 0 -500 Y Axis
-500 0
-500
1000
500
45
500 0
1000 500
0
-500
X Axis
0 500 X Axis
90
120 1000
85% 70 Elevation
Z Axis
60%
35 Elevation 85%
0 Elevation 60% 85%
Global Origin 0
Figure 2.5
C
60%
500
1000X Axis 0
500
1000
Spatial Distribution of Target Locations. The targets were located within the right-hand reach envelope along five azimuths, three elevations and two distances. The near and far targets of each pair are represented by boxes and filled circles, respectively.The global origin is shown as a diamond in each plot (at 0,0,0). A) Perspective view; B) Overhead view, highlighting the five azimuth directions; and C) Side view of a representative Azimuth (90° in this case), highlighting both the elevation projections and the near and far target distances.
During the course of the experimental testing, the subject performed reaches in six blocks of sixteen reaches, resulting in ninety-six total reaches. Each intra-block reach
21
was followed by a one minute rest period. A pause of five minutes followed the end of each block for recording system reset. The entire sequence of ninety-six reaches was randomized according to target position and hand weight. It included a total of seventyeight unique reaches and eighteen repeated trials.
2.4
A Biomechanical Torque and Effort Model of the Shoulder The first part of the model calculates the reactive joint torques at the shoulder
experienced while performing the task, while the second part estimates the perceived shoulder effort associated with the magnitudes of those torques. The torque model was largely adapted from an existing general methodology used to study gait (Vaughan et al., 1991). This model can be described in four key stages: (1) segment property description; (2) linear kinematics; (3) angular kinematics; and (4) computation of joint forces and torques. Following these stages, the calculated torque values were normalized using a strength criterion as described.
2.4.1
Defining Arm Segment Properties for the Biomechanical Model The right arm was modeled as a three-segment chain composed of the upper arm,
forearm, and hand segments. Individual segment mass values and directional moments of inertia were calculated by using the anthropometric data gathered as inputs into the previously reported regression equations of Zatsiorsky et al. (1993).
22
2.4.2
Linear Kinematics
Determination of Joint Center and Segment Center of Mass Locations The data from the two motion-recording systems was synchronized, and then relevant joint center locations were calculated using previously reported methods developed for the described marker set (Nussbaum & Zhang, 2000). These methods allow the estimation of upper extremity joint centers based on the reduced set of surface markers used in this experimental protocol (Figure 2.3). The joint centers estimated were the glenohumeral (represented by the center of the humeral head), the elbow, and the wrist joints. From these estimated joint center locations, the spatial locations of the centers of mass (CMs) of each segment were calculated based upon the Center-ofMass/Segment-Length Ratios reported by Clauser et al. (1969).
Segment Center of Mass Velocity and Acceleration Calculation Velocity and acceleration of the segment centers of gravity were determined using numerical differentiation. The first and second derivatives of the displacement-time data can be simply expressed as the following (adapted from Vaughan et al, 1991):
∂xn x −x = x&n = n+1 n+1 ∂t 2∆t
(2.1)
and
∂ 2 xn x − 2 xn + xn−1 = &x&n = n+1 ∂t (∆t ) 2
(2.2)
Where x = a data input point, n = the nth sampled frame, and ∆t is the time between frames.
23
Prior to performing the numerical differentiation, the raw data were smoothed using a second-order, 2-pass, Butterworth low-pass filter set to 6 Hz. The cut off frequency was determined by the result of an FFT analysis, to insure that the power of the signal was negligible beyond the selected frequency.
2.4.3
Angular Kinematics
Defining local coordinate systems of the arm The neutral position was defined as having the arm extended to the side, perpendicular to the torso, with palms turned downward. A local coordinate system was defined for each segment of the arm considered (Figure 2.6), as delineated in Table 2.
X Z Y
Figure 2.6
Local Coordinate Systems. A) Local coordinate systems in the reference posture; for each segment, the positive x-,y-, and z- axes are shown by the directions of the pyramid, the rectangular solid, and the cylinder, respectively. The x-axis coincides with the long axis of each segment. It has been moved from the joint centers for visualization purposes. B) Segmental orientation in a representative trial initial position.
24
Table 2.2.
Arm Segment Local Coordinate System Definitions Used in the Biomechanical Model of the Arm, in the Defined Neutral Position.
Body Segment Hand (1)
Neutral x-axis Line Connecting wrist joint center with grip center (rotation axis)
Neutral y-axis Line passing through wrist joint center and ulnar protuberance, perpendicular to xaxis (flexion axis)
Neutral z-axis Cross product of x and y axes, pointed cranially (ab/adduction axis)
Forearm (2)
Defined along line between elbow joint center and wrist joint center (rotation axis)
Cross product of x and z-axes, directed superiorly (ab/adduction axis)
Perpendicular of forearm and upper arm x-axes, passes through humeral epheseus (flexion axis)
Upper Arm (3)
Defined along line Between glenohumeral joint and elbow joint center (rotation axis)
Cross product of x and z axes, directed anteriorly (ab/adduction axis)
Perpendicular of the forearm and upper arm x axes, passes through acromion (flexion axis)
Euler Rotational Sequence A derivative of a technique that has been described as the Joint Coordinate System (JCS) [Nigg and Herzog, 1994] was applied to assess segmental rotation. The first rotation is about the flexion/extension axis of the joint (ψ), the second rotation is about the abduction/adduction axis (θ), and the third rotation is about the longitudinal segment axis (φ). These are slightly different permutations for the three different arm segments. Two rotations are needed to gain meaningful angle results. The first, a 3-2-1 transformation was used to describe the orientations of the upper arm and forearm: cosψ cos θ sin ψ cos θ − sin θ ⎤ ⎧ X ⎫ ⎧x ⎫ ⎡ ⎪ ⎪ ⎪ ⎪ ⎢ y ( sin ψ cos φ cos ψ sin θ sin φ ) (cos ψ cos φ sin ψ sin θ sin φ ) cos θ sin φ ⎥⎥ ⎨Y ⎬ = − + + ⎨ ⎬ ⎢ ⎪ z ⎪ ⎢ (sin ψ sin φ + cos ψ sin θ cos φ ) (− cosψ sin φ + sin ψ sin θ cos φ ) cos θ cos φ ⎥ ⎪Z ⎪ ⎦⎩ ⎭ ⎩ ⎭ ⎣
25
(2.3)
The orientation of the hand, due to its different neutral orientation, was described by a [23-1] transformation matrix: cos ψ cos θ sin θ ⎧x ⎫ ⎡ ⎪ ⎪ ⎢ ⎨ y ⎬ = ⎢(sin ψ sin φ − cosψ sin θ cos φ ) cos θ cos φ ⎪ z ⎪ ⎢(sin ψ cos φ + cos ψ sin θ sin φ ) − cos θ sin φ ⎩ ⎭ ⎣
− sin ψ cos θ
⎤⎧ X ⎫ ⎪ ⎪ (cos ψ sin φ + sin ψ sin θ cos φ )⎥⎥ ⎨Y ⎬ (cos ψ cos φ − sin ψ sin θ sin φ ) ⎥⎦ ⎪⎩Z ⎪⎭
(2.4)
In equations 2.3 and 2.4 X, Y, and Z are global 3-D coordinates and x, y, and z are local 3-D coordinates.
Determination of Segment Angular Velocity and Acceleration The segmental angular velocities and accelerations are determined through classical mechanical methods (Vaughan et al., 1991). These quantities are functions of the Euler angles and their first and second derivatives. The derivatives of the Euler angles were determined using the same finite differencing technique described for linear kinematics.
2.4.4
Calculation of External Joint Forces and Torques
Calculation of Joint Forces In the force equilibrium calculation, derived from the linear form of Newton’s second law of motion can be applied for each segment using the following equilibrium for each segment:
∑ F = msegment × aCOM , segment
(5)
Where F = forces, m = mass of segment, and a = acceleration of segment COM. This is be customized for each segment. External forces are composed of the load in the hand and the weight of each of the segments. The small force required to depress the target
26
button was considered negligible. Solving the resulting equation achieves the cumulative external joint load at each proximal joint segment.
Calculation of Dynamic Joint Torques The angular analog of Newton’s second law was applied to each segment, which is summarized as:
∑ M = H&
(6)
Where M = external torque and H& = rate of change of segmental angular momentum. The rate of change of angular momentum can be calculated based upon the segmental moments of inertia and the segmental velocities and accelerations. External torques are calculated based on the cross products of the produced forces and their lever arms. A more detailed summary of this technique can be found in the literature (Vaughan et al., 1991). Assuming segmental equilibrium, the torque can be found at the proximal end of each segment.
Description of Calculated Dynamic External Shoulder Torques Instantaneous resultant torques were computed as the vector norm of the three cardinal axes torques (about the global X,Y,Z axes). Based on these resultant torques, the maximum torque, mean torque, and integrated (torque * time) torques were calculated for each trial. These resultant torques were used exclusively in the construction of the torque/effort regression model described.
27
Strength-Based Normalization of Shoulder Torque Values To compare the relationship between shoulder loading and perception across the pool of subjects, torque values were normalized by reference strength torques for each subject. The reference torque strength value used corresponds to the average of the calculated moment for all three directions tested. Therefore, the shoulder torque metrics used can be thought of as fractions of maximum shoulder torque strength.
2.4.5
Description of the Torque and Perception Regression Model A multiple linear regression model was created to evaluate the ability to predict
shoulder effort perception levels in reaching tasks. The main effects considered in building the model (Table 2.3) included torque metrics for each trial (integrated, maximum, and mean torques), subject-specific variables (stature, strength, and body mass), and task characteristics (target azimuth, target distance as a fraction of reach capability, and target elevation angle).
28
Table 2.3
Parameters used in the statistical effort prediction models.
Parameters (By Type) Shoulder Torques Integrated Torque Maximum Torque Mean Torque
Description of Parameter Time integral of the normalized torque over the entire exertion (s) The maximum instantaneous torque achieved during the task performance The time-averaged torque value for the entire exertion
Anthropometry Stature Body Mass
Standing height (m) Clothed body weight (kg)
Conditioning Shoulder Torque Strength
Measured shoulder MVC strength in test postures, represented by torque produced in MVC exertions (Nm)
Task Variables Target Azimuth Target Elevation Angle Target Distance
Angle of target azimuth measured from the sagittal plane Projection of target elevation from the hip point Near or far target location along azimuth and elevation
Statistical Modeling Procedure for the Perception Model All main effects were listed for a pre-selection process. Pearson correlation coefficients between main effects were calculated in order to screen out highly correlated variables. Stature and body mass were found to correlate closely (r = 0.81). As a result, only stature was considered in the regression equations. For the tasks studied, all three representations of the resultant torque (maximum, mean, and integrated) were highly correlated (r = 0.96-0.98). Hence, the task mean torque was used as a main factor in the regression model. It should be noted that subsequent references to torque with regards to the regression model will refer to this resultant normalized mean shoulder torque. In addition, the torque was found to correlate highly with the load in the hand (r = 0.85). As all other continuous variables were found to correlate weakly with one another, they were
29
included. The regression model was developed using a stepwise mixed inclusion technique available in the JMP-IN software program. The inclusion/exclusion probabilities were set at 0.25/0.25.
2.5
Results
2.5.1
External Dynamic Shoulder Torque Model Results Overall, normalized resultant mean shoulder torques increased significantly with
increased distance to target (p < 0.05) and subject body weight (p < 0.05). Normalized resultant mean torques also significantly decreased with increasing strength (p < 0.05). Stature had no independent effect. The effects of azimuth and elevation angles on the normalized resultant shoulder torques, while also significant, were complex. A typical example of time variation of the computed normalized torque during a single hand loaded (25%MVC) reach trial is shown in Figure 2.7. The integrated torque values for the two curves were found to correlate very closely. The integrated torque is the area under the torque-time curve. From this example, which is typical of the experimental results, it can be seen that the impact of the dynamic terms in this analysis was minimal. This suggests that a static or quasi-static analysis may be sufficient to study torque loading profiles during these types of low-speed load transfer tasks.
30
Figure 2.7.
Normalized Resultant Torque Profile for a Representative Reaching Task. The torque corresponds to the reach for a middle elevation target in the sagittal plane with a 25% MVC hand load. The static and dynamic torques are represented by a solid and a dotted line, respectively. Differences were found to be minimal in the self-paced reaches included in this study.
Across all conditions, the average integrated normalized resultant shoulder torque was 1.97 ± 0.84 sec, which is equivalent to approximately two seconds of maximum strength moment production. The average normalized resultant mean torque was 0.28 ± 0.10, or approximately 28% of maximum shoulder flexion/abduction strength. The average normalized resultant maximum torque over all trials was 0.34 ± 0.12, or 34% MVC.
31
2.5.2
Intra-subject Effort Perception Variation Results Effort perception responses for identical repeated trials displayed some expected
variability. This was quantified through the comparison of effort scores reported on a series of random repetitions of several experimental trials by each subject as described in the experimental methods. Using this approach, the calculated mean intra-subject variabiliblity for repeated trials was 0.85 ± 0.69 on the 10-point effort rating scale used (25% & 75% quartiles: [0.3-1.15]). This variation, represents nearly a full point on the rating scale, and was statistically significant (p < 0.05), which indicates that some natural variability exists in the perception of shoulder effort.
2.5.3
Relationship Between Shoulder Torque and Perceived Effort Results A univariate linear regression model indicated a positive correlation between
shoulder normalized resultant mean torque loading and perceived shoulder effort for the subject population (with r = 0.71, p < 0.05). The relationship is described by the following equation: Effort = 15.0 *τ n
(7)
In this equation, τn is the normalized resultant mean torque for a trial. For individual subjects, the correlation of torque and effort ranged (r = 0.67-0.88). When considering the total subject population, the inclusion of additional main effects in a multiple linear regression model increased the predictive accuracy of the model substantially. An expanded model including subject stature, and strength, and target azimuth, elevation and nominal distance increased the variance explanation from 50% to 70%. Each of these main effects was found to have a significant influence on the
32
perception of effort in the shoulder. Inclusion of interaction terms did not lead to a marked improvement in the model’s prediction accuracy (r2 = 0.71), so they were not included in the final model.
2.5.4
Influence of Main Effects in the Expanded Regression Model Results Aside from the influence of shoulder torque loading on perceived effort, there
were several other factors significantly related to shoulder effort perception. Overall, taller subjects reported lower effort perception than shorter subjects (p < 0.01), with perception decreasing nearly a full point (0.99) for every additional 10 cm of stature. Increased strength was shown to result in higher ratings of perceived effort (p < 0.01) for the levels of normalized strength required in the task. As expected, perceived effort increased significantly (p < 0.01) with the linear distance to the target. Effort perception also increased with elevation above and below the mid target level (p < 0.05). The extreme backward azimuth angle (120°) produced significantly higher effort ratings than all other azimuths, amongst which there were no significant differences. Nevertheless, a trend indicating higher effort ratings with increased deviations from the (45°) plane did exist. It should also be noted that the full regression model presented a condition number of 75, indicating multi-co-linearity in the data set.
33
2.5.5
Coefficients of the Expanded Regression Model for Effort Prediction Results The final regression model had an r2 = 0.70. Its form is the following:
Effort = 7.2 + 18.1 * τ n + 0.1 * τ S − 8.6 * S + 0.014 * E + 0.001 * ( E − 41) 2 + 0.005 * A + 0.0001 * ( A − 55) 2 + 0.016 * D
(8)
Where: Tn = Mean resultant normalized shoulder torque; Ts = Shoulder torque strength (Nm); S = Stature (m); E = Target elevation above seat pan (cm); A = Azimuth (degrees from sagittal plane); and D = Distance to target from center of seat pan (% of maximum reach distance).
2.6
Discussion The primary goals of this investigation were to determine the correlation between
an objective measure (shoulder torque loading) and a subjective response (perceived effort in the shoulder), and the eventual influence of reach location and subject characteristics on effort perception. There is an inherent preference for objective data over subjective data in prospective job analysis. This preference is predicated on the specific quantitative knowledge of critical task characteristics, including personnel demography, spatial workstation layout and end-effector load levels. Indeed, subjective measured variables have attracted criticism as somewhat unreliable in the literature (Annett, 2002). However, a critical aspect of the success of a workplace design is the psychophysical “comfort” of the operator. Therefore, it is of interest to estimate a metric of this “comfort” based on objective calculated data that can be obtained by a mathematical model of shoulder loading.
34
2.6.1
Shoulder Torque as a Predictor of Perceived Effort The present study indicates that the perceived effort in the shoulder associated
with the performance of a dynamic loaded reach task is dependent on a combination of factors, but is primarily related to the torque loading of the shoulder joint. This loadingperception relationship is in agreement with previous findings of relationships between stimuli and perception for the knee joint (Pincivero 2003), and for general sensory perception (Stevens, 1957). If the perception of effort corresponded exactly to the level of shoulder torque loading, the expectation for the value of the coefficient of Equation 7 would be 10, as an increase of 0.1 in the normalized torque would correspond to 10% of MVC, represented as one point on the effort scale (Figure 2.2). However, the coefficient on the linear relationship between torque and perceived effort exceeded was found to be 15 or 50% higher. This result suggests that the torque levels were overestimated by the subjects, when related to their reference strength. This is slightly more pronounced in the final regression equation (Equation 2.8). This result, however, is mediated by the other subject and task-specific factors. This improvement of prediction accuracy with a multivariate model including subject and task specific modulating factors is in agreement with previous attempts to model perceived shoulder effort for a simulated transfer tasks (Kim et al, 2004). An advantage of the present model is the use of biomechanical metrics that further enhanced that regression model’s ability to explain the variance in perceived effort ratings.
35
2.6.2
The Influence of Personal Characteristics on Effort Perception Personal characteristics were shown to significantly affect perception of effort in
the shoulder. Subject stature and an individual’s strength were the specific factors shown to contribute most to this effect. Stature was the primary individual characteristic found to significantly weight the effort prediction equation, as taller subjects tended to report lower effort scores. This result is in agreement with a previous study (Kim et al., 2004). In addition, individual shoulder strength was found to alter perception of effort. Increased strength yielded higher levels of effort perception for equal %MVC loading levles. One explanation for the presence of the stature component effects is the reflection of a gender-based effort perception effect, which has been reported as showing higher effort perception in males than females for similar exertions (Kim et al., 2004). For repeated elbow flexion-extension, contrary results have also been indicated (O’Connor, 2002), in which females were demonstrated to report lower perceived exertion scores for similar tasks. Other studies have found little confirmation of a gender effect for knee extension (Pincivero, 2001, 2003). This array of conflicting results confirms the difficulty associated with interpreting the effects of personal characteristics on effort perception. While the women in this study were generally of smaller stature than the men, the strength values were clearly higher for the men than the women. However, these two characteristics (stature and strength), while positively correlated, had opposite impacts on the regression prediction of effort. Therefore it is difficult to reach a conclusion regarding the role of gender in confounding the results. Another possibility exists for the dependency of effort perception on strength. The influence of strength on the perception of effort may relate to the tendency to report actual tissue load in the
36
shoulder, rather than effectively reporting percentage of muscular effort. In essence, the major finding with respect to subject-specific characteristics is that there is an interpersonal difference in the interpretation of similar relative torque loading in the shoulder, although the exact cause of that difference is unclear.
2.6.3
Influence of Spatial Task Characteristics on Effort Perception The location (elevation, distance and azimuth) of the target relative to the subject
was shown to impact perceived effort. Although the position of the target was the major driver of the motion performed, and thus relates directly to temporal body position and associated torque levels, the impact on perception was significant beyond this contribution to the calculated shoulder torques. This suggests that beyond the joint loading, the various postures required to attain different targets created a qualitatively, if not quantitatively, different experience of shoulder effort perception. This observation leads to the recognition that effort perception is driven not only by the load on the joint of interest, but also by the motion direction –induced posture relative to the initial posture of the individual. The internal relative geometry of the shoulder changes with posture variation, and thus the mechanical advantages of muscle tissues are also altered (Meskers, 1998). As the postures attained during the reach are clearly related to the position and orientation of the target relative to the subject, it is not surprising that the level of perceived effort changes with changes in target location. Within this context, it was observed that as the target location deviated from an ‘optimal’ position, effort ratings increased in all directions (azimuth, elevation, and distance). This is in agreement with
37
previous findings that have demonstrated that as posture deviates from a neutral position, perception of discomfort increases (Kee and Karwowski, 2001). Thus, torque level loading may not sufficiently explain variations in level of perception alone, but must be complemented by some index of body posture. This postulate is further supported by changes in the motor command that occur with posture. As different positions of the arm produce different mechanical advantages about shoulder joints for each attached muscle (Keuchle et al, 1997), so too must the motor program adjust to produce coordinated movement. This is supported by empirical findings that work location in overhead tasks influences both localized muscle fatigue as well as discomfort (Wiker et al, 1989). Additionally, the maintenance of shoulder stability has been recognized as a contributor to the activation of additional antagonistic muscles. This stability requirement also changes as a function of the instantaneous glenoid position. As the motor command must change to respect these requirements, a variable motor command based on different positions for the same level of torque would be expected as was observed in the experiment. Changes in the descending motor command have been associated with the perception of effort in the past (Cafarelli 1982; Gandevia et al, 1993).
2.6.4
Study Limitations While significant insights were developed from this study, it should be mentioned
that the conclusions are based on a distinct subject pool, and are restricted to a particular occupational task, one-handed load transfer. Though it is suggested that these results can be used to gain a general perspective on the loading/perception relationship in the
38
shoulder, care should be taken when interpreting work situations that are substantially different than those studied with this model, such as heavy pushing and pulling. Further research into the impact of the directionality of external loading and posture is thought to be required to have a truly universal model of shoulder effort perception to use for a wider range of task simulations. In addition, the size of the subject population somewhat limits the findings. It is suggested that a larger sample would increase the level of confidence in applying these results to a more general population.
2.6.5
Reconstructing Effort Perception Pathways in the Shoulder The results of this study suggest a possible theoretical framework of effort
perception based on mechanical loading, which is shown in Figure 2.8. In this model, the loading is interpreted psychologically by a process that is modulated by a combination of anthropometric, task related and personal characteristics. This results in a final rating of perceived effort. Aspects of this perception have provided controversy in the literature as to the origin of the perception of effort. One popular opinion is that the perception is largely associated with an efferent perception that is proportional to the magnitude of the descending motor command (Gandevia et al 1990; Burgess et al 1995). However, the role of afferent information in modulating the signal is poorly understood. There is also a supposed relationship between the anticipated physical requirement and the perception of effort. In this experiment, however, subjects were unaware of the load they were manipulating prior to the trial (except in cases where the reaches were unloaded). The results of this experiment are in agreement with relationships between torque and effort observed for other body joints (Burgess et al 1995, 1997, 2002), which argues for the
39
possibility of an afferent complement to the efferent perception of the motor command, which also has evidence of contributing to perception (Kilbreath et al., 1997).
Figure 2.8
2.7
A Theoretical Perspective of Effort Perception in the Shoulder. The final effort rating depends on loading data that is integrated mentally, in the context of modifying factors.
Conclusions This study was successful in showing that subjective perception of effort can be
predicted using objective biomechanical torque data. The predictive ability of a regression model was increased with the inclusion of additional task and subject information for a set of load transfer tasks. The model that resulted from this investigation can be fundamentally useful in the assistance that it provides towards the prospective analysis of workstations in relation to subjective perceived effort. In addition, estimates of instantaneous and cumulative dynamic joint torque loading can be made to assist in the evaluation of the marginal benefits of adjusting existing workstations to reduce overall and peak shoulder torque loading. It is suggested that the relationship between torque loading and posture be further developed to include the 40
effects of directional hand force production and internal bony geometric variation, as the results suggest that variable perceived effort can exist for similar torque loading in different, task-required postures.
41
2.8
References
Allman BL, and Rice, “Neuromuscular Fatigue and aging: central and peripheral factors”, Muscle Nerve 25:785-796 (2002) Allman BL, and Rice CL, “Perceived exertion is elevated in old age during an isometric fatigure test”, European Journal of Applied Physiology 89:191-197 (2003) Annett J, “Subjective rating scales: science or art?”, Ergonomics 45:966-987 (2002) Baker JS, Baily DM, and Davies B, “The relationship between total-body mass, fat-free mass and cycle ergometry power components during 20 seconds of maximal exercise”, Journal of Science and Medicine in Sport 4:1-9 (2001) Borg GA “Perceived exertion”, Exercise and Sport Sciences Reviews 2:131-153 (1974) Borg GA “Psychophysical basis of perceived exertion”, Medicine and Science in Sports and Exercise 14:371-381 (1982) Borg GA “Physchophysical scaling with applications in physical work and the perception of exertion”, Scand J of Work/Environment and Health 16(S1):55-58 (1990) Burgess PR, Cooper TA, Gottlieb GL, and Latash ML, “The sense of effort and two models of single-joint motor control, Somatosensory and Motor Research 12:343-358 (1995) Burgess PR and Jones LF, “Perception of effort and heaviness during fatigue and during the size-weight illusion“, Somatosensory and Motor Research 14:189-202 (1997) Burgess PR, Jones LF, Buhler CF, Dewald JP, Zhang LQ, and Rymer WZ, “Muscular torque generation during imposed joint rotation: torque-angle relationships when subjects’ only goal is to make a constant effort”, Somatosensory and Motor Research 19:327-340 (2002) Capodaglio P, Grilli C, and Bazzini G, “Tolerable exercise intensity in the early rehabilitation of paraplegic patients: a preliminary study”, Spinal Cord 34:684-690 (1996) Chaffin D.B, J.J Faraway, X. Zhang, and C. Wooley, “Stature, Age, and Gender Effects on Reach Motion Postures,” Human Factors, accepted November (1999) Clauser CW McConville JT, and Young JW, “Weight, volume and Center of Mass of Segments of the Human Body, AMRL-TR-69-70, Aerospace Medical Research Laboratories, Dayton, Ohio, (1969) de Looze MP, Kuijt-Evers LFM, and van Dieen J, “Sitting comfort and discomfort and the relationships with objective measures”, Ergonomics 46(10):985-997 (2003) 42
Gandevia, S.C., Macefield G., Burke D., and McKenzie D.K., “Voluntary activation of human motor axons in the absence of muscle afferent feedback. The control of the deafferented hand,” Brain 113:1563-1581 (1990) Genaidy AM, and Karwowski W, “The effects of neutral posture deviations on perceived joint discomfort ratings in sitting and standing postures”, Ergonomics 36:785-792 (1993) Harms-Ringdahl K, Brodin H, Eklund L, and Borg G, “Discomfort and pain from loaded passive joint structures”, Scand J Rehab Med 15:205-211 (1983) Kee D. and Karwowski W, “The boundaries for joint angles of isocomfort for sitting and standing males based on perceived comfort of static joint postures,” Ergonomics 44:614-618 (2001) Keuchle DK, Newman SR, Itoi E, Morrey BF, An KN, “Shoulder muscle moment arms during horizontal flexion and elevation”, J of Shoulder and Elbow Surgery 9:429-439 (1997) Kilbreath S.L., Refshauge K, and Gandevia S.C., “Differential control of the digits of the human hand: evidence from digital anaesthesia and weight matching,” Experimental Brain Research 117:507-511 (1997) Kim, K.H., Martin BJ and Chaffin DB, “Modeling of shoulder and torso perception of effort in manual transfer tasks,” Ergonomics 47(9):927-944 (2004) Lippitt S, Matsen F. “Mechanisms of Glenohumeral joint instability,” Clin Orthopedics 291:20-28 (1993) Nigg BM and Herzog W. Biomechanics of the Musculo-skeletal System, 2nd Edition, Wiley, New York, NY, 326-229 (1994) Nussbaum and Zhang “Heuristics for locating upper extremity joint centres from a reduced set of surface markers”, Human Movement Science 19:797-816 (2000) O’Connor PJ, Poudevigne MS, and Pasley JD, “Perceived exertion responses to novel elbow flexor eccentric action in women and men”, Medicine and Science in Sports and Exercise 34:862-868 (2002) Pincivero DM and Gear WS, “Neuromuscular activation and perceived exertion during a high intensity, steady-state contraction to failure”, Muscle Nerve 23:514-520 (2000) Pincivero DM, Coelho AJ, Campy RM, “Perceived exertion and maximal quadriceps femoris muscle strength during dynamic knee extension exercise in young adult males”, Eur J Appl Physiol 89:150-156 (2003)
43
Rau G, Disselhorst-Klug C, and Schmidt R, “Movement biomechanics goes upwards: from the leg to the arm”, J Biomech 33:1207-1216 (2000) Snook SH, “The design of manual handling tasks”, Ergonomics 21:263-985 (1978) Stevens SS, “On the psychophysical law”, Psychol Rev 64:153-181 (1957) Ulin SS, Armstrong TJ, Snook SH, and Franzblau A, “Perceived exertion and discomfort associated with driving screws at various work locations and at different work frequencies,” Ergonomics 36: 833-846 (1993) Vaughan CL, Davis BL, and O’Connor JC, Dynamics of Human Gait, Champaign, IL: Human Kinetics (1991) Wiker SF, Chaffin DB, and Langoff GD, “Shoulder posture and localized muscle fatigue and discomfort”, Ergonomics 32:211-237 (1989) Yeadon MR and Morlock M, “The appropriate use of regression equations for the estimation of segmental inertial parameters”, J Biomch 22:683-689 (1989) Zatsiorsky V and Seluyanov V, “Estimation of the mass and inertia characteristics of the human body by means of the best predictive regression equations”, Biomechanics IXB, Human Kinetics (1993)
44
CHAPTER III A BIOMECHANICAL SHOULDER MODEL FOR ERGONOMIC ANALYSIS – I. MODEL DESCRIPTION
3.1 Introduction The musculature of the shoulder is known to provide joint stabilization and both produce and resist torque loading for a variety of both occupational and daily living tasks. Knowledge of the force and stress levels present in individual muscles and connected tissues while performing tasks could contribute significantly to the proactive prevention of shoulder disorders. Currently available shoulder ergonomic analysis tools are somewhat limited in scope and detail. Common tools rely primarily on indirect estimates of joint and tissue loading obtained from limited postural and hand load information (McAtamney et al 1993; Moore and Garg, 1995; Latko et al., 1997). They do not, however provide detailed information regarding the specific structural loading that may lead to musculoskeletal disorders and associated pathologies. Advanced computerized software exists for calculating static joint torques and loading in a variety of work positions, such as the University of Michigan’s 3Dimensional Static Strength Prediction Program (3DSSPP) (Chaffin, 1997). This software also generates estimates of the percentage of a given population that would have the required strength to generate the calculated torques. While very useful in designing tasks that are not strength-limiting, this type of model is not designed to provide specific tissue loading information that could be used to evaluate an exertion for potential
45
discomfort, fatigue, or injury. Further, there are many industrial tasks that incorporate a significant dynamic component, which is not considered with static analysis. To address this issue, a dynamic model for the calculation of joint torques and forces in the shoulder is required. In the shoulder region, the complicated 3-dimensional geometry of the associated structures causes difficulty in assessing the impact of general loading on specific tissues. The orientations of the muscles in the shoulder are related to both end-effector (hand) position and the magnitude of hand loads. The extreme range of shoulder bone movement possible results in the ability of individual muscles to produce torques in a range of directions and magnitudes, depending upon instantaneous bone orientations. Therefore, establishing a reliable and robust geometric model of the shoulder is required to allow the characterization of the changing roles of the individual muscles for various postures and load levels that are experienced during work tasks. Several previous studies have addressed the concept of the creation of geometric models of the shoulder and estimation of shoulder muscle forces. (Van der Helm, 1994; Garner and Pandy, 1999; Hogfors 1987; Hughes and An, 1997; Laursen 1998). The models have primarily used electromyography (EMG) (Laursen, 1998), and numerical optimization techniques (Van der Helm, 1994; Hogfors, 1987; Hughes and An, 1997). These models provide valuable information concerning potential force distributions in the shoulder. However, these models were not developed specifically for 1) integration with ergonomic analysis, 2) dynamic tasks, and 3) non-expert use. This prompted creation of a model specifically to be used symbiotically with traditional job analysis techniques.
46
For use in prospective job analysis and design, in which muscle force activations that are estimable from volunteers would be inaccessible, a numerical solution from a prediction model is preferred over an experimental methodology, as required in an EMGdriven model. Consequently, an optimization-based model was created to predict the muscle force levels. While theoretically coincident with preceding numerical prediction modeling approaches (Van der Helm, 1994; Hogfors, 1987), the primary objective of the current work was to create an alternative computerized model, that is: 1) physiologically accurate, 2) inclusive of task dynamics, and 3) ultimately available for non-expert use when combined with other human motion and ergonomic analysis models. During the course of the iterative development of the shoulder model, assumptions and modeling techniques were selected based upon these criteria. This chapter serves as a description of the model components and the rationale that drove the course of their development.
3.2 Biomechanical Model Description The overall model consists of three major components (Figure 3.1): 1) A geometric model representing the bones and muscles of the shoulder; 2) An external dynamic shoulder torque model; and 3) An internal muscle force prediction model. While the geometric and external models are independent of one another, their combined outputs serve as the inputs to the force prediction model. All of the models were constructed and rendered in a commercially available software program, Matlab®. This allowed the construction of a modular structure within the overall program as well as for each subprogram, which will facilitate future model modifications as improved geometric and morphological data is subsequently generated and incorporated.
47
Shoulder Geometry Model
Geometric Properties
1
Motion Data
Subject & Task Data
Figure 3.1
3.2.1
Internal Muscle Force Model
External Dynamic Torque Model
2
3
Muscle Forces
Shoulder Torques
Data flow through the components of the shoulder model. Required inputs to the overall model include motion, subject and task data. Inputs are framed with dashed lines and outputs with double lines. Model segments are indicated by single-lined boxes.
A Geometric Model of the Shoulder Mechanism
Motivation and Previous Studies Due to its extensive range of motion, the geometry of the shoulder is complex when considered in three-dimensional space. This situation, from a modeling standpoint, is further complicated by the general lack of rigorous population data on in vivo muscle moment arm magnitude, muscle size, bone dimensional variation, and the relative motion of bony segments. This complexity does not negate the importance of accurately estimating the changing geometry of the shoulder mechanism when performing different tasks. In order to generate correct estimates of muscle activations, the geometry must be physiologically sound. To this end, several notable experiments have been performed to assess muscle geometric parameters on cadavers (Keuchle, 1994; Hogfors, 1987; Veeger, 1991; Liu, 1996; Breteler 1995), as well as scapular motion, both in vivo (de Groot, 1998
48
& 1999; Hogfors 1991; Karlsson 1992), as well as computationally (Makhsous, 1999). These prior studies were instrumental in providing several parameter values in the construction of the geometric model, as will be discussed.
Geometric Model Motion Data Inputs To meet the usability criteria for dynamic assessment of shoulder stresses it was necessary to formulate the geometric model to accept motion files as input. These files include the temporal Cartesian coordinates of specific body landmarks defined in a laboratory reference system. While the program currently accepts primarily HUMOSIM specific motion data, it can be adapted with modest revisions to use other motion files as drivers of the model. In addition, the geometric program can be implemented in an angle-defined manner, given that the orientations of the upper limb segments as well as the torso are known throughout a motion. These angles are defined by the Euler sequence described in the shoulder rhythm in this chapter, but conversion of other angles is also possible with minor changes.
Geometric Model Construction The geometric model will be described in five parts: 1) definition of bone parameters; 2) implementation of a shoulder rhythm algorithm; 3) muscle definitions; 4) muscle line-of-action construction; and 5) additional geometric definitions. The final output of this geometric model is the position and orientations of all the bones and joints, and the lines-of-action and moment arms for each muscle element in the model. As such,
49
these provide the foundation for all kinematics computations needed for the complete muscle shoulder model.
Definition of Bone Parameters Four bones are used to describe the shoulder mechanism, similar to a method described by Hogfors (1987). The coordinate systems described by Hogfors and his colleagues, however, depend on direct access to several internal bony landmarks that, while identifiable on a cadaveric specimen, are inaccessible in vivo, in particular the ends of bones such as the distal humerus and clavicle. All of our experimental motion data was collected on living subjects. As a result of this variation in the method of obtaining data, minor alterations were made to the previous system, which will be highlighted. The bony system consists primarily of the scapula, the clavicle, the humerus, and the torso, which includes both the dorsal spine as well as the ribcage. In addition, a combined radial/ulnar forearm link is included. Joints between the segments are treated as spherical joints with three degrees of rotational freedom, but no translational degrees of freedom. The exception to this general convention is the elbow joint, which is allowed only one degree of freedom (flexion/extension). Due to the inability to palpate and thus locate all of the landmarks used in the Hogfors model, a new method for replicating the coordinate systems of the five segments was created, and is described in the following section. In all instances, efforts were made to insure agreement between the segmental descriptions for later computational comparisons. All defined coordinate systems were orthonormal and right-oriented for the right shoulder.
50
Shoulder Model Coordinate Systems The geometric model contains six separate coordinate systems, five of which are used to describe individual segment orientations, and one that is necessary to facilitate the calculation of the shoulder rhythm. The six systems included in the model are: sternum, torso, clavicle, scapula, humerus, and forearm (Figure 3.2). The sternum system is used primarily in the calculation of the shoulder rhythm parameters.
Z
Z
Z X
Y X
Z
X
Y
Y
Y
Y
Z X Z
Figure 3.2
X
Y X
Local coordinate systems for the segments of the shoulder model. A. Sternum System, B. Torso System, C. Clavicle System, D. Humerus System, E. Forearm System, F. Scapula System (See text for details)
The sternum system was created using three thoracic landmarks: the sternoclavicular notch, the seventh cervical/first thoracic intervertebral space, and the
51
fifth lumbar/sacral intervertebral space, as palpated experimentally. From these external markers, the following internal locations were obtained: the superior surface of the first thoracic vertebra, the inferior surface of the twelfth thoracic vertebra, and the right sternoclavicular joint (SCJ). The origin of this system was the right SCJ. The x-y plane is oriented horizontal relative to the torso, with the positive y-axis directed anteriorly, the positive x-axis directed laterally to the right, and the positive z-axis directed cranially. The torso system was created using the same derived internal locations with one key variation. The system origin remains the right SCJ. The positive x-axis is still directed laterally, through the middle of both articulating surfaces. The x-y plane, however, now contains both the middle of the first thoracic vertebra, which represents a deviation of approximately 20 degrees forward tilt from the x-y plane of sternum system. As a result, the positive y-axis is thus oriented forwards and slightly downwards, while the positive z-axis is oriented slightly forward and upwards. The clavicle system shares the same origin as the torso system (SCJ), and was created based upon three locations: the SCJ, the acriomoclavicular (ACJ) joint, and the positive z-axis of the sternum system. The positive x-axis is through the center of the acrimoclavicular joint. Due to the inaccessibility of the interior joints in vivo, the other axes of the clavicle were estimated based on the sternum system. The positive z-axis is approximated as equal to the sternum positive z-axis in a reference posture of the clavicle, as it was impossible to measure clavicular axial rotation experimentally. Finally, the positive y-axis is orthogonal to these axes and is pointed forwards in the reference posture.
52
The scapula system has its origin at the ACJ, and was created using three bonefixed points, along with the ACJ position, and later oriented using the shoulder rhythm. The positive x-axis was directed from the ACJ to the inferior angle of the scapula. The xy plane contains the superior angle in its first quadrant, essentially meaning the positive y-axis points near the direction of the scapular spine. The positive z-axis orthogonal to these two axes, and has a generally anterior direction, depending on the placement of the scapula on the ribcage for a given posture. The humerus system has its origin at the center of the humeral head (HHC). This location is approximated experimentally using data from the acromion location combined with the orientation of the sternum system. The positive x-axis is directed through the elbow joint center (EJC). The positive z-axis of the humerus is directed along the lateral cross product of the long axes of the forearm and humeral segments. The positive y-axis is orthogonal to these two axes. The forearm system has its origin at the EJC. The positive x-axis is directed through the center of the wrist joint (WJC). The positive z-axis is equivalent to the positive z-axis of the humerus, and the y-axis is orthogonal to these vectors. While these definitions deviate slightly from those proscribed in the literature (Hogfors 1987), subsequent efforts are made to replicate the geometry seen physiologically, and to allow data obtained from a motion capture system to be used to configure the linkage system in space.
53
Bone Lengths The final parameter specified for each of the bones was their relative length. The length of the thoracic portion of the torso was defined as 70% of the distance between the first thoracic and fifth lumbar vertebrae, as measured with motion tracking. Based upon this estimate of thorax length, scaled segment lengths for each bone were calculated in the proportions described by Makhsous (1999). Alternatively, experimentally measured bone lengths can be used, however, these lengths change somewhat during the course of a movement due to marker movement. This complication led to the use of the scaled lengths which were placed according to the individual segment orientations obtained from experimental motion data.
Implementation of a Modified Shoulder Rhythm It has been well documented that a relationship exists between the positions of the scapula, humerus, clavicle and torso in the human body, and this relationship has been termed the ‘shoulder rhythm’ (Inman, 1944; Doody, 1970; Poppen and Walker, 1974). One shoulder rhythm implementation designed for the described segmental coordinate systems has been reported in the literature (Makhsous, 1999). It was extrapolated from previous reported rhythms (Hogfors, 1991; Karlsson, 1992). The model is predicated on the consistent relationship of relative movement between the humerus and the sternum system, as described by a set of Euler angles. Using the knowledge of the three Euler angles that describe the relative orientation of the humerus in the torso system, we can calculate estimates for the Euler angles for the clavicle and scapula in that system configuration. For the creation of the geometric model, the described shoulder rhythm
54
was implemented with a slight mathematical modification to accommodate the collection of experimental data. The modification was based on physiological considerations. Beyond the original shoulder rhythm, the scapula was constrained to maintain both the inferior and superior angles outside the surface of the ribcage, as defined by a cylindrical surface. This constraint was implemented through modification of the scapular Euler angle predictions sequentially until a physiologically valid position was found. The modification was necessitated by small variations associated with experimental data collection of tracked motion data on different subjects.
Definition of Euler Angles In this model, a (3, -2, 1) set of Euler angles was implemented. These angles are defined: α
=
rotation about the sternum positive z-axis (vertical)
β
=
rotation about the intermediate negative y-axis
δ
=
rotation about the secondary positive x-axis
The original form of the rhythm follows: Clavicle:
α c = −35.15 + 11.15 * cos[0.75 * ( β h + 90)] * (0.08 * α h ) β c = 9 + 18 * {1 − cos[0.8 * ( β h + 90]} γ c = 3 + 30 * {1 − cos[0.75 * ( β h + 90]}
55
(3.1)
Scapula:
α s = 200 + 20 * cos[0.75 * ( β h + 90)] β s = −87 + 42 * cos[−0.75 * β h − 70] * (0.1 * γ h / 90 + 1) γ s = 82 + 8 * cos{(α h + 10) * sin[0.75 * ( β h + 90)]}
(3.2)
(Subscripts refer to the segment for each angle: h = humerus; c = clavicle; s = scapula)
Muscle Definitions In our model, twenty-three muscles in the shoulder are represented. However, the model represents several of these muscles as having multiple mechanical contributors. The muscles modeled as having more than one line of action are as follows, with the number of elements shown parenthetically: latissimus dorsi (2), the serratus anterior (3), the trapezius (4), the subscapularis (3), the infraspinatus (2), the pectoralis major (2), the deltoid (3), the biceps (2), and the triceps (3). A more detailed list of these muscles is contained in Appendix F. An earlier cadaver study established segment-specific muscle attachment site averages for three specimens based on the coordinate systems previously defined in this model (Hogfors, 1987). These sites are represented on each bone mathematically by the location of the muscle attachment described as a fractional distance along each axis, relative to the segment length. Visually, these averages were combined with a geometric representation of the Visible Human dataset to confirm and finalize the muscle attachment sites on each bony component.
Muscle Line-of-Action Construction In general, muscle lines-of-action (L-O-As) are represented by the line connecting the two attachment sites designated for each given muscle. While this is adequate for
56
many muscles in the shoulder, improper lines-of-actions would be obtained for several muscles using this simple straight-line procedure. The complication arises from physiological L-O-As that pass around orthopedic barriers, specifically the ribcage and humeral head. In order to address these particular muscles, variations of spherical (van der Helm, 1994) and cylindrical (Charlton and Johnson, 2001) geometric muscle wrapping techniques were used to describe each affected muscle path. These procedure for making these modifications to the muscle L-O-As, for both cases, are described analytically below.
Spherical Muscle Wrapping The muscles wrapped using a spherical criterion were the following: muscles of the rotator cuff (supraspinatus, infraspinatus, subscapularis), and the deltoid. These muscles were all wrapped around a representative humeral head, which was defined as having a radius proportional to the length of the humerus (Makhsous, 1999). Before wrapping a given muscle line of action, a collision detection procedure is carried out for the given posture. Defining the initial line of action as the connecting vector between the muscle attachment points, M1 and M2, and knowing the location of the center of the sphere about which wrapping may occur (HHC for the case of humeral head wrapping) (Figure 3.3), the distance (D) between the sphere center and the line of action can be calculated through the standard 3D point-line distance equation: r r L2 × L1 r r r L1 = M 1 − M 2 ; L2 = HHC − M 2 ; D = L1
57
(3.3)
W2
W1 r α1
M1
α2
M2
HHC
W2’
W1’
Figure 3.3
Spherical Muscle Wrapping. Shown are the general case (left), and the specific application of wrapping for the middle deltoid around the humeral head. Particularly for the rotator cuff musculature, linear representations of muscle lines of actions are not physiologically consistent with their functions.
This distance, D, can then be compared to the radius of the wrapping surface to determine whether wrapping is required in the current geometric formation. If wrapping is warranted, the following equations are applied, given M1, M2, HHC, and the magnitude of the sphere radius, r. First we calculate the angle α1 using the following analytical relationship: ⎛
r ⎜ HHC − M 1 ⎝
α 1 = arcsin⎜
⎞ ⎟ ⎟ ⎠
(3.4)
Next we can solve for the vector (M1-W1), and thus also for W1 by rotating the vector (HHC-M1) through a rotation of α1 around the axis perpendicular to the plane (n) containing M1, M2, and HHC, and finally scale by cos( α1): r W1 = M 1 + R(n , α 1 ) * (HHC − M 1 ) * cos(α 1 )
(3.5)
W1’ can be found by rotating with an angle of - α1. The process can be repeated for W2 and W2’. The end result of this is two sets of wrapping points, one along each
58
side of the circle defined in the wrapping plane by the sphere. The selection of the appropriate choice of the two sets has been suggested as that which represents the shortest distance (van der Helm, 1992). In our model, physiological muscle paths are emphasized in this model, i.e. the posterior deltoid must always wrap around the posterior aspect of the humeral head, rather than through the shoulder joint space, despite the latter representing a shorter overall distance. This particular example would be most potentially true for right humeral postures obtained when reaching across the body (in the negative x-axis direction in the sternum system). The final calculated muscle L-O-As pass along the surface of the described sphere. The sphere representing the wrapping surface for the deltoids (Figure 3.4) is slightly larger than that for the muscles of the rotator cuff, as the deltoids are more superficial and pass over the deeper cuff muscles. Muscle Parts
Figure 3.4
View of spherical wrapping in the geometric model. In this figure, the three modeled parts of the deltoid are shown (3 dark lines indicated with leader arrows) wrapping around a sphere that represents an obstacle in their straight-line path from origin to insertion. Each muscle is specified to wrap over designated faces of the obstacles modeled by primitive geometric shapes.
59
Cylindrical Muscle Wrapping The cylindrical case is more common for wrapping around the ribcage of the torso, and is implemented in the model for the serratus anterior. As in the case for spherical wrapping, the first task is to determine if wrapping about a surface is necessary for a given position. In order to do this, a collision algorithm is implemented to determine if the line that joins the muscle attachments passes through the cylindrical shape. Knowing the two attachment points, it is possible to construct a line L, described
v v as L = P + u , where P is any point on the line and u is the vector description of the v orientation of L. By knowing the direction of the cylinder to be wrapped around v , and any point on the axis of the cylinder (Q), which are both necessary to define the cylinder,
v the line describing the cylinder C, can be expressed as C = Q + t v Given these lines, the distance is calculable with the following formula: → r r PQ×(u × v ) D= (ur × vr )
(3.6)
This distance, D, can be compared to the radius of the cylinder, r, to determine if wrapping is required. After geometrically deriving all of the values shown in Figure 3.5, the wrapping points can be determined using the following equations (see Charlton and Johnson, 2001 for a complete derivation):
60
r r r r r W1 = M 1 + l1 cos(α 1 ) * VM1C + l1 sin(α 1 ) * VM1P + h1 * VC r r r r W2 = M 2 + l 2 cos(α 2 ) * VM 2C + l1 sin(α 1 ) * VM 2 P − h2 * VC
h
M2
O
IM1 α
h
β
β f
e
α
l1
I
l1 Ψ
Vc r
θ
h1 MO 2
l2
EW1
(3.7)
W 2 EI
E
Pc
W 2 EI rθ
Ψ Ψ
W1
l2
M1 Longitudinal View Figure 3.5
Side View
Graphical representation of cylindrical muscle wrapping. Shown are the longitudinal, side and perspective views of wrapping a geodesic line around a given cylindrical shape.
Using these points, the path along the cylinder can be calculated using another specified coordinate system. Again, the direction of wrapping must be specified. The portions of the serratus anterior that are wrapped can include all three portions (Figure 3.6), the upper, middle, and lower serratus anterior.
61
Muscle Parts
Figure 3.6
Cylindrical wrapping of the serratus anterior (thick dark lines indicated by leader arrows). Each muscle L-O-A is first tested for collision with the wrapping surface, and then mathematically wrapped around the geodesic surface corresponding to the physiological muscle path direction.
Specification of Additional Geometric Parameters In addition to the muscles described, two scapulothoracic contact forces were also specified, to reflect the forces transmitted to the scapula from its contact with the posterior aspect of the ribcage. These locations were obtained from the proposed locations of Makhsous (1999). Ligaments were placed according to data published in the same study, but were not activated in the muscle force prediction model.
3.2.2
The External Dynamic Shoulder Torque Model
Motivation and Theoretical Basis The shoulder torque estimation model was conceived with an inverse dynamics methodology similar to techniques previously used to quantify human gait kinematics
62
(Vaughan, 1991). Several modifications were required to generate an analogous model for the upper extremities from that described for the lower extremities. Conceptually, this model stage is subdivided into four parts: 1) Definition of Segments and Segment Properties; 2) Linear Kinematics; 3) Angular Kinematics; and 4) Calculation of Joint Forces and Torques. The final output of this model is a continuous prediction of dynamic torque and force values caused by external forces throughout an exertion.
Dynamic Torque Model Inputs The dynamic torque model, similar to the geometric shoulder model, also requires body landmark positions through time in the form of a motion file. In addition to this, however, the torque model requires inputs that relate to the external forces that occur during the exertion. This necessitates knowledge of the weight of the subject, as well as any forces at the hand. These are essential components of the calculation of both intersegmental joint torques as well as joint forces that arise from external sources. This information was gathered experimentally for our laboratory studies, and is contained in a database accessed by the dynamic torque model. Alternatively, these values can be given to the model by direct means for simulation purposes.
Definition of Segments and Segment Properties In the external model, the upper extremity is modeled as three connected rigid bodies, arranged distally to proximally: 1) the hand; 2) the forearm; and 3) the upper arm. Joints connecting adjacent segments are considered to be spherical joints. Individual segment mass values and directional moments of inertia were calculated by
63
using the anthropometric data gathered as inputs into the previously reported regression equations of Zatsiorsky et al. (1993).
Linear Kinematics Determination of Joint Center Locations and Segment Center of Mass Locations Using joint center coordinates (for the shoulder, elbow, and wrist), the locations of the centers of mass (CMs) of each segment were calculated based upon the Center-ofMass/Segment-Length Ratios reported by Clauser et al. (1969).
Segment Center of Mass Velocity and Acceleration Calculation Velocity and acceleration of the segment centers of gravity were determined using numerical differentiation. The first and second derivatives of the displacement-time data can be simply expressed as the following:
∂xn x −x = x&n = n+1 n+1 ∂t 2∆t
(3.8)
and
x − 2 xn + xn−1 ∂ 2 xn = &x&n = n+1 (∆t ) 2 ∂t
(3.9)
Where x = a data input point, n = the nth sampled frame, and ∆t is the interframe time. Prior to performing the numerical differentiation, raw data were smoothed using a second order, 2-pass Butterworth low pass filter set to 6 Hz. Before filtering, an FFT analysis was performed to insure that the majority of the signal was in this range.
64
Angular Kinematics Defining local coordinate systems of the arm As mentioned, the arm was subdivided into three linked rigid bodies: the hand, the forearm, and the upper arm. The neutral position was defined as having the arm extended to the side, perpendicular to the torso, with palms turned downward. The linkage, along with local coordinate systems used in modeling is shown in Figure 5, both in the defined neutral posture (Figure 3.7A) as well as in a seated posture (Figure 3.7B). A local coordinate system was defined for each rigid body used to represent the arm (Table 3.1).
65
Figure 3.7
Segmental Coordinate Systems. A) Segmental systems in the reference posture; the positive x-axis for each segment is represented by the flag direction, the positive y-axis by the rectangle direction, and the positive zaxis by the cylinder direction. The x-axis has been moved from the joint centers for visualization purposes only, as it is coincident with segment long axes. B) Segmental orientation for a seated work position.
Table 3.1
Arm Segment Local Coordinate System Definitions. Represented in the defined neutral position
Body Segment Hand (1)
Neutral x-axis Line Connecting wrist joint center with grip center (rotation axis)
Neutral y-axis Line passing through wrist joint center and ulnar protuberance, perpendicular to xaxis (flexion axis)
Neutral z-axis Cross product of x and y axes, pointed cranially (ab/adduction axis)
Forearm (2)
Defined along line between elbow joint center and wrist joint center (rotation axis)
Cross product of x and z-axes, directed superiorly (ab/adduction axis)
Perpendicular of forearm and upper arm x-axes, passes through humeral epheseus (flexion axis)
Upper Arm (3)
Defined along line Between glenohumeral joint and elbow joint center (rotation axis)
Cross product of x and z axes, directed anteriorly (ab/adduction axis)
Perpendicular of the forearm and upper arm x axes, passes through acromion (flexion axis)
66
Euler Rotational Sequence A series of rotations about segmental axes was applied to define segmental rotation. The first rotation is about the flexion/extension axis of the joint (ψ), the second rotation is about the abduction/adduction axis (θ), and the third rotation is about the longitudinal segment axis (φ). These are slightly different permutations for the three different arm segments. Two rotations are needed to gain meaningful angle results. The first, a 3-2-1 transformation was used to describe the orientations of the upper arm and forearm:
cosψ cos θ sin ψ cos θ − sin θ ⎤ ⎧ X ⎫ ⎧x ⎫ ⎡ ⎪ ⎪ ⎪ ⎪ ⎢ y ( sin ψ cos φ cos ψ sin θ sin φ ) (cos ψ cos φ sin ψ sin θ sin φ ) cos θ sin φ ⎥⎥ ⎨Y ⎬ = − + + ⎨ ⎬ ⎢ ⎪ z ⎪ ⎢ (sin ψ sin φ + cos ψ sin θ cos φ ) (− cosψ sin φ + sin ψ sin θ cos φ ) cos θ cos φ ⎥ ⎪Z ⎪ ⎦⎩ ⎭ ⎩ ⎭ ⎣
(3.10)
The orientation of the hand, due to its different neutral orientation, was described by a [2-3-1] transformation matrix:
cos ψ cos θ sin θ ⎧x ⎫ ⎡ ⎪ ⎪ ⎢ ψ φ ψ θ φ θ cos φ y = (sin sin − cos sin cos ) cos ⎨ ⎬ ⎢ ⎪ z ⎪ ⎢(sin ψ cos φ + cos ψ sin θ sin φ ) − cos θ sin φ ⎩ ⎭ ⎣
− sin ψ cos θ ⎤⎧ X ⎫ ⎪ ⎪ (cos ψ sin φ + sin ψ sin θ cos φ )⎥⎥ ⎨Y ⎬ (cos ψ cos φ − sin ψ sin θ sin φ ) ⎥⎦ ⎪⎩Z ⎪⎭
(3.11)
In equations 2.3 and 2.4 X, Y, and Z are global 3-D coordinates and x, y, and z are local 3-D coordinates.
Determination of Segment Angular Velocity and Acceleration The segmental angular velocities and accelerations are determined through the classical mechanical methods (Vaughan et al., 1991). These quantities are functions of
67
the Euler angles and their first and second derivatives. The derivatives of the Euler angles were determined using the same finite differencing technique described for linear kinematics (Vaughan, 1991).
Calculation of External Joint Forces and Torques Calculation of Joint Forces The force equilibrium calculation, which is derived from the linear form of Newton’s second law of motion, can be applied for each segment using the following equilibrium equation for each segment:
∑ F = msegment × aCOM , segment
(3.12)
Where F = forces, m = mass of segment, and a = acceleration of segment COM. This can be customized for each segment. External forces are composed entirely of the external load in the hand and the weight of each of the segments. The small horizontal force required to depress the target button was considered negligible. Solving the resulting equation achieves the cumulative external joint load at each proximal joint segment.
Calculation of Dynamic Joint Torques The angular analog of Newton’s second law was applied to each segment, which is summarized as:
∑ M = H&
(3.13)
Where M = external torque and H& = rate of segmental angular momentum The rate of change of angular momentum can be calculated based upon the segmental moments of inertia and the segmental velocities and accelerations. External
68
torques are calculated based on the cross products of the produced forces and their lever (moment) arms. A more detailed summary of this technique can be found in the literature (Vaughan et al., 1991). Requiring segmental equilibrium, the torque can be found at the proximal end of each successive segment.
3.2.3
Muscle Force Prediction Model The third modeling stage of the overall biomechanical model is the muscle force
prediction model. The shoulder is an indeterminate mechanical system, as evidenced by the large number of muscle elements (38 in this model formulation) compared to the number of equilibrium conditions available (19). This indeterminacy led to the implementation of a numeric optimization solution to solve the load distribution problem amongst the muscles required to resist external torques while maintaining shoulder stability. The muscle force prediction program can be thought of as consisting of five key elements, all of which serve to delimit the solution space and achieve a unique solution to the load-sharing problem for each iteration: 1) mechanical equilibrium constraints; 2) force bounds for individual muscles; 3) a glenohumeral contact, nondislocation constraint; 4) an objective function to minimize in finding the global solution; and 5) a solution methodology.
69
Mechanical Equilibrium Constraints In this numerical optimization solution, there are nineteen equilibrium conditions, both in the angular and linear systems. There are six equilibrium equations for the glenohumeral, acrimoclavicular, and sternoclavicular joints. In addition, there is one equilibrium equation for the elbow flexion/extension torque. The general form of these constraints is analogous to those of the external torque model, with the introduction of several more contributors to both the linear and angular equilibrium formulations. The linear formulation, in the general case, is as follows:
∑F
+ J i −1 + J i = ∑ FE
m ,i
(3.14)
Where Fm,i are the muscles active on segment i, Ji-1 is the joint contact force on the distal joint, Ji is the joint contact force on the proximal joint, and FE are any external forces unaccounted for in the previous segmental calculations. This is of particular importance for the clavicle and scapula systems, which are considered to have zero mass and zero acceleration. The angular general case is also similar to the formulation in the external model:
∑ (ma
i
× Fm ,i ) + τ i −1 + τ i = ∑ τ E
(3.15)
Where mai is the moment arm of the ith muscle, and the torque terms are analogs of the force terms from the previous equation.
70
Force Bounds for Individual Muscles Muscle forces are restricted to act along the lines modeled by the geometric model of the shoulder. The forces are further constrained to remain positive, as they are not allowed to transmit force in tension. Thus the lower bound for all muscle tensile forces is zero. The maximum value of tensile force allowed in a muscle, Fm,i,max is assumed to be in proportion to the physiological cross-sectional area (PCSA) of the specific muscle. The baseline value for our model is estimated to be a specific tension of 88 N/cm2 (Wood, 1989). The values for individual PCSA for each muscle were obtained from the results of a cadaver study (Hogfors, 1987). The maximum force producible values are scaled based upon the experimentally measured strength of the subject in our model.
Glenohumeral Contact Force Constraint In existing models of the shoulder mechanism, the constraint placed on the glenohumeral joint contact force is that it be pointed into an ellipsoid that represents the glenoid (Makhsous, 1999; van der Helm, 1994). This is an approximation of the physiological shape of the glenoid fossa, which is asymmetric along both its anteriorposterior and inferior-superior axes. One unique feature of the current model formulation instead uses empirical data collected on the tolerable level of the shear/compression force in the glenohumeral joint (Lippett and Matsen, 1993). By parameterization of these data into linear form which provides the force associated with joint dislocation into eight equally spaced compass directions, three additional constraints (one in each global
71
direction) are added to the mechanical equilibrium and muscle force constraints previously mentioned. These constraints are of the form:
J gh =
∑c
i =1−8
i
* Si
given that: ci ≥ 0
(3.16)
Where Jgh = glenohumeral joint contact force, ci = stability coefficients, and Si = directional shoulder stability requirements. The values used for the directional stability requirements can be found in the table in Figure 3.8. They are calculated as follows: Si =
Fs Fc
(3.17)
Where Fs is the shear force in specified direction, i, and Fc is the compressive force into glenoid cavity.
0°
315°
45° 270° 90° 225° 180°
Figure 3.8
135°
Direction
Stability Ratio (%)
0
51
45
33
90
29
135
40
180
56
225
43
270
30
315
35
Directional Shoulder Stability Ratios. Ratios indicate the directional shear to compressive force joint dislocation tolerance. It should be noted that the glenoid fossa is not a perfect ellipse but is directionally asymmetric. 72
Objective Function of the Optimization Formulation Several different objective functions based on a variety of physiological rationales have been suggested in the biomechanics literature. A summary of many of these can be found in Dul et al. (1990). The standard objective function implemented in the model is currently the sum of the cubed muscle stresses, which has been used in different model formulations (Crowninshield and Brand, 1981). Mathematically:
⎛ fi Θ = ∑ ⎜⎜ i =1 ⎝ PCSAi 38
⎞ ⎟⎟ ⎠
3
(3.18)
Where Θ is the objective function, f i is the force prediction in an individual muscle, i, and PCSAi is the cross-sectional area of the same muscle i. This cost function provides similar results to cost functions suggested in the literature for shoulder modeling (Hogfors et al, 1987; Van der Helm, 1994). This function has the additional quality of being convex while promoting synergistic muscle sharing amongst agonistic muscles for a variety of loading conditions, compared to other models (Hughes, 1991). The modular structure of the model allows the future consideration of alternative objective functions, including a quadratic muscle force formulation (such as is used by Hogfors et al, 1987 and Van der Helm, 1994) with modest additional programming efforts if desired.
Solution Methodology The indeterminate nature of the mechanical system being modeled is solved sequentially at each time interval throughout a motion, and this solution is then used by the subsequent iteration as the initial conditions for the minimization routine.
73
The set of constraint equations are a standard minimization optimization form:
Minimize Θ s.t.
(3.19)
Ax = B
(linear equality constraints)
(3.20)
In this model there are twenty-two equality constraints, nineteen for mechanical equilibrium and three related to the glenohumeral joint contact force. Solving for sixty unknown variables results in the formation of a 22x60 matrix (A in equation 3.20) and a 22x1 matrix (B in equation 3.20). The forms of these constraint based matrices are shown in Table 3.2.
Table 3.2 Rows in A 1 2-4 5-7 8-10 11-13 14-16 17-19 20-22
Construction of the Optimization Constraint Equations Equilibrium condition Flexion/extension elbow torque equilibrium 3-D global humeral force equilibrium 3-D global humeral torque equilibrium 3-D global scapular force equilibrium 3-D global scapular torque equilibrium 3-D global clavicular force equilibrium 3-D global clavicular torque equilibrium 3-D glenohumeral stability equilibrium
Columns in A Torque producing capabilities of muscles with forearm insertions 1 for muscles with humeral insertions Torque producing capabilities of muscles with humeral insertions 1 for muscles with scapular insertions Torque producing capabilities of muscles with scapular insertions 1 for muscles with clavicular insertions Torque producing capabilities of muscles with clavicular insertions Directional glenohumeral stability ratios
74
Column in B Reactive elbow flexion/extension torque Directional GH reactive force Directional GH reactive torque Directional AC reactive force Directional AC reactive torque Directional SC reactive force Directional SC reactive force Directional GH contact force
In addition, the muscle force bounds indicated are coded as upper bound constraints. Unknown variables that are solved for in the model (x in Equation 3.20) are muscle forces levels (38), ligament forces (3), joint contact forces (9), scapulothoracic contact forces (2), and directional stability coefficients (8), for a total of 60. These solution values form a 60x1 matrix (x in equation 3.20) for each model evaluation. Ligament forces are not currently active in the model. They do not appear in the cost function that determines the load sharing and are most active in extreme range of motion exertions (Lippitt et al.,1993), which were not of the type studied in this work. The model allows these to be incorporated in the future if appropriate for a given analysis.
3.3 Model Outputs 3.3.1
Outputs of the Geometric Model of the Shoulder The principal outputs of the geometric model are the instantaneous orientation
and relative positions of each shoulder bone and muscle attachment sites on each of the bones. In addition, muscle lines of action are numerically determined and act as critical inputs to the muscle force prediction model. The inclusion of wrapped and unaltered muscle results in a geometric model of the shoulder mechanism that has thirty-eight muscle components (Figure 3.9). These can also be visualized using custom designed graphing functions contained within the model. Contact between muscle-tendon line-ofaction and bony surfaces can be visualized for various postures and anthropometry. Derivative quantities, such as muscle length, can also be estimated using the geometric model if desired, with modifications.
75
Figure 3.9
Graphical Representation of the Right Shoulder Mechanism. Muscles are modeled as connecting strings between defined bony attachment sites, with some muscles shown as wrapped around idealized orthopedic surfaces to enhance the realism of the muscle lines of action.
3.3.2 Outputs of the External Dynamic Shoulder Torque Model The major outputs of the dynamic torque model are time-dependent values for both the external torques (Figure 3.10) at each joint of the upper limb (wrist, elbow, and shoulder), as well as the forces in the joints due to external factors. The model incorporates the effects of motion in both the angular and linear dimensions, and as a result produces torque values that are more descriptive than those currently obtainable with static torque calculations. Torque levels in the shoulder and the elbow, as well as the resultant external joint force in the shoulder are required to perform the muscle force prediction associated with the third modeling step.
76
Figure 3.10
Temporal representation of global external dynamic shoulder torques. These torques are expressed in the laboratory coordinate system (+x lateral; +y anterior; +z superior). Torques about axes are shown as follows: x – solid line; y – dotted line; z – dash-dot line. The torques shown are for a task in which the subject reached to a far target located directly in front of them with a moderate hand load, which explains the high torque about the x-axis and minimal torques about the other axes.
The torques and forces are currently obtained with respect to a laboratory reference frame with a seat frame as its basis. This facilitates their later integration into the optimization solver. Torques can also be represented in a given coordinate system through multiplication by an appropriate transformation matrix provided for each of the local joint coordinate systems described earlier.
77
3.3.3
Outputs of the Muscle Force Prediction Model The culmination of this modeling effort is the prediction of individual muscle
forces at distinct time points throughout a dynamic exertion. These results can be expressed in two ways: 1) absolute values for forces produced in a given muscle element (in Newtons) (Figure 3.11A), and 2) normalized muscle forces (percentages of maximum predicted muscle force) (Figure 3.11B). These graphics are intended to represent the changing magnitudes of all of the muscle forces throughout a trial. Individual muscles can be visualized independently as well. Functionally, the end result of the muscle force model calculations for a trail with n time instants is an nx60 matrix of numeric values describing individual muscle forces, joint contact forces, scapulothoracic contact forces, and glenohumeral directional stability coefficients.
Figure 3.11
Sample muscle force predictions generated by the model for a forward reach with a 2.4 kg mass in the hand. A) Muscle Force in Newtons; B) Muscle force as a percentage of capability. At the beginning and end of the trial, force levels are somewhat lower. As the individual moves to the target, the force levels increase to a steady level which represents the phase of the reach where the subject remains in the extended position, after which force levels decrease with the return to the starting position. It should be noted that the relative magnitude of the curves changes when the force producing capability of the muscles is considered. In these plots, 38 traces are shown for each of the muscle forces predicted. Another representation of an individual time instant can be seen in Figure 4.9.
78
When determining the levels of muscle forces required in reaction to the external shoulder torques and for stability, all calculations are performed using absolute force values. To compare performance of a task between individuals, however, it is often more attractive to present the normalized values. The latter data format is also useful in making a comparison between muscles, as a given force level represents a different proportion of muscular potential for different muscles. In other words, one measure (absolute force) identifies the major contributors to producing a required torque, while the other (normalized force) identifies those muscles that may be approaching their predicted maximum capabilities.
3.4
Discussion
3.4.1
Strengths of the Biomechanical Shoulder Model The specific novel strengths of this newly formulated model can be classified into
two major groups: 1) benefits to ergonomic analysis and 2) increased physiological realism. There are seven key benefits to ergonomic analysis that this model provides. Imperative to the utility of an ergonomic tool is its applicability to a diverse population, rather than for a specific group of individuals, such as clinical research may target. The developed model addresses this by having model scalability in two important areas. The first is the overall scalability of the geometry of the shoulder. Both bone lengths and muscle attachment sites are scalable to overall anthropometry. This allows the simulation of specific absolute loading/demographic work task conditions for a wide range of anthropometries (for example, it is capable of estimating the difference in specific muscle tissue loading between a 1.8 m tall male and a 1.5 m female when lifting 5 kg from a
79
workbench to an assembly line workstation). In addition to this geometric scaling, the model also scales for strength in determining muscle capability, rather than using a generic cross-sectional value. This addresses documented variations in population shoulder strength numerically (Stobbe, 1982; Kumar, 1991). It is clear that many industrial tasks have a dynamic component to them. One distinct advantage of the biomechanical model developed in this research is the inclusion of dynamic terms in the estimation of the shoulder joint torques. These dynamic contributions also influence the muscle force prediction estimates, which are based directly upon the torques calculated. These first steps towards a fully dynamic model are critical in replicating a large subset of industrial tasks. Three programming structural aspects of the model enhance its performance as an ergonomic analysis tool. First, the computational muscle force distribution optimization program was designed to yield a single global solution. This was accomplished through the use of an objective function and constraints that were convex. This deterministic nature is critical in achieving consistent results for simulations of workstations and ergonomic interventions. The three major components of the model (external torque, internal geometry, muscle force), were designed in a modular framework in an open programming language, Matlab®. This will facilitate the enhancement of the model as more complete data becomes available regarding specific tissue properties and their interactions with motion, posture, and task demands. It also allows variation of model parameters for which some uncertainty still exists, such as the magnitude of the glenohumeral stability constraint. This high level of flexibility is difficult to achieve with models that are in a compiled format that do not allow direct source code editing.
80
Finally, the creation of the models in a widely available commercial package results in a level of accessibility previously unachieved to both the overall model as well as to its components. Another benefit of the current model design is its intentional interface with a model of perceived effort in the shoulder (see Chapters II and V for details of this modeling effort). Beyond generating metrics of general and specific muscle tissue loading in the shoulder, this also allows the a priori estimation of the effects of the physical workload on the person performing the job. This adds a new layer of functionality to the models, as they allow an ergonomist to design work tasks to be both safer as well as more comfortable, to a given tolerance level. Currently described models of the shoulder do not incorporate this type of psychophysical metric as part of their output, though it has been demonstrated to relate to increased occupational risk of undesirable shoulder musculoskeletal outcomes (Yeung et al, 2003). A final contribution of the model to ergonomic analysis is the generation of a realistic geometric model of the human shoulder. The graphical representation of previous models is indistinct (Makhsous, 1999), and for novice users is essentially impossible to interpret. Inclusion of photorealistic orthopedic structures allows the user to envision the potential contributions of the shoulder musculature in a given posture, as well as a means to understand the complex interplay between the bones in the shoulder girdle during a motion. This level of clarity is unprecedented in available model packages. Beyond the contributions of the model to ergonomic analysis, there are also two major improvements to the realistic portrayal of the physiological behavior of the
81
shoulder musculature that are implemented in the model. The first novel aspect is the inclusion of an anisotropic, empirically-derived glenohumeral stability index. This index is an improvement in that it represents the directional non-dislocation requirements of the glenohumeral interface in a quantitative, non-heuristic way. Previous attempts to address the issue of glenohumeral stability mathematically have included using ellipses to model a cone into which the glenhumeral contact force was constrained to be directed (van der Helm, 1994; Hogfors 1987; Niemi, 1996). Rather than apply these heuristic measures, which assume a non-physiological glenoid elliptical shape, a linear formulation of the empirical asymmetric stability requirements was created. This formulation has the added advantage of creating convex constraints to insure a global solution. These are not similarly obtained with an ellipsoidal stability assumption. A second mathematical component of the model implemented in this research is the concept of collision detection when determining muscle wrapping requirements. This explicit description of a technique to determine if modification of the muscle line-of action to respond to orthopedic obstructions is needed adds to the models realism. Though the modeling of muscles as strings has inherent limitations, as will be discussed, the ability of the model to respond to variations in posture and impose wrapping at appropriate times enhances the model’s consistency and also may help to identify insufficiencies in wrapping algorithms.
3.4.2
Limitations of the Biomechanical Model Although the model created does include many favorable aspects for ergonomic
analysis and model utility, there are also several limitations to facets of the mathematical
82
model. These limitations originate from variable sources, but can be summarized as relating to two key areas: 1) possible under representing of physiological phenomena and 2) potential insufficiency in characterizing the level of biological variability in a population. These issues will be addressed separately, along with the rationale for their presence in the final model produced. The study of human physiology requires the acknowledgement that many factors combine to achieve the high level of function and integration exhibited by the musculoskeletal system. The shoulder mechanism provides an excellent opportunity to appreciate the level of complexity involved in living tissue systems. The nature of the mechanical behavior of muscle tissue exposed to different loading conditions has attracted significant attention by many scientists. Two major outcomes of this research have been concepts known as the length-tension and velocity-tension relationships. In the current model of the shoulder, these relationships are not considered. In doing this, the muscle is assumed to have the same amount of force-producing capability regardless of its instantaneous length and contraction velocity. As a result, these effects are not present in the model and their absence may somewhat compromise model predictions for particular postures or rapid exertions. For the most part, the exertions studied experimentally as a part of this research occurred at a relatively low velocity, so the velocity-tension effect is thought to be minimal, however the length-tension effect may be more pronounced, though unlikely, as variations in the shape of the length-tension curve caused changes in model predictions by less than 0.9% in a simulated study (Nieminen et al., 1995).
83
Another issue related to physiology that may be questioned is the exclusion of soft tissue contributions in the model. The decision to exclude ligamentous and capsular contributions was based on the type of exertions performed. In order to allow the large range of motion demonstrated by the shoulder, ligaments must have a sufficient level of intrinsic laxity. This leads to the conclusion that they primarily provide stability at the extremes of glenohumeral motion (Lippitt et al., 1993; Terry et al., 1991). The types of exertions that were studied in this research were loaded reaches to points in space that were not at the extremes of the subjects’ demonstrated capabilities. Due to the nonparticipation of the ligamentous forces in the objective function calculation, it was decided that these quantities would more likely skew the results towards an incorrect profile of muscle loading. A second mechanism is in place to account for potential contributions of the soft tissues however: the glenohumeral stability constraint. By assuming that there is some level of capsular and soft tissue contribution to shoulder stability, a less conservative metric can be applied, in that muscle contributions to stability may not need to satisfy the stability requirement alone. This can be accomplished mathematically by varying the degree to which the stability constraint is activated. Regardless, it may be the case that inclusion of these soft tissue contributions would permit further insight into the mechanisms of load sharing, stability, and force distributions amongst both muscle and connective shoulder tissues. An additional concept that has not been integrated into the shoulder model is the concept of arm stiffness. Stiffness may be a result of synergistic activation of antagonistic muscles when performing a task that requires precision, such as reaching to a given target in space (Hogan, 1990; Winters, 1990). In the current model, there is no
84
implementation of a stiffness constraint for the humerus. Stiffness constraints have been shown to effect the predictions of a similar optimization based shoulder model (Nieminen et al., 1996). The general effect was to increase the levels of several of the muscles. A limited amount of redistribution of the pattern of activated muscles was also demonstrated, but only for two simulation cases and no generalization across their data set was made. However, it is reasonable to believe that inclusion of a stiffness constraint on the humerus would affect predictions of muscle activity and could impact the accuracy of model predictions. The complication with modeling arm stiffness would be an accompanying assumption concerning its uniform requirement for different postures and task requirements. Hence, stiffness was not included in the model developed for this research. In the biomechanical model developed, muscles are represented by strings that at span between defined insertion points on bones. Although these strings are altered to respect the obstruction to muscle paths caused by bones, there has been concern expressed that dividing muscles a priori into functional mechanical units may be presumptuous (van der Helm, 1994). The major advantage to implementing the muscle data used in this model, which were obtained from the literature (Hogfors et al., 1987) was its scalability to link length. For muscles with relatively small attachment sites, a single string may do well in modeling the mechanical ability to produce torque by that muscle. A methodology for defining elements of more complicated broad muscles has been suggested by van der Helm (1991). Either solution results in complications with two issues: 1) Are muscle functional units actually functionally separate?, and 2) Are the cross-sectional areas of muscle functional units reliably determined across a population?
85
In this model, the assumption that a line or several lines can represent the mechanical function of a muscle is made. This method has been used in the major available 3-D models of the shoulder mechanism (Hogfors, 1987; van der Helm, 1994), so it has support in the literature. Suggestions have been made that it would be more appropriate to use a volumetric approach to modeling muscles (Nieminen, 1996), but to date no detailed technique to do so in the shoulder has been promoted. This particular method would be exceedingly difficult, as more information regarding relative tendon/muscle length for a population would necessarily have to be obtained. The model does not currently allow any translation of the glenohumeral head. The humeral head has been shown to translation during arm elevation and abduction (Graichen et al.,2000; Poppen and Walker, 1976), but there have been counter examples that for certain postures and tasks, there is little translation of the humeral head (Howell et al, 1988; Deutsch et al., 1996). Implementation of a soft constraint that allows humeral head translation may impact force predictions. Based on one study, however, variation of the distance between the glenohumeral joint center and the acromion has a negligible effect on predicted muscle forces (Nieminen et al.,1995). This motivated the exclusion of glenohumeral translation from our model. It should be noted, however, that if one were interested in studying pathological cases, it may be useful to allow a fairly large degree of translation to occur. Finally, the concept of a physiological cost function that can be implemented mathematically may be flawed. Theoretically speaking, if the sole objective during task performance were the minimization of a level of mechanical stress only, then no antagonistic co-contraction would occur, provided that monotonically increasing
86
objective functions are used. This is related to the penalization by these models of each additional increment of muscle force produced (Hughes et al., 1995). Hence, an objective function such as the one implemented may not fully address the prediction of antagonistic co-contraction. Several researchers have attempted to solve this problem using a combination of EMG and optimization to predict muscular activity (Cholewicki et al., 1995; Marras et al., 19). Using EMG in this model was not practical as the model must function based only upon position and subject data, in order to be implemented as a prospective ergonomic device. However, these models do provide a means by which to obtain results that both satisfy defined equilibrium requirements and reflect empirically gathered EMG data. Beyond physiological realism are other limitations that have their geneses in making the model scalable across a population. These assumptions were necessary to achieve a flexible model, but may discount the impact of biological variability on the pattern of muscle use by various individuals while performing a task. Firstly, a common average shoulder rhythm described in the literature (Hogfors et al., 1991; Makhsous, 1999) was applied to all subjects. Although the shoulder rhythm has been shown to be consistent for an individual, there have been interindividual differences noted (Hogfors et al., 1991). Using a common rhythm, therefore, may not truly reflect the relative scapular and clavicular motion of a given individual. This, however, is not thought to lead to systematic errors in model predictions (Hogfors et al., 1995). In addition, the shoulder rhythm adopted displays inconsistent results for some overhead exertions. Similar arguments can be made regarding the assumption that the proportionality of the crosssectional areas of the shoulder muscles is constant across a population. In this model,
87
that assumption is made. However, it has been shown that variations in the total crosssectional area (CSA) of the shoulder muscles have a negligible effect on the results of an optimization model of the shoulder (Nieminen et al., 1995). Detailed data regarding the different ratios of the CSAs for a population are unavailable. Finally, a final scaling assumption made in the model is that the relative insertion locations of each muscle on each bone are similar for the population. Again, this is believed to have little impact on model predictions. However, due to the relatively small moment arms of several shoulder muscles, particularly those of the rotator cuff, any differences present may be magnified.
3.4.3
Perspective on the Utility of the Biomechanical Shoulder Model Our model represents a significant step forward towards the integration of tissue
biomechanics with ergonomic analysis in the shoulder. Though several models of the shoulder precede this current implementation (Hogfors, 1987; Van der Helm, 1994; Hughes and An, 1997; Laursen, 1998), they are not constructed or available in a form that can be readily used by practicing ergonomists. The current model allows a high degree of flexibility in implementation along with the ability to execute the programs on a common software platform, Matlab®. In addition to these practical novelties, the model also incorporates the integration of submodels to address the shoulder rhythm (Makhsous, 1999), cylindrical wrapping (Charlton and Johnson, 2001), the glenohumeral contact force direction and magnitude (Lippitt et al., 1993), as well as explicit explanations concerning the details of the model parameters. The major functional contribution of the constructed model is the production of a flexible ergonomic analysis and design tool for
88
the study of shoulder loading while performing a variety of physical exertions. With the model constructed, the next logical step is the systematic empirical evaluation of the designed model, which is described in Chapter IV.
89
3.5
References
Breteler MDK, Spoor C, Van der Helm FCT, “Measuring muscle and joint geometry parameters of a shoulder for modeling purposes”, J Biomech 32:1191-1197 (1995) Chaffin, D.B. Development of Computerized Human Static Strength Simulation Model for Job Design, Human Factors and Ergonomics in Manufacturing, 7(4):305-322, (1997) Charlton IW and Johnson GR, “Application of spherical and cylindrical wrapping algorithms in a musculoskeletal model of the upper limb”, J Biomech 34:1209-1216 (2001) Clauser CW McConville JT, and Young JW, “Weight, volume and Center of Mass of Segments of the Human Body, AMRL-TR-69-70, Aerospace Medical Research Laboratories, Dayton, Ohio, (1969) de Groot JH, “The scapulo-humeral rhythm: effects of 2-D roentgen projection”, Clin Biomech 14:63-68 (1998) de Groot JH, Valstar ER, and Arwert HJ, “Velocity effects on the scapulo-humeral rhythm”, Clin Biomech 13:593-602 (1999) Doody SG, Freedman L, Waterland JC, “Shoulder movements during abduction in the scapular plane”, Archives Physical Medical Rehabilitation 51:595-604 (1970) Garner BA and Pandy MG, “Musculoskeletal model of the upper limb based on the visible human male dataset”, Computer Methods in Biomechanics and Biomedical Engineering 4:93-126 (2001) Hogfors C, Sigholm G, Herberts P, “Biomechanical model of the human shoulder – I. Elements”, J Biomech 20:157-166, (1987) Hogfors C, Peterson B, Sigholm G, Herberts P, “Biomechanical model of the human shoulder – II. The shoulder rhythm”, J Biomech 24:699-709 (1991) Hughes RE, “Empirical evaluation of optimization-based lumbar muscle force prediction models”, Doctoral Dissertation, University of Michigan (1991) Hughes, R.E. and K.N. An, “Monte Carlo Simulation of a Planar Shoulder Model,” Medical & Biological Engineering & Computing, 9:544-548 (1997) Inman VT, Saunders JB, Abbott LC, “Observations on the function of the shoulder joint”, J. Bone Surgery 26:1-30 (1944)
90
Karlsson, D. and B. Peterson, “Towards a Model for Force Predictions in the Human Shoulder,” J Biomech 25(2):189-199 (1992) Keuchle DK, Newman SR, Itoi E, Morrey BF, An KN, “Shoulder muscle moment arms during horizontal flexion and elevation”, J of Shoulder and Elbow Surgery 9:429-439 (1997) Latko WA, Armstrong TJ, Foulke JA, Herrin GD, “Development and evaluation of an observational method for assessing repetition in hand tasks,” AIHA, 58:278-285 (1997) Laursen B, Jensen BR, Nemeth G, Sjogaard G, “A model predicting individual shoulder muscle forces based on relationship between electromyographic and 3D external forces in static position”, J. Biomech 31:731-739 (1998) Laursen B, Sogaard B, Sjogaard G, “Biomechanical model predicting electromyographic activity in three shoulder muscles from 3D kinematics and external forces during cleaning work”, Clin Biomech 18:287-295 (2003) Lippitt S and Matsen F, “Mechanisms of glenohumeral joint stability”, Clinical Orthopedics and Related Research 291:20-8 (1993) Liu J, Hughes RE, Smutz WP, Niebur G, An K-N, “Roles of deltoid and rotator cuff muscles in shoulder elevation”, Clin Biomech 12:23-38 (1997) Makhsous M, “Improvements, Validation and Adaptation of a Shoulder Model”, Doctoral Dissertation, Chalmers University of Technology, Gothenburg, Sweden (1999) McAtamney L, Corlett EN, “RULA: A survey method for the investigation of workrelated upper limb disorders,” Applied Ergonomics 24(2):91-99 (1993) Moore JS and Garg A, “The strain index: A proposed method to analyze jobs for risk of distal upper extremity disorders”, AIHA 56(5):443-458 (1995) Poppen NK and PS Walker, “Normal and abnormal motion of the shoulder”, J Bone & Joint Surgery 58A:195-201 (1974) Terry GC, Hammon D, France P, Norwood LA, “The stabilizing function of passive shoulder restraints”, Am J. Sports Med 19:26-34 (1991) Van der Helm FCT, “A finite-element musculoskeletal model of the shoulder mechanism”, J Biomech 5:551-569 (1994) Vaughan CL, Davis BL, O’Connor JC, Dynamics of Human Gait, Human Kinetics, (1992)
91
Veeger HEJ, Van der Helm FCT, Van der Woulde LHV, Pronk GM, Rozendal RH, “Inertia and muscle contraction parameters for musculoskeletal modeling of the shoulder mechanism”, J Biomech 24:615-629 (1991) Wood JE, Meek SG, Jacobsen SC, “Quantification of human shoulder anatomy for prosthetic arm control – I. Surface Modeling”, J Biomech 22:273-292 (1989) Zatsiorsky V and Seluyanov V, “Estimation of the mass and inertia characteristics of the human body by means of the best predictive regression equations”, Biomechanics IXB, Human Kinetics, (1993)
92
CHAPTER IV A BIOMECHANICAL SHOULDER MODEL FOR ERGONOMIC ANALYSIS II. EMPIRICAL EVALUATION
4.1
Abstract Evaluation of biomechanical models of the shoulder has been limited to several
comparisons of static shoulder positions in constrained positions. While clinically relevant and imperative to the basic understanding of shoulder musculoskeletal function, assessment of the performance of shoulder mathematical models for more complex tasks has been limited. In this investigation, a methodology is proposed and applied to evaluate the performance of a biomechanical model of the shoulder based on optimization principles and defined physiological criteria. This evaluation was on a series of loaded reach tasks to targets in the right-handed reach envelope. Model performance was examined for both a static hold posture as well as for a dynamic reach to target exertion. The evaluation of the model showed that model predictions positively correlated with experimentally collected electromyographic (EMG) data for those muscles identified as prime movers for the task (deltoid, infraspinatus, and biceps), and varied somewhat between subjects and task parameters. Further, the model showed a relative inability to reliably predict the measured activity of other muscles that were shown to be active. The model did, however, largely predict inactivity for those muscles shown to be inactive by EMG. Variation of model parameters was shown to have an effect on the predictions of the mathematical model. The implementation of additional
93
mathematical constraints in the solving of the problem of the distribution of muscle forces may allow a more accurate prediction of muscular activity profiles in the shoulder.
4.2
Introduction High muscle forces have been cited as a risk factor in developing musculoskeletal
disorders in the upper extremity (Herberts, 1984). Thus, a reliable characterization of the stresses and forces present in muscles while performing various tasks would assist in the identification of potential hazardous loading conditions for the shoulder tissues. To address this need, several research groups have developed biomechanical models of the shoulder musculature (Dul, 1988; van der Helm et al., 1994a,1994b; Hogfors et al., 1987,1991,1995; Laursen, 1998, 2000; Charlton and Johnson, 2001; Garner and Pandy, 1999; Hughes et al., 1997). These models have concentrated primarily on the production of static muscle force predictions, for both planar and three-dimensional situations.
4.2.1
Review of Biomechanical Shoulder Model Development While these models provide valuable information regarding potential tissue
loading and shoulder kinematics, they are not specifically designed to be integrated with ergonomic analysis of task performance. Hence, a novel mathematical model of the shoulder mechanism was developed, based partially on an existing shoulder model (Hogfors et al., 1987), but with several unique characteristics. This model has three principal parts: 1) a dynamic shoulder torque calculator, 2) an internal geometry constructor, and 3) an optimization-based muscle force prediction algorithm. This model is explained in detail in Chapter III. The essential function of the model is to combine
94
motion data, anthropometric data, and specific physiological criteria to generate a unique distribution of forces in the shoulder tissues for a given motion and loading combination.
4.2.2
Objective of Study The objective of this investigation was to evaluate the performance of a
mathematical shoulder muscle force prediction model for a variety of different lifting and reaching tasks. The model predictions were compared with electromyographic data collected from 11 several muscles during the performance of 78 reaching tasks performed by 8 individuals.
4.2.3 Previous Electromyographic Shoulder Muscle Model Investigations Electromyography (EMG) has been used to study muscle activity in the shoulder by several researchers (Jarvholm et al., 1989; Nieminen et al., 1995; Meskers, 1999; de Groot et al, 2004). Some of these results have been less than impressive, with noncompelling relationships between predictions and measured EMG levels. It has even been suggested that EMG can only be used to verify on/off muscle patterns (van der Helm, 1994). In another study (Makhsous, 1999), the relative difference between estimated muscle activity and measured EMG ranged from 0% to over 200% for various muscles across averages for a group of 17 subjects. However, as there are no current feasible means by which to measure muscle forces in the shoulder directly, EMG activity remains a useful surrogate, albeit with complications (de Groot, 1999). The complexity of interpreting experimental EMG data, particularly for dynamic motions contributes to
95
the difficulty of the validation procedure (Basmainian and de Luca, 1987). As a result, a perspective on the use of dynamic EMG is given consideration in the discussion section.
4.3
Methods
4.3.1
Mathematical Muscle Force Model Description The developed mathematical shoulder model fuses geometric data from previous
studies (Hogfors, 1987; Makhsous, 1999) with a novel empirically-derived glenohumeral stability constraint (Lippitt and Matsen, 1993). This information is implemented in a set of equations that describe 22 assigned degrees of freedom in the shoulder. Finally, these 22 equations are provided as a set of equilibrium constraints for an optimization program to distribute forces amongst the muscles of the shoulder. A complete description of the composition of these equations is given in Chapter III. The model inputs are motion kinematics, subject, and task data, which are processed differentially through three cascading models (Figure 1), eventually yielding estimates of the force levels in all major identified shoulder muscles throughout the performance of a task.
96
Shoulder Geometry Model
Geometric Properties
1
Motion Data
Internal Muscle Force Model
Subject & Task Data
Figure 4.1
External Dynamic Torque Model
2
3
Muscle Forces
Shoulder Torques
Data flow through shoulder model components. Model inputs are framed with dashed lines and outputs with double lines. Intermediate model stages are indicated by single-lined boxes and encircled numbers.
4.3.2 Subject Description Eight college-age subjects (4 male, 4 female) participated in the validation experiment. All subjects were free from chronic shoulder musculoskeletal impairments and signed consent forms. Summary data describing the subjects are in Table 1.
Table 4.1. Summary subject characteristics. Gender Male Female
4.3.3
Mean S.D. Mean S.D.
N 4 4
Age (Years) 24.3 3.3 24.3 5.5
Stature (cm) 182.0 9.2 163.5 6.4
Weight (kg) 78.2 11.7 57.9 0.2
Strength (Nm) 52.32 11.0 30.0 0.9
Anthropometric and Maximum Reach Measurements A series of anthropometric measurements were taken for each subject, including
stature, bodyweight, and body segment dimensions. Each subject was also tested for
97
their maximum directional reach distance capability (lateral and forward), which was then used to scale target positioning, as explained later.
4.3.4
Subject Strength Testing The strength testing protocol explained in Chapter II was used to determine each
subject’s maximum extended arm flexion, abduction, and flexion/abduction force producing capabilities. These directional strength capabilities were averaged to produce an index of force-generating strength for each subject. The index was calculated as the mean maximum force produced over the three exertion directions. In addition, for each subject, an average torque strength value was derived, which incorporates the position and weight of the arm in addition to the force generated in the computation of maximum torque levels produced by each subject.
4.3.5
Experimental Task Description Each subject performed a randomized set of one-handed loaded reaches to various
targets in space. All experimental trials were performed with the right hand. Each trial had three distinct segments: 1) reach from beginning location to target; 2) static hold at the target; and 3) return from target to the beginning location. In the remainder of this chapter, these motion segments will be referred to as the reach, hold, and return phases. Each experimental trial lasted approximately 7-8 seconds. The ‘hold’ phase in the middle of the trial was held for a minimum of 3 seconds, with the balance of trial time representing motion between the beginning location and the target (the reach and return phases).
98
4.3.6
Variation of Reach Task Parameters Parameters for the reaching task were adjusted to facilitate a potentially high
range of muscle activations throughout the reach space. The target locations used were analogous to those explained in Chapters II of this dissertation. An objective of distributing the target locations in the right-handed reach envelope was to assess the consistency of the exposure/perception relationship dimensionally. In addition to testing a range of target locations, different hand load levels were used in order to generate a range of loading conditions. Table 5.2 summarizes the variation of these task parameters, three for the spatial variation of the target, and one for the force requirements of the task. Trials along the 0° elevation projection angle were omitted for the -30° and 0° azimuths, as these reaches were occluded by the simulated workbench used in the experiment. An example of a reach to a forward target with a hand load is shown in Figure 4.2.
Table 4.2 Dimension
Experimental Variation of Reach Task Parameters Description Values
Radial Azimuth
Target location along five azimuths, measured from sagittal plane, in degrees
-30° ,0°, 45°, 90°, 120°
5
Elevation Projection
Projection of target along line beginning at center of seat pan, expressed as degrees above horizontal
0°,35°,70°
3
Reach Distance
Percentage of maximum reach distance along specified azimuth and projection
60%, 85%
2
Hand load
Percentage of demonstrated maximum shoulder abduction/flexion strength
0%, 25%, 50%
3
99
Levels
Figure 4.2
4.3.7
Several views of an experimental reach for a given condition are shown. The exertion shown is a loaded reach to a target designated as (0, 70, 60) for azimuth, projection, and reach distance, respectively.
Experimental Setup During the experiment, several data streams were synchronously collected and
then integrated for input into the biomechanical model and inclusion in the model validation procedure. These data streams involved motion tracking, surface electromyography monitoring, as well as task characteristics, such as target position and hand load. Experimental task variations (Table 4.2) generated (5X3X2X3 = 90 – 2X3X2X2 = 78) 78 unique exertions tested per subject. 18 trial repetitions were also performed. Each subject performed the battery of 96 reaches in one session. Between subsequent trials, the target was positioned to the next location relative to the subject, resulting in a minimum of a thirty second rest period. The order of the trials was also randomized over the entire testing period to avoid fatigue due to repeating similar tasks in
100
order. The 96 reaches were also segmented into eight blocks of 12 reaches to allow recording system reset. This provided a minimum of a two minute interblock rest period.
4.3.8
Relative Target Position During the experiment the subjects were seated in an industrial-style chair which
rotated to position the target along the proscribed azimuths. The target was attached to an articulating arm that allowed precise placement both vertically and horizontally with respect to the center of the seat pan. The reach target was a small green pushbutton (2.5 cm diameter), which was located on the end of the articulating arm on a small cube (Figure 2). A small shelf, simulating a workbench, was located directly in front of the subject, allowing them to rest their upper extremity and any hand load prior to the execution of each trial.
4.3.9
Motion Tracking During the experimental movements, specific body landmarks were tracked using
a combination optical (MacReflex®) and electromagnetic (Flock of Birds®) system. The recording frequency during trials was 25 Hz. Sensors for both recording systems were placed bilaterally on the acromion processes, the lateral distal humeral tuberosities, the center of the dorsal side of the hand, the junction of the fifth lumbar vertebra and the sacrum, and the suprasternal notch (Figure 3). The relative positions of additional landmarks to the recorded six degree of freedom electromagnetic sensors were also recorded, specifically the junction between the seventh cervical and first thoracic vertebrae, the center of the wrist, and the right sternoclavicular joint. Control of the
101
collection of motion data was accomplished through the use of switches located on the workbench. Recording began at time point 1, when the weight or hand was lifted off of the simulated workbench starting location (shown as the dark disk on the workbench in Figure 2). Recording finished for each trial when the weight or hand returned to the workbench starting location, and thereby activating a switch connected to a digital trigger.
Figure 4.3
Tracked motion sensors for the experimental task. This paper focuses primarily on those sensors on the right side of the body and torso. Spheres in the figure indicate the placement of optical sensors, while squares likewise indicate the placement of electromagnetic sensors.
4.3.10 Electromyographic Measurements In addition to tracking the motions of the subjects while performing the reaches, surface electromyography was used to track the muscular activation of several accessible superficial shoulder muscles. Custom in-house fabricated silver-silver chloride pellet active bipolar electrodes of 4 mm diameter with a spacing of 1.5 cm were used. The skin was debreived using alcohol pads and the electrodes were affixed to the skin using
102
custom double-sided adhesive rings. The electrodes were further secured to their locations using athletic tape. A total of eleven muscle sites were monitored on the right (actively reaching) side of the body (Table 2), using placements described in the literature (Cram and Kasman, 1998). The recorded signals were preamplified at the source (gain of 100), band-pass filtered (10-1000 Hz). The root mean square (RMS) of the signal was determined using a time step of 55 ms. The RMS signal was recorded at 50 Hz. Prior to experimental trials, reference maximum voluntary contractions (MVCs) were performed in standard postures (Cram and Kasman, 1998) to determine the maximal activation of each muscle in that posture. EMG data recording was synchronized with motion recording with a digital trigger that began data collection when the switch on the work bench was activated and stopped collection when the switch was activated a second time. Table 4.3 Muscle sites monitored with surface EMG electrodes Muscle Electrode Position Latissimus Dorsi Approximately 4 cm below inferior scapular tip, halfway between spine and lateral torso edge Pectoralis Major, Approximately 2 cm medial from axillary fold, horizontal Sternal Insertion Pectoralis Major, 2 cm below the clavicle, medial to axillary fold and at an Clavicular Insertion oblique angle towards the clavicle Upper Trapezius Parallel to muscle fibers, along shoulder ridge, halfway between seventh cervical vertebra and acromion Lower Trapezius Approximately 5 cm below scapular spine, on medial edge, at 55-degree oblique angle, immediately lateral to spine Middle Deltoid On the lateral aspect of the arm, approximately 3 cm below the acromion, parallel to muscle fibers Posterior Deltoid Approximately 2 cm below the scapular spine, parallel to the muscle fibers at an oblique angle to the arm Anterior Deltoid Approximately 4 cm below the clavicle parallel to muscle fibers on the anterior aspect of the arm Infraspinatus Parallel to scapular spine, approximately 4 cm below and on the lateral aspect Biceps Brachii Parallel to muscle fibers and in the center of the muscle belly Triceps Brachii, Approximately 2 cm medial to arm midline, approximately Long Head halfway between acromion and olecranon
103
4.4
Data Analysis
4.4.1
Motion Data Processing Following collection of the raw spatial positions of the described marker set, the
data was processed through a series of algorithms to determine joint center locations based on a reduced set of markers (Nussbaum and Zhang, 2000). These data were filtered using a low pass digital filter with a cutoff frequency of 6 Hz. Prior to filtering, an FFT analysis was performed to identify the appropriateness of this cutoff frequency. The resulting file of joint center locations then served as input to the biomechanical model.
4.4.2
EMG Processing The average RMS EMG was next normalized to the reference MVC contractions
performed for each muscle trials to calculate a percent maximal voluntary contraction (%MVC) activation. These normalized EMG data were subsequently synchronized to the motion data to achieve a congruent data set.
4.4.3
Calculation of Muscle Force Predictions The biomechanical model was used to analyze the collected motion, subject, and
task data. The model, described in detail in Chapter III, was used for all trials available from the experiment. Occasional trials were excluded from the analysis due to irregularities in either motion tracking or EMG data streams. The model generated sequential muscle force predictions for each recorded time instant specified during an experimental trial. The model calculates muscle force estimates for 38 muscle functional
104
units, which represent 23 shoulder muscles. The muscles for which loading estimates are generated are listed in Appendix F. In addition, it provides an estimate of the net joint contact forces present at the glenohumeral, acrimoclavicular, and sternoclavicular joints. The predicted muscle force levels were further scaled based upon the demonstrated strength of each subject. In this calculation, the strength measured in the reference postures was assumed to be applicable to all shoulder muscles. The final proportional MFP levels, therefore, were calculated as follows:
nMFPi =
MFPi Fmax,i
(4.2)
Where nMFPi is the normalized muscle force prediction for muscle i, MFPi is the force predicted in muscle i in Newtons, and Fmax,i is the subject-specific maximum value of force production for muscle i.
4.4.4
MFP/EMG Comparison Method All trials began and concluded with the hand at the same starting position (Figure
4), but the target location was varied considerably. For the first comparison of model predictions and experimental data, the focus was placed on identifying the similarity of the muscle force predictions and the EMG measurements while maintaining body posture with the hand at the target position. One important element to note in the experimental is that the subjects were not instructed to constrain their torso movements, but rather to move ‘naturally’. The reasoning behind this was to generate motions that mimic true movements that would occur in a workplace when performing the same type of tasks.
105
The statistical comparison analyzed two of the three portions of the experimental trials: 1) the static hold phase and 2) the reach phase. Mathematically, each of these two
Figure 4.4
Trial Starting Position. The subject began each reach with their left hand located on the simulated workbench at a proscribed location. In addition,
comparisons consisted of three parts: 1) determining the time windows of interest for each task; 2) calculating the EMG activity and muscle force prediction (MFP) levels for muscles within these time windows; and 3) performing a statistical comparison for the EMG and MFP values for each muscle. These stages will be discussed for each of the movement phases analyzed.
106
4.4.5
Analysis of the Static Hold Phase: Correlation Coefficients The first analysis involved linear comparison of EMG and MFP levels during the
middle hold phase of the experimental trials. The level of EMG activity in the 11 recorded muscles was first summarized in a series of tables (Appendix B) and histograms (Appendix C). Based on these histograms, it was concluded that several muscles were primarily inactive during the hold phase and thus comparisons of their values would be fruitless. These muscles included the two parts of the pectoralis major, and the latissimus dorsi. This resulted in 8 muscles of primary interest. The muscles with the highest visible range of values were muscles that are primary abductors and elevators of the humerus in many positions, specifically the three parts of the deltoid and the two parts of the infraspinatus. These muscles especially displayed a large range of EMG values with respect to their MVC levels. For the task type performed, these muscles are believed to be the most critical, as the task consisted primarily of lifting and moving light to moderate handloads. The statistical analysis also expanded to include other moderately active muscles (two parts of the trapezius, the biceps and triceps brachii). The algorithm applied defines the relationship between the predicted and measured variables through a correlation coefficient “r” that describes their degree of linear association. As r approaches 1.0, the relationship is closer to linear. An ideal perfect relationship when comparing variables on the same scale would be r = 1.0, with a slope of 1.
Determining the Time Window for Static Hold Phase Analysis The time window used was centered temporally within the static hold phase. The hold phase was chosen as this time window represents the most extreme posture obtained
107
for each exertion, and the largest range of muscle activations between trials. In addition, the static nature of the hold phase eliminates many complications associated with the interpretation of dynamic EMG values. Each experimental trial lasted approximately 7 seconds, with the following average time breakdown: 1.5 seconds to reach to the target, 3 seconds of sustained hold at target, and 1.5 seconds to return to the beginning location. These times varied slightly between subjects and target locations and weights, with time variance similar to that documented previously for similar tasks (Faraway et al., 2001). The motions can be described by a torque-time profile that consists of three visually distinct phases that mirror the defined motion segments (Figure 5). A time window was next specified within the hold phase for data quantification.
A X
Reach ~1.5 sec
Hold 3 sec
Return ~1.5 sec Z Y
Figure 4.5
Time-torque curve for a sample reach. The torque is given with respect to three globally-defined axes. Within these curves, three distinct segments can be identified, signifying 1) the reach to target; 2) the hold at target; and 3) the return from target. “A” is the period used for the statistical comparison of model predictions and EMG data.
108
Calculating Hold Phase Average EMG and Muscle Force Prediction Levels The middle second of the intermediate phase (marked A in figure 5) was used to quantify the EMG and MFP values for comparison. This was done by computing the middle data point of each data file, and then including data from the preceding and subsequent ½ second intervals. The arithmetic mean of these values was calculated to arrive at a value for each of the two metrics. These values were calculated for all experimental data files, and combined to produce a matrix containing all corresponding muscle predicted forces and measured muscle EMG levels for the hold phase.
Construction of the Correlation Coefficient Matrix A correlation coefficient was constructed for each muscle and subject, stratified by azimuth. These were then compiled into a large matrix. In addition, three correlation coefficients were determined for three muscle unit groups: the composite trapezius; the composite deltoid; and the composite pectoralis major. These full matrices appear in Appendix A.
4.4.6 Analysis of the Dynamic Reach Phase: Muscle Activity Concordance Following the initial comparison for the hold phase of the exertion, an analysis was performed on the dynamic reach portion of the movement, which is shown as the first segment in Figure 5. This comparison was made to assess the ability of the model to identify those muscles which are demonstrated to be active by their recorded EMG levels. This analysis included the muscles previously identified as inactive at the static hold phase. A concordance model was selected for this analysis than a correlation
109
comparison, as the EMG levels of the muscles vary greatly with movement and posture, as do the predicted muscle forces. The concept of concordance relates to the consistency of the model to predict active and inactive muscular activity status. If both the prediction and measurement indicate that a muscle is active, or both the prediction and measurement indicate that a muscle is inactive, then concordance (active and inactive, respectively) is achieved. Conversely, if the two metrics do not agree, discordance exists. The concordance ratio is defined as a 1 X 4 matrix as shown in table 4.3. Both rows and columns sum to the total number of trials processed.
Table 4.3. Concordance matrix, per muscle Off Discordance: On Concordance: On Discordance: Number Off Number On Number On MFP/Number On MFP/Number Off MFP/Number On EMG EMG EMG
Off Concordance: Number Off MFP/Number Off EMG
For perfect agreement (ideal concordance), for each muscle the matrix would be:
Table 4.4. Matrix for perfect concordance 1.0 0.0 0.0 1.0
Approaching this standard indicates the congruence of the shoulder mathematical model predictions with measured muscular activity.
Determination of Time Window for Reach Phase Analysis The time window of each trial that represented the dynamic reach phase of the motion was determined based on the variation in the externally calculated shoulder torque
110
levels (an example of the torque-time curve is shown in Figure 2.6). The end of the reach phase is defined as the time point at which the torque maintained a level that was consistent for 200 ms, which typically occurred at about one-third of the overall trial duration.
Concordance Methodology The four cells of the matrix described in Table 3 were calculated for each of the 8 subjects over all 11 muscles that were monitored with EMG. A muscle was considered to be ‘on’ by MFP standards if the MFP level at any point during the dynamic reach was greater than 5% (Pedersen et al., 1987) of the muscle’s maximum force producing capability as estimated based on the strength distribution of the subjects. This was calculated for a given muscle, i, as follows: if
MFPi > 0.05, then MFPon = 1 MFPi ,max
conversely, if
(4.3)
MFPi < 0.05, then MFPoff = 1 MFPi ,max
(4.4)
For the EMG data to indicate that a muscle was ‘on’, a value of 5% over a measured resting activity level of the value recorded in a reference maximum voluntary contraction was used. This can be expressed by simply changing equations 4.2 and 4.3 to the following: if
EMGi > 0.05, then EMGon = 1 & EMGi ,max
(4.5)
if
EMGi < 0.05, then EMGoff = 1 EMGi ,max
(4.6)
111
This allows the calculation of the four statistical components of the concordance matrix to be determined for a given muscle, j: NT
Cell (1): On concordance
∑ MFP
on , j
C on , j =
1 NT
∑ EMG
(4.7)
on , j
1
NT
Cell (2): On discordance
∑ MFP
off , j
Don , j =
1 NT
∑ EMG
(4.8) on , j
1
NT
Cell (3): Off discordance
Doff , j =
∑ MFP
on , j
1 NT
∑ EMG
(4.9)
off , j
1
NT
Cell (4): Off concordance
C off , j =
∑ MFP
off , j
1 NT
∑ EMG
(4.10)
off , j
1
Further, it allows the calculation of several other metrics that describe both the patterns of muscle activations in the trials, as well as the overall accuracy of the model predictions: NT
PAEMG , j =
Proportion of EMG activity, muscle:
∑ EMG
on , j
1
NT
(4.11)
NT
PAMFP , j =
Proportion of MFP activity, muscle:
Concordance Ratio, muscle:
CR j =
(C (D
on , j
on , j
112
∑ MFP
on , j
1
+ Coff , j )
+ Doff , j )
NT
(4.12)
(4.13)
Note that if CR j < 1 , then the model is more discordant than concordant for a
given muscle, j. If CR j > 1 , then the model is more concordant than discordant. Ideally,
all values of CR j would be high. PA values indicate the level to which a particular muscle was shown to be active by muscle force predictions (MFP) and experimental monitoring (EMG).
4.5
Results
4.5.1
Shoulder Biomechanical Muscle Model Output
The muscle model outputs time-dependent muscle force predictions for thirtyeight muscle functional units at each time instant specified. Of these thirty-eight muscles for which predictions are made, representative measurements of 16 were monitored using surface EMG. A force-time plot for the middle deltoid for a representative trial is shown in figure 6. While continually active in this example, there is a clear increase in activity during both the dynamic reach as well as a sustained elevated level during the middle static hold phase. The coefficient of determination for repeated trials when comparing muscle force predictions over all muscles for the subject population was r2 = 0.89, indicating some variability in the predictions. This is in agreement with variations seen in the EMG data, the specifics of which are discussed in the next section.
113
Trial Midpoint
Static Hold Analysis Interval (1 sec)
Figure 4.6
Muscle force prediction throughout a trial, for one muscle. The analysis interval for the prediction/EMG comparison is shown as the middle second of the trial duration, as indicated by the time enclosed by the heavy vertical lines. This second is centered about the trial midpoint, and contains data from the intermediate hold stage of the trial exclusively.
4.5.2 Electromyographic Data Structure
As mentioned, dynamic surface electromyography was collected for eleven sites during the experimental trials. When compared to the data output of the muscle force prediction model, the recorded EMG data is less smooth due to noise and other factors. An example of an EMG trace for the middle deltoid for the same trial seen in figure 6 is shown in figure 7. Note that the empirical recordings of muscle EMG activity vary somewhat throughout the analysis window, while applied torques are relatively constant during the static hold position, as there are no inertial contributions to the torques during this phase. Comparisons were also made of repeated trial EMG similarity. Across all
114
measured muscle groups and subjects, the coefficient of determination was r2 = 0.86, indicating that a degree of natural variability in muscular activity during tasks.
Trial Midpoint Analysis Interval (1 sec)
Figure 4.7
4.5.3
Middle deltoid EMG activity during an experimental trial. (See text for filtering and integration values). The highlighted analysis interval corresponds to the middle second of the trial, when the subject was maintaining a posture while extended toward the target. Multi-step averaging is crucial to the interpretation of these curves, as the variation of the EMG throughout the exertion is evident.
Static Hold Correlation Coefficient Matrix
Following the extraction of the time-averaged values for both muscle force prediction and empirical EMG readings, comparisons were made for muscle concordance and correlation, of which those for the muscles most active are summarized (Table 4.5). The relationships shown are based on the additive total of the muscle subunits for each
115
respective muscle. Complete tables for all tested muscles, stratified by the azimuth direction of the target, are shown in Appendix A.
Table 4.5. Static correlation coefficients (r) for arm elevators, combined azimuths Subject 1 2 3 4 5 6 7 8 Average
Deltoid 0.52 0.55 0.69 0.61 0.48 0.47 0.54 0.41 0.53
Infraspinatus 0.70 0.82 0.77 0.28 0.53 0.56 0.65 0.71 0.63
A plot of the comparison between the theoretically calculated muscle forces and the recorded EMG values for one subject for 8 muscles and the total deltoid activity over all recorded trials is shown in Figure 8. The plot is limited to one subject to allow adequate visualization of the trends. Though there is some spread in the data, correlations are clearly more distinct for certain muscles. Correlations were significant within each subject, and the overall correlation for the agonists was also significant. A comparison of all recognized active muscles for each subject is contained in Appendix B.
116
Figure 4.8
4.5.4
Muscle force predictions plotted against electromyographical data for the 9 muscles and muscle parts for all trials for one subject. Relationships vary between the muscles. Additional differences in correlations were identified when the data was segmented by azimuth. These data are summarized in Appendix A.
Dynamic Reach Concordance Analysis
The dynamic reach concordance analysis resulted in the generation of large matrices of data for each subject. Results for one subject are shown in Table 4.6. Tables for all subjects are contained in Appendix E. The muscles shown to be most active by
117
EMG analysis were defined the ‘prime movers’. The concordance ratios are higher for those muscles that are prime movers in the task (including the three deltoid components and infraspinatus), but lower for other muscles, particularly the clavicular branch of pectoralis major. This disparity may be related to the mathematical structure of the developed optimization model, as will be discussed. The values in the matrix are also sensitive to the threshold values established for defining a muscle as “on” or “off”.
Table 4.6 Concordance Matrix Averaged Over Subjects
Muscle Name
Concordance Matrix On C
Latissimus Dorsi Pec Maj, Stern Pec Maj, Clav Lower Trapezius Upper Trapezius Middle Deltoid Posterior Deltoid Anterior Deltoid Infraspinatus Biceps Brachii Triceps Brachii Prime Movers
4.5.5
0.48 0.01 0.01 0.87 0.84 1.00 0.72 0.76 1.00 1.00 0.23 0.88
On D 0.52 0.99 0.99 0.13 0.16 0.00 0.28 0.24 0.00 0.00 0.77 0.12
Off D
Off C
0.27 0.00 0.00 0.69 0.58 1.00 0.58 0.60 1.00 1.00 0.28 0.79
0.73 1.00 1.00 0.31 0.42 0.00 0.42 0.40 0.00 0.00 0.72 0.21
EMG MFP Concordance On On Discordance Fraction Fraction Ratio 0.69 0.39 1.26 0.29 0.00 7.58 0.59 0.01 0.78 0.95 0.87 8.13 0.82 0.80 4.08 0.93 1.00 27.58 0.81 0.67 2.56 0.92 0.74 3.45 0.94 1.00 25.17 0.73 1.00 4.59 0.60 0.26 1.01 0.87 0.87 4.02
Influence of the Stability Parameter on Model Predictions
One important component of the new shoulder model is the implementation of the stability constraint derived from empirical shoulder dislocation data (Lippitt and Matsen, 1993). The importance of this constraint on the muscle force prediction algorithms was evident through means of a sensitivity analysis. A stability multiplier, ϑ , was applied to
118
the shear forces permissible in the glenohumeral joint with respect to a reference compressive force. This takes the form: Fscaled = ϑ * Fdislocation
(4.14)
Where 0 < ϑ < 1 Thus, as ϑ approaches 1, the shear force allowed per unit compressive force in the glenohumeral joint nears the directional dislocation shear force criteria. Conversely, as
ϑ approaches 0, the allowed shear force approaches zero as well. Thus, the smaller the value of ϑ is, the more conservative the stability metric applied is. It is unlikely that the shoulder is operated on the cusp of dislocation in common activities. If this were the case, dislocations would be far more rampant than they are amongst the population. As a result, several different levels of ϑ were tested for one subject to determine the impact of the stability constraint on muscle prediction. The first observation was that the stability constraint is active in the solution of the muscle force distribution problem, indicating an influence on predictions. Secondly, the model predictions are sensitive to the control metric, as predicted muscle forces increased with a more conservative requirement, and demonstrated changes in the specific muscles activated during a task. This can is represented by way of two examples of muscle force predictions for loaded reaches to different locations, but with the same hand load (Figure 4.9). For muscle definitions corresponding to the numbers on the graphs, please see Appendix F. A complete comparison of the stability-influenced predictions generated by a detailed sensitivity study to experimental EMG recordings must be performed to determine the appropriate setting value for ϑ . Inclusion of parameter estimations of the relative contributions of soft tissues to glenohumeral stability may enhance this value selection further. While
119
modeled in a slightly different manner, glenohumeral stability has been shown to affect muscle force prediction distributions in a similar optimization-based shoulder muscle force prediction models (Niemi et al, 1996).
120
A
B
Figure 4.9
Effect of Varying the Stability Multiplier, ϑ , on Muscle Force Predictions. ϑ was varied from 0.40-1.00 for two loading scenarios: A) A forward reach along the 0° azimuth; B) A reach to the right and backward along the 120° azimuth. Variation in both the active muscles and their magnitude of force production is evident, but more pronounced for the forward reach.
121
4.6
Discussion
Validation of any model is a complicated exercise. This exercise is made more difficult when the model to be validated is dimensionally complex, with many components and experimental comparison metrics that are difficult to estimate reliably. The muscles of the shoulder act in a concerted way to both resist torques and insure glenohumeral stability. To describe this behavior, a mathematical shoulder model was constructed that incorporated relative bone motion, muscle attachment data, a muscle stress based objective function, as well as limits on the force producible by muscles. In the construction of the model, many assumptions and existing data sets were used to acquire necessary model parameters, all of which can affect the model results. The implications of these required simplifications will be discussed in terms of their impact on the quality of the mathematical model performance. In addition, the unique challenges associated with experimental EMG collection and the nature of experimental model evaluation will be highlighted.
4.6.1
The Influence of Model Characteristics on Predicted Muscle Forces
Prediction of Synergistic Agonist Activity The implementation of optimization paradigms to solve indeterminate biomechanical load-sharing formulations is not a novel concept. The use of muscle stress as an objective function in optimization models is designed to activate the majority of agonists for a given task (Bean et al., 1988). In this model formulation, the sum of the cubed muscle stresses was used as the objective function. This objective function is attractive in that it has been shown to perform robustly across a range of similar exertions
122
in previous low back musculoskeletal models (Hughes et al., 1994). The current model also demonstrated a distribution of the force required to resist external torque loading amongst those muscles thought to be primary agonists (in particular, the deltoid and infraspinatus are nearly always concurrently activated in extended postures). This is consistent with previous conclusions that nonlinear cost functions distribute the required internal muscle forces widely whence compared to linear formulations. For this reason, the model is believed to behave sufficiently when predicting the activity of agonistic muscles, which are situated to most effectively respond to an external torque load. A complication in the shoulder region, however, is the degree of asymmetry and changing functions of muscles. This is particularly true for the elements of the rotator cuff, due to their small humeral moment arms and their movement along with humeral displacements. This has led to the execution of model evaluation experiments in which the movement of the arm is severely constrained (Meskers, 1999), in order to minimize this effect. While informative, these data cannot be easily extended to apply to generic postures such as those encountered in the workplace.
Prediction of Antagonistic Co-Contraction Activity One key criticism of the use of optimization models to predict muscle forces is their demonstrated inability to predict muscular co-contraction accurately. This tendency has been documented in optimization models of the low back (Hughes et al., 1995; Cholewicki et al., 1996; Gagnon et al., 2001). The reasoning behind the under prediction of antagonistic activity measured with EMG may be linked to their use of monotonically increasing objective functions. These functions penalize incremental
123
increases in muscle force (Hughes et al., 1995), and the objective function used in this model falls into that category. It is possible, however, that the assumed objective function does not exactly mimic the physiological pattern of muscle activation in the shoulder. It was thought that the empirical glenohumeral constraint implemented in this shoulder model may have elicited increased muscular co-contraction, as requiring stiffness has been shown to produce antagonistic activity in other body joints. However, the impact of changing this parameter mostly impacted load sharing amongst agonists (Section 4.5.5), without altering the pattern of overall muscles activated, in a similar fashion to a prior study (Nieminen et al., 1995). The relatively low linear correlations with measured EMG patterns for the antagonists over most subjects showed a consistent insufficiency in the model to accurately predict several muscles not identified as primary agonists. A major complication in interpreting the prediction of co-contracting muscles is that there is no taxonomy for which shoulder muscles are defined as agonists and antagonists for a given posture/load combination. Similar confusion has been noted in the low back region (Hughes, 1991). For instance, in the model the pectoralis major is almost entirely absent from the muscle force predictions. There is no definitive evidence, however, to say that this muscle was either an agonist or an antagonist for the entire duration of trials. This results in an inability to comment on the importance of these low correlations. The preponderance of zero muscle force predictions for several muscles that demonstrated EMG activity also makes it difficult to come to conclusions regarding the correlation coefficients obtained for those muscles (triceps, pectoralis major, and
124
latissimus dorsi). In general, based on these results, it can be stated that the model under predicts antagonistic co-contraction. A potential resolution to this inability could be the application of different glenohumeral stability or humeral stiffness requirements. Previous studies have shown that variation of an imposed stiffness constraint has an effect on model predictions (Nieminen et al., 1995), but did not comment on the impact on model performance in matching EMG measurements.
The Influence of Muscle Unit Definitions Another, equally important facet of the model which may contribute to the misestimate of produced muscle forces is the representation of the muscle mechanical functional units. The modeled muscles were divided a priori into a minimum set of functional units with respect to their mechanical contribution to torque production (Hogfors, 1987). In certain arm postures, however, this minimum set does fully reflect the flexibility of a given muscle to resist loads while maintaining a level of glenohumeral stability. A solution to this conundrum, however, is not elementary. Different methods of dividing muscles into subunits have been developed by several researchers (van der Helm, 1991; Garner and Pandy, 1999). There is no clear consensus on how this is most effectively performed, but the possibility exists that the current model is not sufficiently complex in terms of functional units to replicate all physiological shoulder function. This situation is difficult to resolve with the currently used data set, as it encompasses all of the generically scaled muscle attachment site data available. Further, there is no explicit guarantee that adding layers of complexity to the model would universally address this complication.
125
Variations in muscle lines-of-actions have been found to significantly impact the muscle force predictions made for lower back musculature as well (Nussbaum et al., 1996). In our model, an analysis was done for one subject where the infraspinatus muscle was removed from the model. Despite increasing the upper limits of the muscle bound to five times their initial values, the model returned infeasible solutions to a majority of the load situations. This underscores the necessity to adequately model key muscles in a mathematical shoulder model, and also is a testament to the critical mechanical contribution of the infraspinatus to reach tasks. No other combination of muscles was able to resolve the load sharing adequately. An additional facet within the muscle modeling approach used is the assumption of independence of muscle subunits, which has not been firmly established throughout the shoulder region. Finally, there a potential difficulty due to the limited data regarding the respective cross-sectional areas of this expanded set of muscle units, but small variations in this regard have been shown to have negligible effects on previous model formulation predictions (Nieminen et al, 1995).
4.6.2
Electromyography as an Evaluation Tool for Muscle Force Predictions
Currently, there is no feasible method by which to measure muscle force levels in the shoulder tissues in a living subject. This restriction necessitated the use of an alternative measure, surface muscle electromyography. Muscle EMG measures the electrical activity within a muscle, not muscle force directly. Experimentally, surface EMG data could only feasibly be collected on a fraction of the shoulder muscles. This is partially due to the inaccessibility of several of the deep muscles of the shoulder,
126
particularly those of the rotator cuff (subscapularis and supraspinatus), and partially due to the hindrance to natural motion that an excess of surface EMG electrodes and attached wires would create. A major criticism of experimental surface EMG has been shown to be sensitive to a number of factors, including muscle length, velocity of muscle shortening, fatigue, electrode placement, intersubject anatomical differences, muscle crosstalk and the non-trivial processing of the acquired signal (Basmanjian and DeLuca, 1989). Efforts were made to minimize the effects of these many factors on the quality of the EMG signal, including standardized electrode placements and MVC postures and exertions, but the use of surface EMG still remains an inexact measurement tool in assessing individual muscle activity. As such, all of the comparison values must be approached with the perspective that they reflect the ongoing muscle activity, albeit imperfectly.
EMG Population Comparison Complications The utility of an across-subject comparison of muscle force predictions and EMG activity is predicated on a consistency of normalization for these values. Although the MVC exertions were standardized for all subjects, it is difficult to guarantee that a maximal exertion was obtained for each subject on each test. Additionally, there is limited confidence that the same proportion of muscle activity as measured by EMG relative to MVC indicates the same level of force production in different individuals. (Cram and Kasman, 1998). As a result, it is difficult to make strong statements about the correlations for the entire subject pool, although it should be noted that the population correlations for the deltoid and infraspinatus are positive and significant (r = 0.53 deltoid;
127
r = 0.63 infraspinatus; both p < 0.01). In addition, the concordance ratio remained high for the most active muscles across subjects for the dynamic reach phases of the motion. Thus, the model consistently predicted a level of activity in muscles that were also shown to be highly active by the EMG recordings.
The Dependence of EMG on Body Posture Another complication arising from recording surface EMG activity in a variety of positions is the movement of the electrode relative to the muscle mass for which activity is recorded. As the muscle changes length and orientation with variation of arm posture, the electrode, which is firmly affixed to a specific skin location through means of adhesive, may be recording the activity of either a different part of the same muscle, or potentially even the activity of adjacent muscles (Cram and Kasman, 1998). Similarly, the logistical difficulty of obtaining MVC values for each muscle for each target position required the use of reference MVC data, collected for proscribed reference positions. This may have adversely affected the reliability of the consistency of the EMG recorded activity. Essentially, this means that for two postures the same level of EMG activity in a muscle may not correspond to the same level of force production in that muscle, and thus a different contribution to the resistance of torques and production of stabilizing forces. Hence, the EMG recordings must be interpreted with the knowledge that several confounding factors affect the collected data. Inclusive in this confounding is the inherent variability in experimental EMG. Repeated trials showed a coefficient of determination of r2 = 0.86, demonstrating that even a perfect model cannot explain all of the variance observed in the recordings.
128
4.6.3
The Influence of Experimental Parameters on Model Evaluation
The Influence of the Experimental Subject Pool on Empirical Data By design, the subject pool used in this study demonstrated a range of anthropometry, particularly for the male subjects. To ensure model congruency and to avoid incompatible geometry due to specific link scaling, a general model of segment scaling was employed. This essentially creates one scalable shoulder representation, which changes size based on the stature of the subject. The necessity for this implementation stems from the use of a common shoulder rhythm, which was created through previous work (Hogfors, 1991; Karlsson, 1992; Makhsous, 1999). In reality, it is probable that both segment lengths and shoulder rhythm vary independently throughout the population. Differences in the shoulder rhythm that may exist between subjects due to anthropometry are not demonstrable using the current model formulation. Similar difficulties have been noted when using a common shoulder rhythm across experimental subjects (Nieminen et al., 1995). A second assumption that underlies the current utility of the model for population analyses is the establishment of maximum muscle force producing upper bounds. Currently, this is set as a constraint across all subjects based upon a demonstrated strength level in a reference exertion. While an improvement over a single value, which has been suggested previously (Hogfors et al, 1987), this still may not be a sensitive enough metric to be considered robust. This contention is supported by findings that strength values can vary amongst the population on the order of 10:1! (Chaffin et al, 1999). Therefore, the maximum producible force limits may misestimate the true capabilities of a given person’s musculature. This is only relevant for those exertions for
129
which the required muscle forces approach the maximum defined level, which in this experiment was rare. A third critical unifying assumption in the model is the similarity of muscle attachment sites across the population. The model currently places muscle attachment sites on each bone with respect to the overall length of the bone. As an example, every subject is modeled as having their deltoid muscles attach to the humerus at the same proportional distance from the center of the humeral head (approximately 37% of humeral length from the humeral head center). While it is unlikely that differences in muscle attachment sites do occur in the population, the relatively small magnitude of the moment arms of the shoulder muscles (often < 2-3 cm) magnifies any inaccuracies that may be present. However, no direct measurements of moment arms were available for our study participants, so such an assumption is needed. The insensitivity of model predictions to small variations in moment arms has been suggested in the literature (Hogfors et al, 1995). All of these assumptions implemented to address a population may cause the model to be somewhat insensitive to the confounding nature of biological variability, but this sensitivity is thought to not be a primary source of error.
Task-Specific Model Evaluation A complicating aspect of evaluating the performance of a model is recognizing the influence of the conditions under which the confirmatory experiments occur. The shoulder musculature is responsible for a wide variety of directional force production and thus, variable associated muscle recruitment. In this experiment, the task consisted of the movement of loads with one hand in a defined space. The primary external force
130
involved in this task was gravity, both on the load in the hand, as well as on the weight of the body segments involved in placement of the hand. While ergonomically interesting tasks, the experimental trials represent a subset of the large range of forces and postures that the arm and hand can achieve. As a result, it is plausible only to make firm statements concerning the recruitment (or inactivity) of those muscles believed to be contributors and non-contributors to the given task. This task, however, is further complicated by the dependency of shoulder muscle function on relative humeral/scapular position. Due to the placement of the rotator cuff particularly on the humeral head, in a hemispherical pattern, depending on arm position the muscles can contribute to a variety of torque- and stability-producing activity. This flexibility allows the performance of a broad spectrum of activity, but the inherent asymmetry confounds the definition of muscles or muscle groups as being part of a group of agonists or antagonists without first defining the associated arm segment positions and orientations. However, for the tasks performed, two muscles, the deltoid and the infraspinatus, were considered to be agonists for the elevation of the loaded arm, and thus the validation comparison focused on the model predictions for these tissues. The relevance of these muscles to the performed task is evidenced by the relatively higher correlations (Table 4.5) for these muscles, as they demonstrated stronger correlations than other measured muscles. Although the correlations were not remarkably strong for the two muscles (deltoid range r = 0.37-0.69; infraspinatus range: r = 0.56-0.80), the relationships for these muscles were significant for all subjects. The magnitudes of the correlation coefficients degraded when comparisons were made across the subject pool, likely as the result of differential muscle use among subjects, which has been documented in the literature, even for the same
131
subject (Palmerud et al., 1995). In addition, the dynamic concordance analysis (Table 4.6 and Appendix E) showed much higher concordance ratios for those muscles believed to be the prime movers in the loaded reach tasks studied (average CR = 4.02 across subjects for prime mover group).
4.6.4
Next Steps for a More Robust Biomechanical Shoulder Model
The next necessary steps in enhancing the performance of the model, and indeed its universality, is further sensitivity analysis of the model parameters and empirical validation for a wider range of tasks. This methodology has been used previously to differentiate between the performance of various optimization algorithms on the prediction of muscle forces (Hughes, 1991). By specifying tasks that elicit particular muscle use, however, the model can be evaluated for more general ergonomic application. These tasks would involve horizontal and vertical push/pull exertions, as well as manual torque production, all of which occur during manufacturing work as well as in the course of daily activities. Directional principal action determination has been performed for constrained arm postures (de Groot et. al, 2004), and offers another potential method to evaluate the ability of a mathematical model to correctly predict activity in agonistic muscles for a given task. In essence, the current model can be viewed as a milestone along the path towards a more comprehensive mathematical model of the shoulder mechanism. Although it has demonstrable ability to replicate a subset of physiological activity, further investigations into the nature and specific setting of model parameters must occur, together with rigorous validation of more loading conditions. The model was designed with potential
132
enhancement in mind, and has been constructed in a modular format that allows adjustment and improvement of subcomponents easily. The end result is a useful tool with which to analyze complex loading scenarios that the shoulder tissues are subjected to, and an emergent opportunity to understand the intricacies of shoulder muscular function.
4.7
Conclusions
It is evident that the shoulder mechanism and shoulder function are complicated topics. The model developed in this thesis attempts to consider much of the known complexity. Concomitant with this complexity is difficulty in interpreting the output of the model with respect to specific tasks. In this study, validation data are presented for a set of industrial types of tasks, specifically spatial one-handed load transfer. The model has been shown to consistently predict activity for those muscles which are shown to be active through empirical measurements of muscle EMG activity. Further, the model has demonstrated significant positive correlations between predicted muscular activity and observed EMG activity for several of the most active muscles in reaching tasks (deltoid, infraspinatus, biceps). The model did not sufficiently predict co-contraction of antagonistic muscles for many of the reaches. This is partially due to intrinsic properties of the optimization model used, and partially due to the flexible functions of certain muscles. The model has been shown to be useful in the estimation of muscle forces for the dominant agonistic shoulder muscles active during industrial reaching tasks. With further enhancement, and the consideration of additional model parameters, the model
133
may be improved to estimate other contributors to motion, stability and stiffness in the shoulder.
134
4.8
References
Basmajian JV and de Luca C, Muscles Alive, Williams & Wilkins, Baltimore (1987) Bean JC, Chaffin DB, and Schultz AB, “Biomechanical model calculation of muscle contraction forces: a double linear programming method”, J Biomech 21:59-66 (1988) Chaffin DB, Andersson GBJ, and Martin BJ, Occupational Biomechanics, 3rd Edition, John Wiley, New York (1999) Charlton IW and Johnson GR, “Application of spherical and cylindrical wrapping algorithms in a musculoskeletal model of the upper limb”, J Biomech 34:1209-1216 (2001) Cholewicki J, McGill SM, and Norman RW, “Comparison of muscle forces and joint load from an optimization and EMG assisted lumbar spine model: towards development of a hybrid approach”, J Biomech 28:321-331 (1995) Cram JR and Kasman GS, Introduction to Surface Electromyography, Aspen, Gaithersburg, MD (1998) de Groot JH, “The shoulder: a kinematic and dynamic analysis of motion and loading”, Doctoral Dissertation, Delft Technical University (1999) de Groot JH, Rozendaal LA, Meskers CA and Arwert HJ, “Isometric shoulder muscle activation patterns for 3-D planar forces: A methodology for musculo-skeletal model validation”, Clin Biomech 19:790-800 (2004) Dul J, “A biomechanical model to quantify shoulder load at the workplace”, Clin Biomech 3:124-128 (1988) Gagnon D, Lariviere C, and Loisel P, “Comparative ability of EMG, optimization, and hybrid modeling approaches to predict trunk muscle forces and lumbar spine loading during dynamic sagittal plane lifting”, Clin Biomech 16:359-372 (2001) Garner BA and Pandy MG, “Musculoskeletal model of the upper limb based on the visible human male dataset”, Computer Methods in Biomechanics and Biomedical Engineering 4:93-126 (2001) Herberts P, Kadefors R, Hogfors C, and Sigholm G, “Shoulder pain and heavy manual labour”, Clin Orthopaedics 191:161-178 (1984) Hogfors C, Sigholm G, Herberts P, “Biomechanical model of the human shoulder – I. Elements”, J Biomech 20:157-166 (1987)
135
Hogfors C, Peterson B, Sigholm G, and Herberts P, “Biomechanical model of the human shoulder – II. The shoulder rhythm”, J Biomech 24:699-709 (1991) Hogfors C, Karlsson D, and Peterson B, "Structure and internal consistency of a shoulder model", J Biomech, 28:767-777, (1995) Hughes RE, “Empirical evaluation of optimization-based lumbar muscle force prediction models”, Doctoral Dissertation, University of Michigan (1991) Hughes RE, Chaffin DB, Lavender SA, and Andersson GBJ, “Evaluating muscle force prediction models of the lumbar trunk using surface electromyography”, J Orth Res 12:698-698 (1994) Hughes RE, Bean JC, and Chaffin DB, “Evaluating the effect of co-contraction in optimization models”, J Biomech 7:875-878 (1995) Hughes, RE and An K-N, “Monte Carlo simulation of a planar shoulder model,” Medical & Biological Engineering & Computing 9:544-548 (1997) Jarvholm U, Palmerud G, Herberts P, Hogfors C and Kadefors R, “Intramuscular pressure and electromyography in the supraspinatus muscle at shoulder abduction”, Clin Orthopedics 245:102-109 (1989) Karlsson D and Peterson B, "Towards a model for force predictions in the human shoulder", J Biomech 25:189-199 (1992) Laursen B, Jensen BR, Nemeth G, and Sjogaard G, “A model predicting individual shoulder muscle forces based on relationship between electromyographic and 3D external forces in static position, J Biomech 31:731-739 (1998) Laursen B and Jensen BR, “Shoulder muscle activity in young and older people during a computer mouse task,” Clin Biomech 15:S20-S33 (2000) Lippitt S, Matsen F. “Mechanisms of Glenohumeral joint instability,” Clin Orthopedics 291:20-28 (1993) Makhsous M, “Improvements, validation and adaptation of a shoulder model”, Doctoral Dissertation, Chalmers University of Technology, Gothenburg, Sweden (1999) Meskers CGM, “Quantitative assessment of shoulder function in a clinical setting: methodological aspects and applications”, Thesis, Leiden University Medical Center (1998) Niemi J, Nieminen H, Takala EP, Viikari-Juntura E, “A static shoulder model based on a time-dependent criterion for load sharing between synergistic muscles”, J Biomech 29:451-460 (1995)
136
Nieminen H, Niemi J, Takala EP, Viikari-Juntura E, “Load-sharing patterns in the shoulder during isometric flexion tasks”, J Biomech 28:555-566 (1995) Nussbaum MA, Chaffin DB, and Rechtien CJ, “Muscle lines-of action affect predicted forces in optimization-based spine muscle modeling”, J Biomech 28:401-409 (1996) Nussbaum MA and Zhang X, “Heuristics for locating upper extremity joint centres from a reduced set of surface markers”, Human Movement Science 19:797-816 (2000) Palmerud G, Kadefors R, Sporrong H, Jarvholm U, Herberts P, Hogfors C, Peterson B, “Voluntary redistribution of muscle activity in human shoulder muscles”, Ergonomics 38:806-815 (1995) Pedersen DR, Brand RA, Cheng C, Arora JS, “Direct comparison of muscle force predictions using linear and nonlinear programming”, J Biomech Eng 109:192-199 (1987) Van der Helm, FCT and Veenbaas R, "Modeling the mechanical effect of muscles with large attachment sites - application to the shoulder mechanism", J Biomech 24:11511163 (1991) Van der Helm FCT, "Analysis of the kinematic and dynamic behavior of the shoulder mechanism", J Biomech 27:527-550 (1994) Van der Helm FCT, "A finite-element musculoskeletal model of the shoulder mechanism", J Biomech 27:551-569 (1994)
137
CHAPTER V QUANTIFICATION OF SHOULDER LOADING AND ITS RELATIONSHIP TO THE PERCEPTION OF EFFORT – II. SHOULDER MUSCLE FORCE PREDICTIONS
5.1
Abstract
The mechanism of effort perception during light industrial load-delivery tasks was examined in this experiment. Three hypotheses regarding the etiology of perceived effort in the shoulder were evaluated. Effort perception in the shoulder was shown to be quantitatively related to total physical loading in a studied population, both for overall and specific shoulder loading at the resultant torque (r2 = 0.50) and cumulative muscle tissue (muscle force prediction: r2 = 0.42, electromyography: r2 = 0.26) levels. These relationships were more pronounced in individuals than when considering a population of subjects. Muscle activity information did not explain the variation in perception of effort better than general shoulder joint torque loading. The inclusion of subject and task descriptive variables, along with specific muscle tissue loading information, in models of the loading/perception relationship consistently improved the ability of each model to explain variability across a population (Torque: r2 = 0.74; muscle force prediction: r2 = 0.67, electromyography: r2 = 0.64). These results suggest that effort perception may not be fully explained by the motor command as quantified by muscle tissue contributions, but rather is a complex integrative quantity that is affected by factors that include posture, subject characteristics, and specific task goals.
138
5.2
Introduction
In the design of workstations, multiple competing objectives must be recognized. Primary amongst these objectives is the safety of the person performing tasks within the work environment. Beyond this first goal, which should be unequivocally attained, are secondary goals that include the comfort of the operator. Thus, a designer must also be sensitive to the impact of performing physical tasks on the worker’s perception of their duties. Perception of effort has been studied for a variety of physical exertions, often with respect to exercise and rehabilitation programs (Borg, 1974; Cafarelli, 1982). Beyond this, proportional effort perception rating scales have been developed for application in the workplace (Borg, 1982). Additional applied industrial studies have investigated the effects of the spatial arrangement of work on the level of perceived effort associated with the task (Ulin et al., 1990; Kim et al., 2004). These studies represent important attempts to relate objective, measurable work descriptors to the level of effort perceived while performing a specific type of task. The reliable estimation and reporting of effort perception may also be useful in evaluating potential links between physical exposures and risk of musculoskeletal outcomes. Odds and risk ratios for these outcomes in the shoulder have been shown to be more closely related to perceived effort than work dissatisfaction, for instance (Yeung et al., 2003). The etiology of the perception of effort is a topic that is still under debate in the scientific community, though there is some agreement upon aspects of the perception pathway. The perception is thought to be related to the afferent pathway of neural muscle activation (Cafarelli 1982; Gandevia et al, 1990; Burgess et al., 1995). This was supported by previous work relating the level of resultant shoulder torques and perception
139
of effort in the shoulder (Chapter II). In addition to the afferent pathway, however, the possibility of an efferent feedback component was suggested, as has been mentioned in the literature (Kilbreath et al., 1997). In this study, a different metric is used to describe the loading present in the shoulder while performing a task. A 3-dimensional mathematical model of the musculoskeletal components of the human shoulder was used to estimate the loading occurring in specific shoulder tissues while performing a series of loaded reach tasks. The muscle activation was also recorded in these tasks through the use of surface electromyography (EMG). The research reported in this chapter seeks to answer three principal hypotheses:
1)
The perception of loading in the shoulder is quantitatively related to the specific muscle tissue loading achieved while performing a task.
2)
The perception of loading in the shoulder is quantitatively described more accurately with a spectrum of muscle tissue loading predictions than with a subset of muscles monitored experimentally with EMG.
3)
Specific tissue loading can provide a more accurate description of the perception of effort pathway than possible with general joint loading metrics, specifically resultant dynamic shoulder torques.
These hypotheses were addressed through an experimental protocol similar to that explained in the previous chapters. Comparisons of the resulting muscle force prediction/effort perception model with the results of the shoulder torque/effort
140
perception model presented in Chapter II, along with comparisons to an EMG/effort perception model are a central part of this analysis.
5.3
Methods
5.3.1
Subjects
Eight college-aged subjects participated in this study. Summary data for these subjects is in Table 5.1. All subjects were free from documented history of chronic musculoskeletal health problems. All subjects signed informed consent forms and performed all experimental trial in one experimental session.
Table 5.1. Summary subject characteristics. Gender Male
Female
5.3.2
Mean S.D. Mean S.D.
N 4
4
Age (Years) 24.3 3.3 24.3 5.5
Stature (cm) 182.0 9.2 163.5 6.4
Weight (kg) 78.2 11.7 57.9 0.2
Strength (Nm) 52.32 11.0 30.0 0.9
Experimental Task Studied
The experimental tasks performed in this investigation were loaded and unloaded one-handed reaches to specified targets in space. These tasks are common in manufacturing environments, in which parts for assembly and hand tools are often moved around an environment to interface with the object being worked on. Each reach began at a common starting location, which was an instrumented panel on a simulated workbench. Upon being given both aural and visual cues, the subject moved from the starting location to a target in space. Upon arriving at the target, the subject depressed a small pushbutton, beginning a static hold phase. During this hold phase, the body posture attained at the
141
specified target was held for a minimum of three seconds. After three seconds had elapsed from the pushing of the button, the subject was given additional aural and visual indicators to return to the beginning location. Following the reception of these indicators, the subject returned to the starting position, at which point they were able to rest until the beginning of the next trial. A minimum of thirty seconds of rest separated experimental trials. In addition, the trials were subdivided into blocks between which an additional 3-5 minutes of rest occurred. Trials were ordered over the entire experimental session randomly with respect to task parameters, which were systematically varied to produce a range of posture/loading combinations.
5.3.3
Variation of Experimental Task Parameters
The target locations used in this experiment were analogous to those explained in Chapters II and IV of this dissertation. The objective of distributing the target locations in the right-handed reach envelope of the subjects was to assess the consistency of the exposure/perception relationship dimensionally. In addition to testing a range of target locations, different hand load levels were also used in order to generate a range of loading conditions. Table 5.2 summarizes the variation of these task parameters, three regarding the spatial variation of the target, and one concerning the force requirements of the task. It should be noted that trials along the 0° elevation projection angle were omitted for the 30° and 0° azimuths, as these reaches were occluded by the simulated workbench used in the experiment. This resulted in (5X3X2X3 = 90 – 2X3X2X2 = 78) 78 unique exertions tested per subject.
142
Table 5.2.
Experimental Variation of Reach Task Parameters
Dimension
Description
Values
Radial Azimuth
Deviation of target location along five azimuths, measured from the saggital plane, in degrees
-30° ,0°, 45°, 90°, 120°
5
Elevation Projection
Projection of target along line beginning at center of seat pan, expressed as degrees above horizontal
0°,35°,70°
3
Reach Distance
Expressed as a percentage of measured 60%, 85% maximum reach distance along specified azimuth and projection
2
Hand load
Expressed as a percentage of demonstrated maximum shoulder abduction/flexion strength
3
5.3.4
Levels
0%, 25%, 50%
Anthropometry
A series of anthropometric measurements were taken on all subjects. The critical measurements taken for use in the described mathematical shoulder model were subject stature and bodyweight.
5.3.5
Strength Testing
Prior to performing the experimental trials, the subjects were evaluated for their maximum shoulder strength in three directions: 1) forward flexion; 2) abduction; and 3) combination flexion/abduction [at 45° clockwise from the saggital plane]. Each test included of a ramp up and recording phase. Two trials were performed in each position.
143
If the agreement between the tests was not within +/- 5%, a third trial was performed. The demonstrated strengths were averaged across the three exertions to attain a single value describing subject strength. This value was then used to set the three hand load levels indicated in Table 5.2.
5.3.6
Effort Perception Scale and Calibration
The scale used to measure effort perception was a modified Borg CR-10 scale (Borg, 1982). Each division on the scale represented a 10% increase in perceived muscular effort relative to an established maximum value. For example, a rating of 5 would indicate a perception of muscular effort equal to 50% of the maximal exertion. A representation of the scale is shown in Figure 5.1.
Figure 5.1
Rating Exertion Scale. Each number on the scale corresponds to 10% of the perception of effort during calibration shoulder exertions. A moveable cursor was used by the subject to indicate the level of perceived effort during the task, which was connected to an electronic sensor which reported the effort to the hundredths place. 144
Prior to performing the experiments, the subjects were also calibrated to the effort scale. Initially, feedback was given with a visual digital scale which displayed the level of force produced with a given strength task. Subjects were asked to reproduce the levels by producing variable amounts of force. Following initial calibration, the feedback scale was obstructed and subjects attempted to produce specified force levels that corresponded to percentages of their maximal strength (%MVS). This continued until the subjects were able to reproduce a %MVS to +/- 5%.
5.3.7
Experimental EMG Data Collection
During the experiment, EMG data was collected on 11 shoulder muscles on the right side of the body. The locations for EMG electrodes are explained in Chapter IV, and are listed in Table 4.2. EMG average values during the static hold phase of the reach exertion were used in the analysis as representative of the recorded EMG levels during the trials. The definition of the time window defined for the static hold phase is discussed in detail in Section 4.3.
5.3.8
The Mathematical Shoulder Model
The construction of the mathematical shoulder model used to quantify muscle tissue loading for this study was described in Chapter III and evaluated in Chapter IV. The model uses motion, task, and subject data streams to generate predictions of force levels throughout a task. For this analysis, the muscle force prediction levels studied were those levels associated with the static hold phase of the reach tasks. The values
145
calculated for this portion of the reaching task were used in building a statistical model to relate shoulder muscle tissue loading to perceived effort.
5.3.9
Regression Model Construction
To evaluate the influence of three classes of factors on the perception of effort several statistical multiple regression models were created. These models were based on three characterizations of shoulder loading, each expressed at two descriptive levels, one general and one specific. Table 5.3 summarizes the model types developed. These models were developed using a commercial statistics software package, JMP®. All models were created using a mixed stepwise selection process with inclusion/exclusion criteria set at 0.25/0.25. To keep the model as parsimonious as possible, only main effects were considered in building the initial models. Previous work with resultant torque data showed little improvement achieved with the use of interaction terms (Chapter II). Table 5.4 contains the task factors that were also considered in building the multifactor models. In this study, subject was treated as a random variable. The radial directionality of the task used in this experiment quickly emerged as a factor both in the muscle use as well as interpretation of relatively simple metrics such as torque, which was represented in a global system. As a result, individual models were run for the entire subject pool and across all model formulations as seen in Table 5.3 for the five azimuths tested, as well as over the entire dataset from all azimuths, which is defined as the composite index.
146
Table 5.3
Regression Model Component Descriptions
Loading Metric
Dynamic Shoulder Torques (DSTs)
Temporal Muscle Force Predictions (MFPs)
Electromyographic Levels Recorded (EMG)
Specificities
Factor Type Inclusion Levels
Resultant DSTs
Loading only
Loading and Loading, subject, subject and task
Directional DSTs
Loading only
Loading and Loading, subject, subject and task
Sum of MFPs for all model muscles
Loading only
Loading and Loading, subject, subject and task
Individual MFP levels
Loading only
Loading and Loading, subject, subject and task
Sum of EMG for all recorded muscles
Loading only
Loading and Loading, subject, subject and task
Individual muscle EMG levels
Loading only
Loading and Loading, subject, subject and task
Table 5.4 Specific Task Effects Considered in the Multiple Regression Models Task Variables Target Azimuth Angle of target azimuth measured from the sagittal plane Target Elevation Angle Projection of target elevation from the hip point Target Distance Target location along azimuth and elevation (expressed as percentage of maximum reach)
147
5.3.10 Regression Model Comparison Technique
These major model types were based primarily upon loading metrics, which were complemented by the other factors considered (subject and task parameters). Accordingly, the method used was to compare the ability of these different models based upon the proportion of the variance in the perceived effort responses given by the subjects. This measure of model performance is commonly known as r2. Models were deemed more effective in representing the underlying predictors of effort perception if they explained higher levels of the variance, and consequently had high r2s.
5.4
Results
5.4.1 Main Effects/Dependence on Torque & Directionality of Task
All loading metrics (torques, muscle force predictions, and EMG) were positively correlated to the perception of effort. As the level of loading increased, the effort increased proportionately. Variations in these relationships did exist, however, between subjects and between the different directions of reach motions. As in Chapter II, within subject effort was highly correlated to loading metrics. The purpose of this investigation, however, was to quantify the overall performance of different loading descriptors as well as other characteristics included in multiple regression models to predict perceived effort levels. For this reason, the data was examined across the subject pool. It should be noted as reported in Chapter II that the intrasubject variability in reporting effort was 0.85 ± 0.69 on the 10-point effort rating scale used, which represents approximately 8.5% of the effort perception corresponding to a maximal effort. This inherent variability should be accounted for when interpreting model results.
148
Four data permutations were first considered: 1) resultant shoulder torques, 2) directional shoulder torques, 3) total muscle force prediction, and 4) total EMG recorded. Figure 5.2 contains the variance explanations (r2 values) for each of these permutations. The panels in the figures correspond to the top view perspective. Each ray extending from the center of each subfigure represents an azimuth for which experimental trials were performed. The zero degree azimuth corresponds to the sagittal plane of the body. Along each azimuth, the model r2 values are shown in a set of three stacked boxes corresponding, top to bottom, to the model formulations for each loading metric studied with three progressively inclusive factor inclusions. These correspond to the following:
Top:
Shoulder Loading Metric Alone
Middle:
Shoulder Loading Metric Combined with Subject Data
Bottom:
Shoulder Loading Metric Combined with Subject and Task Data
In addition, in the bottom left corner of each subfigure is a description of the model’s r2 composite performance over the combined azimuths, as described in the methods section. These various model formulations allow the assessment of the relationship of different loading characterizations and their interaction with subject and task-specific data sources. Following the construction of these initial models, more specific models were developed to examine the EMG and MFP data. These models were constructed directionally and inclusive of both subject as a random effect and spatial task parameters. A composite index for the directional models is presented in Table 5.5.
149
Muscles that were present in the equations for multiple azimuths in the MFP data were the latissimus dorsi, the anterior deltoid, the infraspinatus, the trapezius, the subscapularis, the biceps, the brachialis and brachioradialis. The overall model for the MFP data included the latissimus dorsi, the clavicular branch of the trapezius, the middle deltoid, the infraspinatus, the subscapularis, the supraspinatus, the biceps, the triceps, the brachioradialis, target azimuth, target elevation, and subject as a random effect. In the EMG based model, the lower trapezius, middle and posterior deltoid, infraspinatus and biceps were in multiple azimuth formulations. For the overall EMG data, the factors were the latissimus dorsi, the lower trapezius, the posterior deltoid, the infraspinatus, the biceps, the triceps, target azimuth, target elevation, and subject as a random effect.
150
A
B
Resultant Torque Models -30° 0.39 0.54 0.73
Directional Torque Models
0° 0.41 0.56 0.75
-30° 45°
0.45
0.50
0.62
0.69
0.72
0.69
0° 0.39 0.54 0.75
45° 0.49 0.69 0.69
90° 0.47 0.64
Resultant Torque Composite
0.64 120°
0.50
0.69
0.69
0.55
90°
0.47
0.32
0.55
0.50 EMG Composite 0.26 0.47
0.49
0.60
0.69
Total Electromyographic Data Models 0° -30° 0.28 0.30 45° 0.59 0.62 0.24 0.67 0.65 0.48
0.38
0.55
0.50
0.73
90°
120°
0.42
120°
0.60
D
0.60
MFP Composite
0.65
0.45
0.66
Total Muscle Force Prediction Models 0° -30° 0.44 0.29 45° 0.60 0.51 0.49 0.70 0.65 0.68
0.65
Directional Torque Composite
0.65
0.74 C
0.49
0.48
0.61
90°
0.53
0.53
0.56 120° 0.28 0.40 0.40
Loading Alone Legend
Loading and Subject Factors Loading, Subject and Task Factors
Figure 5.2
Relative Performance of Effort Prediction Models. For each azimuth, a box indicates the r2 value for different model types. A) Resultant Torque Models, B) Directional Torque Models, C) Total Muscle Force Predictions, D) Total EMG Predictions.
151
Table 5.5
Specific EMG and Muscle Force Prediction Model Performance
MFP Based Effort Prediction Model Azimuth r2 value -30° 0.72 0° 0.75 45° 0.77 90° 0.68 120° 0.73 Overall 0.67
5.4.2
EMG Based Effort Prediction Model Azimuth r2 value -30° 0.68 0° 0.76 45° 0.72 90° 0.64 120° 0.59 Overall 0.64
Performance of the Perception Models
All loading metrics showed significant positive correlations with reported effort scores. When considering the coarsest measures of total torque, total MFP, or total EMG, it was found that a resultant torque model outperformed the other models. (Torque model: r2 = 0.50; Muscle force prediction model: r2 = 0.42; EMG prediction model: r2 = 0.26). This discrepancy, while smaller when considering subject and spatial task characteristics, persisted in showing the resultant torque model as yielding the highest fit for the effort responses (Torque model: r2 = 0.74; Muscle force prediction model: r2 = 0.60; EMG prediction model: r2 = 0.53). The muscle force models achieved a slightly higher fit than the EMG model for both conditions. Increasing the power of exponent on the individual muscle forces in the total MFP and EMG models did not increase the prediction accuracy of the model, though both quadratic and cubic summations were considered. In the specific MFP and EMG effort prediction models, the MFP-based model slightly outperformed the EMG model again (r2 = 0.67 vs. 0.64). Both of these fits, however, were still inferior to either the resultant or directional torque overall performance (r2 = 0.74 and 0.72, respectively), but only slightly.
152
5.5
Discussion
The most salient finding derived from the models was a positive correlation between several different derived shoulder loading metrics and the perception of muscular effort in the shoulder, thus confirming the first hypothesis. This is in agreement with studies that have found positive correlations between effort perception and the magnitude of loads manipulated in two-handed weight transfers in a right-to-left horizontal lifting task (Ljungberg et al., 1982). The relationship of perceived effort has often been described as following a power function (Stevens, 1957) of the form:
ψ = KΦ n
(5.1)
where ψ is the sensation magnitude, Φ is the physical stimulus intensity, and K and n are parameters relating to the particular sensation. An accelerating power function has
also been used to model effort perception for constrained knee exertions more recently (Pincivero et al., 2002, 2003). In our experiment, a power relationship did not explain the variance in the effort response to any greater fidelity than a simple linear construction, and in fact for torque loading was somewhat less able to explain the variability in effort perception (linear model r2 = 0.50; power model r2 =0.44). This is likely due to the processing of data collected. In our study, the experimental trials, while conceptually similar, were performed for a variety of different end goal targets. The variation associated with these target locations may have resulted in different movement patterns for different targets. This, combined with variation in the stature and strengths of the participating subjects, complicates interpretation. To construct the power relationships discussed, the authors averaged experimental trials for
153
consistent physical exertion levels. This was not possible given the wide range of the predictor variables in our study, even for a similar set of tasks. Despite these small methodological differences, the general trends are analogous: perceived effort increases in proportion to increases in loading.
5.5.1
Performance Criteria for Different Metrics of Shoulder Loading
Although all of the loading metrics studied in this experiment were significantly related to effort perception, there was a range of effectiveness in predicting perceived effort amongst them. This is apparent when comparing the dynamic shoulder joint torques to either the muscle force predictions or the EMG model performance. Relatively speaking, the resultant dynamic torque is a coarse indicator of the overall loading experienced at the shoulder when compared to load estimates in individual shoulder tissues. Accordingly, it was initially hypothesized that the enhanced fidelity of the muscle force model in describing shoulder loading would enable a more accurate prediction of perceived effort scores (Hypothesis 3). This was not the case, as the torque model, in a general sense, outperformed both of the muscle activity based model forms, essentially explaining approximately 18-24% more of the variance than the other two models, if only loading metrics are considered. This difference is marginally reduced if complementary factors related to subject and spatial task characteristics are used (1421%). However, if the loading in specific tissues is considered, this difference becomes only 7-10%, but the torque model still performs at the highest level.
154
The reasons behind this variable model performance may be related to two factors: 1) the recorded EMG signal was acquired for a subset of shoulder muscles; and 2) the mathematical muscle force prediction model may be of insufficient accuracy to depict shoulder loading fully. Addressing the first point, EMG was collected for 11 shoulder muscles, which represent less than 1/3 of the total number of muscle elements modeled mathematically. Thus, using these values as representative of the efferent copy of the motor command may be suspect. In the shoulder, several critical muscles are difficult to access using surface electrodes, but may play a role in effort perception. This is especially true for the components of the rotator cuff, which are active as agonists and antagonists during many manual tasks (Meskers, 1998). The lack of a full complement of shoulder muscle electromyographic data may have compromised the ability of this metric to fully account for the impact of the efferent motor command on the perception of effort. The second point also bears further discussion. While based on physics and requiring several equilibrium equations be satisfied, the muscle force prediction model displayed some inability to replicate EMG data for all recorded muscles (Chapter IV). This was suggested to be attributable to several factors, one of which is the general inability of optimization models that employ monotonically increasing cost functions to reliably predict antagonistic co-contraction (Hughes et al, 1991). This leads to potential underestimation of the overall level of force produced by the muscle tissues of the shoulder. However, with the increased number of predictive forces contributing, the model can provide some sensitivity unavailable with a sampling of EMG data. These two aspects combine to achieve an effort perception prediction model based on muscle
155
force predictions that performs slightly better as a muscle EMG model when applied to the entire data set with task and subject variable influences considered (EMG model r2 = 0.60; MFP model r2 =0.53). Hypothesis 2 is therefore difficult to evaluate conclusively. There is less evidence for the superiority of the MFP model when considering specific muscles as predictors, as the difference between these models was just 3%.
5.5.2
Influence of Subject and Task Parameters on Effort Perception
Beyond the use of loading metrics, it was shown that the model performance improved with the inclusion of subject and spatial task characteristics as detailed in tables 5.3 and 5.4. Inclusion of additional mediating variables improved the overall predictions. It should be noted that in each prediction model, the loading metrics were the most significant factors in explaining variance in the effort levels across the population. This is in agreement with earlier studies (Yeung et al., 2003). In their study, the weight was found to be the most important factor in determining perceived effort for lifting tasks. In addition, they also found significance contributions in a multiple regression model of task factors including directional distances to targets, which were also observed for a series of equally-weighted load transfers (Kim et al., 2004) and in our own study. Further, the study by Kim et al. also demonstrated that certain subject characteristics (age, gender, stature, and body weight) contributed to the perception of effort, which was also indicated by the results of this study across all loading metrics, albeit with subject defined as a random effect. This was done in the study because including subject as a random variable instead of representing it with stature and strength explained somewhat more of
156
the variance for the three main models than stature and strength alone (Chapter II), suggesting that other subject variation may have also impacted effort perception.
5.5.3
The Influence of Radial Azimuth on Model Prediction Performance
The radial direction of the task performed was shown to affect the performance of each of the developed models. This result is intuitive in that muscles in the shoulder have been shown to be differentially able to produce forces in different directions for a given posture (de Groot, 2004; Meskers, 1998). This difference is more pronounced in the muscle-based models. The EMG and MFP models show superior performance for tasks located along particular azimuths. For example, the models were shown (Figures 5.2 and 5.3) to explain a larger proportion of the variance in effort perception for those azimuths located in front of the subject (0°, 45°, and 90°). This may be due to the increased familiarity of the subjects with exertions in these directions, as the more extreme azimuths are encountered less often in typical occupational and daily life tasks. This may have been the mechanism that led to a slight increase in the consistency of the loading/effort relationship in these directions.
5.6
Conclusion
The underlying mechanisms of the perception of effort or exertion have been studied closely for several decades. A large majority of preceding work was directed at relating central factors to levels of perceived effort. In this study, an attempt was made to examine which local physical loading metrics for the shoulder were most effective in producing accurate predictions for perceived effort at the shoulder. Of the three metrics used to quantify shoulder loading exposure during the experiment, the most effective for
157
general application was the resultant dynamic shoulder torques, which was also the coarsest measure of shoulder loading used. This leads to the conclusion that effort perception is an integrative process that may include components not captured by the motor command alone. Figure 5.3 depicts a representation of the pathways that may generate shoulder loading and accompanying effort perception. With modulation by subject and spatial task characteristics also considered, the coarse measure of resultant joint torques consistently yielded the best fit with experimental effort perception data overall, though this difference was modulated when specific tissue loading was considered. Spatial aspects of the task, subject characteristics, and the loading impacted the perception of effort amongst the various models (Figure 5.4). This investigation serves in a functional way to justify the use of simple loading metrics to estimate potential discomfort. This is fortuitous in that models that calculate joint torques are widely available to practicing ergonomists. Implementation of a relatively simple expression to predict effort based on shoulder torques and a limited number of task and subject data could provide guidance in the design of workstations. In addition to its pragmatic contribution to prospective job analysis, this study also provides new insight into the mechanism of effort perception in the shoulder, and confirmation of the theory of effort perception as an integrative phenomenon. The nature of this integration, however, requires further attention to clearly describe its characteristics.
158
Figure 5.3
An Integrative Model of Effort Perception. General joint loading is determined by extrinsic factors. The conversion of these loads into internal forces is modulated by intrinsic factors. Finally, these loading metrics are interpreted under the influence of task and subject characteristics to achieve an effort perception level.
159
Figure 5.4
Progression of Variance Explanation Through Effort Models. In general, the simplest measure, the resultant elbow torques (EJT), explained the variance in perceived effort most completely. This is less pronounced when specific tissue activity is considered, however. In the tissue-based models, higher specificity led to more variance explanation. This was not seen in the torque-based models. Inclusion of subject and task variables increased model performance across all formulations.
160
5.7
References
Borg GA “Perceived exertion”, Exercise and Sport Sciences Reviews 2:131-153 (1974) Borg GA “Psychophysical basis of perceived exertion”, Medicine and Science in Sports and Exercise 14:371-381 (1982) Burgess PR, Cooper TA, Gottlieb GL, and Latash ML, “The sense of effort and two models of single-joint motor control, Somatosensory and Motor Research 12:343-358 (1995) Cafarelli E, “Peripheral contributions to the perception of effort”, Medicine and Science in Sports and Exercise 14:382-389 (1982) de Groot JH, Rozendaal LA, Meskers CA and Arwert HJ, “Isometric shoulder muscle activation patterns for 3-D planar forces: A methodology for musculo-skeletal model validation”, Clin Biomech 19:790-800 (2004) Gandevia SC, Macefield G, Burke D, and McKenzie DK, “Voluntary activation of human motor axons in the absence of muscle afferent feedback. The control of the deafferented hand”, Brain 113:1563-1581 (1990) Hughes RE, “Empirical evaluation of optimization-based lumbar muscle force prediction models”, Doctoral Dissertation, University of Michigan (1991) Kilbreath S.L., Refshauge K, and Gandevia S.C., “Differential control of the digits of the human hand: evidence from digital anaesthesia and weight matching,” Experimental Brain Research 117:507-511 (1997) Kim, K.H., Martin BJ and Chaffin DB, “Modeling of shoulder and torso perception of effort in manual transfer tasks,” Ergonomics 47(9):927-944 (2004) Ljungberg AS, Gamberale F, and Kilbom A, “Horizontal lifting – physiological and psychological responses”, Ergonomics 25:741-757 (1982) Meskers CGM, “Quantitative assessment of shoulder function in a clinical setting: methodological aspects and applications”, Thesis, Leiden University Medical Center (1998) Pincivero DM and Gear WS, “Neuromuscular activation and perceived exertion during a high intensity, steady-state contraction to failure”, Muscle Nerve 23:514-520 (2000). Pincivero DM, Coelho AJ, Campy RM, “Perceived exertion and maximal quadriceps femoris muscle strength during dynamic knee extension exercise in young adult males”, Eur J Appl Physiol 89:150-156 (2003) Stevens SS, “On the psychophysical law”, Psychology Review 64:153-181 (1957)
161
Ulin SS, Armstrong TJ, Snook SH, and Franzblau A, “Perceived exertion and discomfort associated with driving screws at various work locations and at different work frequencies,” Ergonomics 36: 833-846 (1993) Yeung SS, Genaidy A, Deddens J, Leung PC, “Workers’ assessments of manual lifting tasks: cognitive strategies and validation with respect to objective indices and musculoskeletal symptoms”, Int Arch Occup Environ Health 76:505-516 (2003)
162
CHAPTER VI CONCLUSIONS
6.1
Overview
This dissertation reports on the creation and testing of several mathematical models of the shoulder mechanism. These models were designed to both assess general and specific loading of shoulder components, as well as to quantitatively relate these loading metrics to the perception of effort in the shoulder. The principal novelty of the developed models is their targeted use in biomechanical ergonomic analyses, though several other new theoretical concepts are introduced. Additionally, several elements were incorporated into the formulation of the models to facilitate their enhancement as superior parameter information becomes available through future investigations. The testing of the models was accomplished through experiments that simulated a subset of work tasks. These experiments established the feasibility of using the model to predict a range of muscular activity. It should be noted, however, that the experiments were performed with a relatively small subject pool, and the results of the analyses should be interpreted with this in mind. Further studies, as discussed, with an expanded demographic base and additional variation of task requirements would allow the application and validation of the developed models across a broader range of work activities. Nonetheless, the experiments performed assisted in evaluating the developed models, and also led to areas of potential future research.
163
6.2
Principal Research Contributions
The dissertation addresses the five research objectives stated in Chapter I. This body of work makes several contributions to the fields of biomechanics and ergonomic analysis, specifically for the shoulder: 1) Creation of a novel biomechanical shoulder model for ergonomic analysis. Prior to this dissertation work, most ergonomic analyses of shoulder loading
have relied on static postural analysis, such as the 3DSSPP program (Chaffin et al, 1997). The created model goes beyond these initial evaluations and is sensitive to the impact of inertial terms on the loading experienced in the shoulder during dynamic activities, both at general and specific tissue levels. Though the exertions studied in this dissertation showed relatively small contributions due to dynamics, the theoretical model can be applied to more dynamic tasks, including many assembly, exercise and sport exertions. A key aspect of the model is its ability to be scaled to populations, which is critical in studying occupational tasks. While several models exist to solve the load distribution problem in the shoulder, these models are not specifically built for this type of application. The developed model can be used to assess differences in tissue loading that occur for different workstation layouts, manipulated loads, and task frequency. Further, the model is scalable with respect to anthropometry and strength characteristics. This flexibility makes the tool very useful in the evaluation of tasks in which the requirements may remain constant, but the personnel are represented by a diverse population. Secondly, the development of a virtual internal shoulder geometry emulator was a large portion of the model development. An established difficulty in studying the
164
mechanics of the shoulder is the relative inability to document scapular and clavicular motion through the measurement of external markers. This has been addressed by several groups, primarily through the description of a form of ‘shoulder rhythm’ that quantitatively relates the relative orientation of the humerus with respect to the torso and the scapular and clavicular orientations (de Groot, 1999; Hogfors, 1991). The geometry emulator developed as a part of this dissertation implements a rhythm previously defined (Makhsous, 1999). A major benefit is the creation of a 3-D graphic representation of the internal musculoskeletal geometry that includes muscle attachment sites and lines-ofaction as well as bone orientations. This allows detailed visual and numeric scrutiny of instantaneous shoulder geometry beyond what was previously available. The emulator can also provide visual feedback concerning the alterations made to muscle paths using various wrapping algorithms. This degree of online interactivity is difficult to achieve in many existing models. In essence, a new formulation of a series of linked biomechanical models of the shoulder specifically geared towards ergonomic analyses has been achieved. One major benefit of this model is its ability to address specific hypotheses about identified tasks and generate conclusions based on different metrics of shoulder loading. This is a step towards improving the fidelity of job evaluation for potential risks to the shoulder. In particular, the model allows the estimation of tissue-specific stresses and forces. This unique knowledge could enhance the specificity of identification of shoulder-related ergonomic risks as well as potentially link the loading associated with performing work tasks to musculoskeletal pathologies that are observed clinically, which would represent an important step in future ergonomic analyses.
165
2) Inclusion of an empirically-based glenohumeral stability constraint in the load-distribution solution. Previous shoulder model formulations have relied on
heuristic methods by which to delimit permissible shear force in the glenohumeral joint (van der Helm, 1994; Makhsous, 1999). While largely theoretically consistent with glenohumeral joint stability, a method based on experimental findings is preferred. In the developed biomechanical model, empirical data collected on glenohumeral dislocation characteristics of cadaveric specimens (Lippitt and Matsen, 1993) is used to define this constraint. This inclusion generates a novel, physiologically accurate, anisotropic glenhumeral stability requirement that is implemented as a series of linear equations, thus assuring a global solution to the optimization problem. In addition, the use of this constraint can be modified using a scaling factor for either conservative or liberal stability maintenance requirements. 3) A methodology for the quantitative evaluation of muscle force predictions for a set of industrial reaching tasks. Existing models have used several techniques for
their evaluation. Often, highly constrained tasks, such as loaded shoulder abduction and flexion (van der Helm, 1994) are used to evaluate model performance. Further studies have examined the directional accuracy of musculoskeletal models in a given posture (de Groot, 2004). While essential to the understanding of the internal function and performance of a model, these analyses are not directly transferable to exertions performed dynamically or in a range of postures. This study represents a significant accomplishment in that it attempts, despite numerous documented complications, to perform an analysis on industrially relevant, complex tasks. Effectively, an analysis method designed to assess models of the shoulder musculature in such situations is
166
described and applied. This method is extendable and can be modified to study model performance in other, unstudied postural loading conditions. 4) Establishment of the quantitative relationship between physical shoulder loading and the psychophysical perception of that loading for load transfer tasks.
One goal of the dissertation was to relate the subjective psychophysical perception of effort to objective measures of shoulder loading. This allows the prospective evaluation of changes in a workplace interface on the worker’s ultimate perception of effort for the job. In designing workplaces and human interfaces, it is important to design not only to attempt to limit potential injury but also to minimize the perceived physical experienced. This implementation allows designing a task or environment to meet or exceed a designated level of effort. Conversely, this relationship can be used to approximate the physical stresses associated with a given task based upon psychophysical metrics, which are far more convenient to obtain in the field, and are also less intensive to process. 5) An integrated, accessible modular shoulder analysis package. A final
unique contribution of the dissertation work is the integration of several models into one analysis package developed in a common software environment. This allows the cohesive assessment of a given task in multiple areas: external joint loading, posture, muscle tissue loading, and effort perception. This can be done without the sequential implementation of several separate programs by the program user, permitting analysis of a work task from visualization of external body landmarks through to predicting perceived effort and interpreting forces in individual muscles. The modular structure of the program also allows new parameter estimates to be incorporated as they become available through the course of planned future research. This will allow the package to
167
adapt with forthcoming increases in the global shoulder biomechanics knowledge base. In essence, the developed shoulder analysis package addresses the broad central motivation for this entire dissertation: to enable better analysis of the impact on the shoulder associated with performing various tasks. An interface that allows simultaneous visualization of posture and muscle force estimates is shown in Figure 6.1.
Figure 6.1
6.3
Visual Interface Allowing Comparison of Posture and Muscle Force
Future Research Directions
A research project often raises as many interesting questions as it can answer. Through the course of completing the dissertation work, many critical areas were identified for which the acquisition and application of additional knowledge could help to improve the understanding of shoulder biomechanics. Several directions of future research, which will ultimately lead to superior methods for estimating and evaluating mechanical loads in the shoulder, are summarized below, in three key areas: 1) model enhancement; 2) model evaluation; and 3) model implementation.
168
6.3.1
Model Enhancement 1) Systematic inclusion of soft tissue contributions to the maintenance of
glenohumeral stability in the mathematical models. The flexibility and range of
motion of the shoulder joint is purchased through a decreased level of joint stability. The maintenance of stability is thought to be achieved through a combination of both active muscle activation and soft tissue stabilization. Although the motions studied in this dissertation were estimated to be unlikely to require or invoke tightening of the shoulder ligaments, due to non-extreme postures, ligaments and the glenohumeral joint capsule play a role in ensuring stability for a range of exertions. For the model to be more universally applicable, estimation of the posture-dependent contributions of these tissues will be required. Investigation of the significance of the included stability parameter showed that it had a marked effect on the muscle force predictions (Chapter IV). The remaining uncertainty revolves around the estimation of an acceptable stability threshold and the relative contributions of the soft tissues and active shoulder muscle activation to the maintenance of stability. 2) Refinement of model parameter values for subject-specific strength capabilities. Currently, the model operates under the assumption that strength is
normally distributed across the subject population analogously to the world population and the musculature and is dependent on muscular cross-sectional area (which is currently scaled by demonstrated shoulder strength). This assumption is somewhat approximate. There is well documented variability in human strength, both with respect to gender (Stobbe, 1982; Kumar, 1996), age (Roebuck et al., 1975), and lean muscle mass (Bishop et al, 1987). Inclusion of scaling factors for determining each subject’s
169
maximum force producible based on these relationships may improve model predictions. The use of anthropometry alone to determine these values, however, would not be advised as the relationships are not sufficiently correlated for such applications (Chaffin, 1999). 3) Inclusion of muscle and joint intrinsic characteristics. Currently the model
does not contain several documented phenomena related to muscle physiology and joint mechanics. Inclusion of the length-tension and velocity-tension relationships may help to interpret model predictions and their dependency on these factors. A second application of intrinsic relationships would be the inclusion of a joint stiffness maintenance factor in the glenohumeral joint, and possibly the elbow. Inclusion of a stiffness parameter was shown to effect the results of a previous optimization model of the shoulder (Nieminen, 1995), however, comparison of the altered results with empirical data was not reported. In other body joints, such as the lumbar spine, stiffness parameters have been used to predict muscle activity levels (Cholewicki et al., 1995; Crisco and Panjabi, 1991), and may prove useful in enhancing this shoulder model as well. 4) Improvement of scapular orientation prediction for extreme postures.
The method currently used to predict scapular orientation is an extrapolated mathematical “shoulder rhythm” previously presented in the literature (Hogfors et al., 1991). This method has been shown to be less reliable in predicting scapular positioning for elevated arm postures. An extension of the original rhythm was presented later to improve this inaccuracy (Makhsous et al., 1999), but the method still retains some artifacts of its original construction. More recently, a series of 3-D shoulder rhythm regression equations have been presented (De Groot et al., 2001) that represent an alternative
170
rhythm formulation. The modular format of the shoulder model allows consideration of either this or future experimentally-derived scapular position prediction methods. Alternatively, directly measured scapular position could be used, provided a quasistatic analysis is performed. In light of the low impact of dynamic terms on joint torque loadings for the exertions studied, this may be a fruitful approach. The shoulder rhythm used, however, performed adequately for the subset of motions performed as a part of this dissertation.
6.3.2
Model Evaluation 5) Further empirical evaluation of the shoulder muscle force prediction
model. As stated in Chapter IV, the model has been initially evaluated using a set of
simulated industrial motions. These motions were exclusively one-handed load transfer tasks performed in a seated position. While the locations to which the loads were transferred were varied spatially, the force profiles obtained throughout the motions and at the extended position were largely gravitational (dependent on the weight of the arm and the weight in the hand). This led to the selective activation of specific muscles over others in the shoulder, particularly those responsible for arm elevation (deltoid, infraspinatus), and little to no activation for other muscles (latissimus, pectoralis major). An expanded set of exertion and force production tasks would enable holistic model evaluation. The model’s structure allows future consideration of other types of exertions, such as generating push and pulls forces. However, these tasks were not specifically studied in this work. Applying the model to novel situations could provide additional fundamental insights into the interplay between body posture, force generation, and
171
muscle activation in the shoulder, as well as the model’s overall performance. Reported experiments have investigated directional force production and its effect on shoulder muscle activation (Meskers, 1999; Laursen, 1999; de Groot 2004). However, these experiments required constrained postures as well as used isometric exertions. While able to simplify data analysis, the tasks did not replicate a range of common industrial tasks. In pursuing the further evaluation of the mathematical models for various tasks, a careful balance must be achieved that allows recording of relevant exertion types while permitting a high level of data interpretation. 6) Comparison of model outputs with existing shoulder models. Although
several mathematical models of the shoulder mechanism have been reported in the literature by groups world-wide (van der Helm, 1994; Hogfors et al., 1987, 1991, 1995; Charlton and Johnson, 2001; Hughes et al., 1999; Garner and Pandy, 2001; Laursen et al., 1998), little has been done to compare the outputs of these models for non-planar tasks. The complex geometry of the shoulder demands such a 3-D approach to be successful. Applying a rigorous set of loading conditions to each model, as well as replicating these loading conditions experimentally, could provide insight into the strengths and limitations of each model formulation, as well as guide future research towards the most fruitful unresolved areas. This exercise is somewhat limited through accessibility and inevitable format differences, however, could prove useful. A new proposal by the International Shoulder Group has suggested a convention by which to standardize reporting of shoulder kinematics (Wu et al., In Press), which should facilitate any comparisons. Even a limited comparison between available models would assist in advancing the field of shoulder biomechanics.
172
6.3.3
Model Implementation 7) Development of an angle-based shoulder analysis package for enhanced
simulation capability. One limitation of the current shoulder model is that it requires
specific spatial body landmarks as inputs to define the internal geometry of the model, as well as to calculate the external torques on the system. Early development has begun on an angle-based shoulder analysis package that requires only the relative orientations of the body segments, along with anthropometric data. This would allow the user of the program to simulate given postures without first generating a matrix of 3-D positional data, which can be a labor-intensive task. This will be a useful model feature until robust human motion simulation algorithms are integrated into Digital Human Modeling software, at which point this data would be freely accessible. 8) Creation of a reduced shoulder force prediction algorithm for use in Digital Human Modeling (DHM) software. Although the model generates a large
amount of ergonomically relevant data regarding specific and overall shoulder loading while performing a task, the infrastructure and time requirement associated with the optimization-based muscle force distribution problem would make it impractical for rapid evaluation of prospective work interfaces in DHM software. Accordingly, an accelerated statistical method to estimate the output of the combined shoulder analysis package is suggested. This statistical method would use an algebraic formulation of the shoulder model to determine muscle activity levels. Creation of an algorithm that performs this estimation based on a combination of shoulder torque levels and upper arm postural data is the next step in this process.
173
6.4
Summary
A biomechanical model of the shoulder has been presented and evaluated for a set of industrially relevant tasks. The model outputs several exposure variables of interest when evaluating a work scenario, including general dynamic shoulder torque loading, specific shoulder muscle tissue loading estimates, and an index of perceived muscular effort in the shoulder. Through a quantitative evaluation of the model, it was demonstrated that the model generates predictions that are consistent with the characteristics of an optimization-based musculoskeletal model. Development of the model in a flexible modular fashion will allow subsequent modifications and improvements with comparative ease. In addition, the model generates many useful intermediate outputs, including bony and muscular orientations and positions. The experiments used to evaluate the developed models demonstrated their potential for application to a wider range of work tasks and populations. Further study of model performance in this larger scope will enhance understanding of the behavior of the shoulder mechanism, as well as point towards those aspects of man-machine interfaces that most significantly impact the development of musculoskeletal disorders, the quality of work performance, and the nature of worker effort perception.
174
6.5
References
Bishop PK, Cureton K, and Collins M, “Sex Differences in muscular strength in equallytrained men and women”, Ergonomics 30:675-687 (1987) Chaffin, D.B. Development of Computerized Human Static Strength Simulation Model for Job Design, Human Factors and Ergonomics in Manufacturing, 7:305-322 (1997) Chaffin DB, Andersson GBJ, and Martin BJ, Occupational Biomechanics, 3rd Edition, John Wiley, New York (1999) Charlton IW and Johnson GR, “Application of spherical and cylindrical wrapping algorithms in a musculoskeletal model of the upper limb”, J Biomech 34:1209-1216 (2001) Cholewicki J, McGill SM, and Norman RW, “Comparison of muscle forces and joint load from an optimization and EMG assisted lumbar spine model: towards development of a hybrid approach”, J Biomech 28:321-331 (1995). Crisco JJ and Panjabi MM, “The intersegmental and multisegmental muscles of the lumbar spine”, Spine 16:793-799 (1991) de Groot JH, “The shoulder: a kinematic and dynamic analysis of motion and loading”, Doctoral Dissertation, Delft Technical University (1999) de Groot JH and Brand R, “A three-dimensional regression model of the shoulder rhythm”, Clin Biomech 16:735-743 (2001) de Groot JH, Rozendaal LA, Meskers CA and Arwert HJ, “Isometric shoulder muscle activation patterns for 3-D planar forces: A methodology for musculo-skeletal model validation”, Clin Biomech 19:790-800 (2004). Garner BA and Pandy MG, “Musculoskeletal model of the upper limb based on the visible human male dataset”, Computer Methods in Biomechanics and Biomedical Engineering 4:93-126 (2001) Hogfors C, Sigholm G, Herberts P, “Biomechanical model of the human shoulder – I. Elements”, J Biomech 20:157-166, (1987) Hogfors C, Peterson B, Sigholm G, Herberts P, “Biomechanical model of the human shoulder – II. The shoulder rhythm”, J Biomech 24:699-709 (1991) Hogfors C, Karlsson D, and Peterson B, "Structure And Internal Consistency Of A Shoulder Model", J Biomech, 28:767-777 (1995)
175
Hughes RE, Rock MG, and An K-N, “Identification of optimal strategies for increasing whole arm strength using Karush-Kuhn-Tucker multipliers”, Clin Biomech 14:628-634 (1999) Lippitt SB, Vanderhooft JE, Harris SL, Sidles JA, Harryman DT, and Matsen FA, “Glenohumeral stability from concavity-compression: a quantitative analysis”, J Shoulder and Elbow Surgery 2:27-35 (1993) Stobbe TJ, “The development of a practical strength testing program in industry”, Doctoral Dissertation, University of Michigan (1982) Kumar S, “Isolated Planr Trunk Strengths Measurement in Normals: Part III – Results and Database”, Int. J. Ind. Erg 17:103-111 (1996) Roebuck JA, Kroemer KHE, and Thomson WG, Engineering Antrhopometry, WileyInterscience, New York (1975). Nieminen H, Niemi J, Takala EP, Viikari-Juntura E, “Load-sharing pattrns in the shoulder during isometric flexion tasks”, J Biomech 28:555-566 (1995). Laursen B, Jensen BR, Nemeth G, and Sjogaard G, “A model predicting individual shoulder muscle forces based on relationship between electromyographic and 3D external forces in static position, J Biomech 31:731-739, 1998. Makhsous M, “Improvements, validation and adaptation of a shoulder model”, Doctoral Dissertation, Chalmers University of Technology, Gothenburg, Sweden (1999) Meskers CGM, “Quantitative assessment of shoulder function in a clinical setting: methodological aspects and applications”, Thesis, Leiden University Medical Center (1998). Van der Helm FCT, "A Finite-Element Musculoskeletal Model Of The Shoulder Mechanism", J Biomech 27:551-569, 1994. Wu G, Van der Helm FCT, Veeger HEJ, Makhsous M, van Roy P, Anglin C, Nagels J, Karduna AR, McQuade K, Wang X, Werner FW, and Buchholz B, “ISB recommendation on definitions of joint coordinate systems of various joints for the reporting of human joint motion—Part II: shoulder, elbow, wrist and hand”, J Biomech (In Press)
176
APPENDICES
177
APPENDIX A INDIVIDUAL SUBJECT MUSCLE PREDICTION CORRELATIONS STRATIFIED BY AZIMUTH
Data Description: For each subject, the Pearson correlation coefficient between the model predicted muscle force and the experimentally measured electromyographic activity is indicated for each azimuth and muscle recorded. For the trapezius, deltoid, and infraspinatus totals, the additive combination of the muscle parts was used. These are followed by the correlations of the individual muscle parts. Subject #1 Muscle Latissimus Pectoralis Major Trapezius Lower and Middle Upper and Clavicular Deltoid Middle Deltoid Posterior Deltoid Anterior Deltoid Infraspinatus Biceps Triceps
-30 NPA NPA 0.21 0.80 -0.22 0.58 0.33 NPA 0.66 0.67 0.71 -0.36
0 0.64 NPA -0.09 0.74 -0.38 0.40 0.37 NPA 0.55 0.83 0.70 -0.11
Azimuth 45 0.60 NPA 0.22 0.49 -0.10 0.64 0.42 0.32 0.13 0.45 0.71 -0.32
90 0.20 0.00 0.06 0.16 -0.28 0.53 0.51 -0.02 0.33 -0.25 0.79 -0.43
120 0.07 0.00 0.56 0.34 0.22 0.56 0.58 0.20 0.64 -0.18 0.61 -0.43
Total 0.23 0.00 0.17 0.54 -0.12 0.52 0.38 0.26 -0.03 0.70 0.62 -0.34
-30 0.86 NPA 0.31 0.48 0.08 0.39 0.58 NPA -0.41 0.98 0.89 -0.11
0 0.87 NPA 0.83 0.90 0.34 0.76 0.80 0.76 0.06 0.98 0.48 0.01
Azimuth 45 0.45 NPA 0.41 0.74 0.16 0.47 0.75 -0.01 -0.15 0.86 0.86 -0.10
90 0.37 0.00 0.51 0.81 0.14 0.55 0.15 0.72 0.06 0.68 0.55 -0.17
120 0.53 0.00 0.53 0.65 -0.09 0.43 0.11 0.35 0.39 -0.15 0.74 -0.05
Total 0.50 0.00 0.40 0.57 0.13 0.55 0.48 0.25 0.14 0.82 0.68 -0.06
Subject #2 Muscle Latissimus Pectoralis Major Trapezius Lower and Middle Upper and Clavicular Deltoid Middle Deltoid Posterior Deltoid Anterior Deltoid Infraspinatus Biceps Triceps
178
Subject #3 Muscle Latissimus Pectoralis Major Trapezius Lower and Middle Upper and Clavicular Deltoid Middle Deltoid Posterior Deltoid Anterior Deltoid Infraspinatus Biceps Triceps
0 NPA NPA 0.51 0.42 -0.01 0.63 0.56 0.30 -0.20 0.71 0.76 0.29
Azimuth 45 0.80 0.00 0.18 0.51 -0.17 0.87 0.79 0.54 0.66 0.63 0.76 0.03
90 0.74 0.00 0.42 0.71 -0.49 0.77 0.80 0.56 0.15 0.60 0.67 0.06
120 0.75 0.00 0.59 0.66 0.07 0.78 0.82 0.71 0.38 0.80 0.72 -0.17
Total 0.68 0.00 0.41 0.43 0.36 0.69 0.66 0.40 0.35 0.77 0.63 0.08
-30 0.56 NPA -0.11 -0.01 -0.24 0.73 0.71 0.13 0.66 0.67 0.78 -0.20
0 0.43 NPA 0.11 0.25 -0.34 0.67 0.65 0.62 0.51 0.78 0.85 -0.48
Azimuth 45 0.39 NPA -0.13 0.62 -0.15 0.53 0.24 0.35 0.85 0.28 0.88 -0.45
90 0.66 NPA 0.31 0.77 -0.04 0.45 0.18 0.46 0.68 0.15 0.68 -0.45
120 0.62 NPA 0.38 0.60 0.03 0.81 -0.15 0.81 0.92 -0.13 0.78 -0.56
Total 0.23 0.00 0.41 0.42 0.29 0.61 0.31 0.66 0.57 0.28 0.61 -0.42
-30 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
0 NPA NPA 0.80 0.75 0.19 0.87 0.67 NPA 0.91 0.97 0.87 -0.02
Azimuth 45 0.83 NPA -0.03 0.68 -0.50 0.91 0.69 0.60 0.49 0.71 0.86 -0.47
90 0.75 NPA 0.13 0.55 -0.45 0.37 0.39 -0.18 0.06 0.53 0.53 -0.38
120 0.15 NPA 0.19 0.80 -0.50 0.43 0.51 0.13 0.48 0.57 0.82 -0.46
Total 0.60 0.00 0.14 0.63 -0.19 0.48 0.39 0.12 0.58 0.53 0.73 -0.35
-30 NPA NPA 0.15 -0.04 0.01 0.07 -0.02 0.01 0.37 0.81 0.72 0.11
Subject #4 Muscle Latissimus Pectoralis Major Trapezius Lower and Middle Upper and Clavicular Deltoid Middle Deltoid Posterior Deltoid Anterior Deltoid Infraspinatus Biceps Triceps Subject #5 Muscle Latissimus Pectoralis Major Trapezius Lower and Middle Upper and Clavicular Deltoid Middle Deltoid Posterior Deltoid Anterior Deltoid Infraspinatus Biceps Triceps
179
Subject #6 Muscle Latissimus Pectoralis Major Trapezius Lower and Middle Upper and Clavicular Deltoid Middle Deltoid Posterior Deltoid Anterior Deltoid Infraspinatus Biceps Triceps
0 0.02 NPA 0.43 0.63 0.59 0.86 0.74 0.59 0.12 0.49 0.64 0.25
Azimuth 45 0.09 -0.12 0.09 0.78 0.10 0.65 0.24 0.06 0.50 0.46 0.79 -0.51
90 0.63 -0.28 0.48 0.86 -0.45 0.61 0.42 0.26 0.61 0.48 0.62 -0.57
120 -0.05 0.17 -0.02 0.52 -0.18 0.41 0.39 -0.15 0.08 0.13 0.14 -0.51
Total 0.03 0.00 0.18 0.68 0.01 0.47 0.44 0.08 0.04 0.56 0.51 -0.34
-30 -0.01 NPA 0.53 0.55 -0.56 0.61 0.57 0.15 0.51 0.76 0.72 0.41
0 0.42 NPA 0.44 0.41 -0.26 0.49 0.42 0.02 0.35 0.76 0.46 -0.04
Azimuth 45 0.28 NPA 0.64 0.60 -0.52 0.33 0.30 -0.01 0.18 -0.05 0.62 -0.13
90 0.41 0.00 0.73 0.81 -0.36 0.51 0.42 0.64 0.13 0.41 0.62 -0.18
120 0.42 0.00 0.57 0.55 -0.38 0.72 -0.09 0.71 0.68 0.49 0.41 -0.21
Total 0.26 0.00 0.54 0.57 -0.25 0.54 0.18 0.64 0.32 0.65 0.53 -0.05
-30 NPA NPA 0.12 0.19 0.59 0.35 -0.18 0.70 -0.07 0.72 0.66 0.65
0 NPA NPA 0.31 0.33 0.73 0.85 0.89 NPA 0.00 0.51 0.78 -0.34
Azimuth 45 0.11 NPA -0.19 0.08 -0.21 0.42 0.45 -0.16 0.08 0.40 0.66 -0.22
90 0.08 0.00 0.35 0.46 0.26 0.52 0.49 -0.18 0.49 0.61 0.66 -0.38
120 0.35 NPA -0.35 0.31 0.07 0.44 0.11 0.32 0.39 0.53 0.35 -0.55
Total 0.03 0.00 -0.11 0.35 -0.18 0.41 0.50 0.10 0.07 0.71 0.53 -0.08
-30 NPA NPA 0.45 0.79 -0.07 0.61 0.69 0.10 -0.35 0.33 0.55 0.45
Subject #7 Muscle Latissimus Pectoralis Major Trapezius Lower and Middle Upper and Clavicular Deltoid Middle Deltoid Posterior Deltoid Anterior Deltoid Infraspinatus Biceps Triceps Subject #8 Muscle Latissimus Pectoralis Major Trapezius Lower and Middle Upper and Clavicular Deltoid Middle Deltoid Posterior Deltoid Anterior Deltoid Infraspinatus Biceps Triceps
180
APPENDIX B EMG CHANNEL VARIABILITY BY SUBJECT ACROSS ALL RECORDED TRIALS
Data Description: For each subject, the minimum, maximum, and average level of EMG activity is reported as a percentage of the EMG activity for a maximal exertion. Subject #1 Muscle Latissimus Dorsi Pectoralis Major, Total Pectoralis Major, Sternal Insertion Pectoralis Major, Clavicular Insertion Trapezius, Total Trapezius, Lower Trapezius, Upper Deltoid, Total Middle Deltoid Posterior Deltoid Anterior Deltoid Infraspinatus Biceps Brachii Triceps Brachii
EMG Activation % MVC Minimum Maximum Average 4% 45% 27% 3% 33% 20% 2% 9% 4% 4% 58% 13% 12% 37% 21% 4% 35% 20% 14% 49% 22% 11% 47% 26% 10% 91% 35% 4% 31% 12% 16% 71% 31% 7% 52% 19% 1% 20% 8% 3% 17% 8%
Subject #2 Muscle Latissimus Dorsi Pectoralis Major, Total Pectoralis Major, Sternal Insertion Pectoralis Major, Clavicular Insertion Trapezius, Total Trapezius, Lower Trapezius, Upper Deltoid, Total Middle Deltoid Posterior Deltoid Anterior Deltoid Infraspinatus Biceps Brachii Triceps Brachii
EMG Activation % MVC Minimum Maximum Average 4% 27% 12% 2% 17% 9% 2% 27% 3% 1% 27% 7% 11% 45% 23% 6% 58% 23% 9% 44% 22% 1% 13% 9% 1% 10% 5% 1% 16% 8% 1% 11% 5% 3% 25% 10% 1% 33% 9% 1% 17% 2%
181
Subject #3 Muscle Latissimus Dorsi Pectoralis Major, Total Pectoralis Major, Sternal Insertion Pectoralis Major, Clavicular Insertion Trapezius, Total Trapezius, Lower Trapezius, Upper Deltoid, Total Middle Deltoid Posterior Deltoid Anterior Deltoid Infraspinatus Biceps Brachii Triceps Brachii
EMG Activation % MVC Minimum Maximum Average 2% 13% 6% 2% 12% 5% 2% 8% 4% 1% 15% 3% 5% 38% 19% 6% 63% 24% 4% 29% 14% 10% 74% 29% 9% 104% 40% 4% 76% 24% 4% 52% 24% 3% 31% 13% 0% 13% 4% 2% 14% 5%
Subject #4 Muscle Latissimus Dorsi Pectoralis Major, Total Pectoralis Major, Sternal Insertion Pectoralis Major, Clavicular Insertion Trapezius, Total Trapezius, Lower Trapezius, Upper Deltoid, Total Middle Deltoid Posterior Deltoid Anterior Deltoid Infraspinatus Biceps Brachii Triceps Brachii
EMG Activation % MVC Minimum Maximum Average 20% 33% 23% 3% 31% 14% 5% 27% 9% 1% 35% 4% 4% 40% 20% 3% 41% 16% 4% 59% 24% 6% 61% 27% 8% 76% 31% 4% 52% 17% 6% 77% 34% 4% 24% 10% 2% 32% 11% 2% 20% 8%
182
Subject #5 Muscle Latissimus Dorsi Pectoralis Major, Total Pectoralis Major, Sternal Insertion Pectoralis Major, Clavicular Insertion Trapezius, Total Trapezius, Lower Trapezius, Upper Deltoid, Total Middle Deltoid Posterior Deltoid Anterior Deltoid Infraspinatus Biceps Brachii Triceps Brachii
EMG Activation % MVC Minimum Maximum Average 10% 20% 15% 2% 61% 8% 3% 121% 19% 1% 9% 2% 6% 45% 22% 4% 57% 22% 6% 49% 22% 6% 36% 16% 9% 65% 27% 3% 13% 6% 4% 62% 16% 1% 8% 5% 1% 29% 7% 4% 23% 10%
Subject #6 Muscle Latissimus Dorsi Pectoralis Major, Total Pectoralis Major, Sternal Insertion Pectoralis Major, Clavicular Insertion Trapezius, Total Trapezius, Lower Trapezius, Upper Deltoid, Total Middle Deltoid Posterior Deltoid Anterior Deltoid Infraspinatus Biceps Brachii Triceps Brachii
EMG Activation % MVC Minimum Maximum Average 5% 166% 31% 2% 43% 34% 1% 34% 4% 1% 29% 3% 14% 100% 29% 15% 83% 38% 11% 171% 20% 14% 70% 30% 13% 105% 39% 5% 60% 24% 8% 60% 26% 2% 48% 5% 1% 30% 7% 2% 33% 10%
183
Subject #7 Muscle Latissimus Dorsi Pectoralis Major, Total Pectoralis Major, Sternal Insertion Pectoralis Major, Clavicular Insertion Trapezius, Total Trapezius, Lower Trapezius, Upper Deltoid, Total Middle Deltoid Posterior Deltoid Anterior Deltoid Infraspinatus Biceps Brachii Triceps Brachii
EMG Activation % MVC Minimum Maximum Average 7% 23% 11% 2% 16% 8% 3% 9% 5% 1% 22% 5% 12% 64% 27% 2% 105% 31% 16% 32% 23% 16% 120% 41% 12% 212% 46% 7% 168% 30% 12% 114% 47% 2% 25% 10% 1% 34% 7% 2% 17% 5%
Subject #8 Muscle Latissimus Dorsi Pectoralis Major, Total Pectoralis Major, Sternal Insertion Pectoralis Major, Clavicular Insertion Trapezius, Total Trapezius, Lower Trapezius, Upper Deltoid, Total Middle Deltoid Posterior Deltoid Anterior Deltoid Infraspinatus Biceps Brachii Triceps Brachii
EMG Activation % MVC Minimum Maximum Average 10% 59% 26% 6% 45% 30% 4% 10% 6% 2% 37% 5% 12% 51% 30% 7% 67% 34% 15% 47% 25% 9% 51% 25% 11% 62% 32% 4% 37% 16% 5% 66% 28% 2% 11% 5% 1% 37% 11% 4% 39% 14%
184
APPENDIX C DISTRIBUTION OF EMG LEVELS DURING HOLD PHASE BY SUBJECT AND MUSCLE IN PERCENT MVC
Data Description: For each subject, a histogram of EMG levels for the hold phase of the movement is reported. The EMG levels are indicated as a percentage of EMG of that achieved for a maximal reference exertion. Subject #1
185
Subject #2
186
Subject #3
187
Subject #4
188
Subject #5
189
Subject #6
190
Subject #7
191
Subject #8
192
APPENDIX D MUSCLE FORCE AND ELECTROMYOGRAPHIC COMPARISON BY SUBJECT AND MUSCLE – GROUPED FOR ALL AZIMUTHS
A comparison plot of model-predicted muscle force and recorded electromyographic data is shown for each subject for eight muscle parts and the composite deltoid. Subject #1
193
Subject #2
194
Subject #3
195
Subject #4
196
Subject #5
197
Subject #6
198
Subject #7
199
Subject #8
200
APPENDIX E DYNAMIC CONCORDANCE MATRICES, BY SUBJECT
Data Description: The tables contain the dynamic concordance matrices described in Chapter IV, and they summarize the agreement of the model muscle force predictions (MFP) and experimentally recorded electromyographic (EMG) data in recognizing muscle activity above a threshold. “On C” indicates activity for both EMG and MFP. “On D” indicates activity by EMG, but not by the MFP. “Off D” indicates no EMG activity, but activity predicted by MFP. “Off C” indicates no activity in either EMG or MFP. “EMG on fraction’ is the proportion of the trials that EMG activity was above the threshold. “MFP on fraction’ is analogous for the MFP model. Finally, the “Concordance/Discordance Ratio” represents the relationship between the portion of the time the data were concordant and discordant for a given subject and muscle. The “Prime Movers” row values are calculated as an average of the deltoids and infraspinatus. Subject #1 Muscle Name
Latissimus Dorsi P. Maj., Stern. P. Maj., Clav. Lower Trapezius Upper Trapezius Middle Deltoid Posterior Deltoid Anterior Deltoid Infraspinatus Biceps Brachii Triceps Brachii Prime Movers
Concordance Matrix On C
On D
0.17 0.00 0.00 0.78 0.95 1.00 0.70 0.90 1.00 1.00 0.14 0.91
0.83 1.00 1.00 0.22 0.05 0.00 0.30 0.10 0.00 0.00 0.86 0.09
Off D Off C 0.00 0.00 0.00 0.00 0.80 1.00 0.30 0.43 1.00 0.29 0.62
1.00 1.00 1.00 1.00 0.20 0.00 0.70 0.57 0.00 0.71 0.38
201
EMG MFP On Concordance/ On Fraction Discordance Fraction Ratio 0.95 0.16 0.27 0.09 0.00 10.29 0.77 0.00 0.30 0.96 0.75 3.65 0.75 0.91 3.16 0.86 1.00 6.18 0.58 0.53 2.29 0.91 0.86 6.90 1.00 1.00 ∞ 0.80 1.00 3.94 0.65 0.19 0.52 0.84 0.86 4.70
Subject #2 Muscle Name
Latissimus Dorsi P. Maj., Stern. P. Maj., Clav. Lower Trapezius Upper Trapezius Middle Deltoid Posterior Deltoid Anterior Deltoid Infraspinatus Biceps Brachii Triceps Brachii Prime Movers
Concordance Matrix On C
On D
0.74 0.00 0.00 0.86 0.69 1.00 0.59 0.74 0.97 1.00 0.00 0.83
0.26 1.00 1.00 0.14 0.31 0.00 0.41 0.26 0.03 0.00 1.00 0.17
Off D Off C 0.46 0.00 0.00 0.00 1.00 0.80 0.55 1.00 1.00 0.24 0.77
0.54 1.00 1.00 1.00 0.00 0.20 0.45 0.00 0.00 0.76 0.23
EMG MFP On Concordance/ On Fraction Discordance Fraction Ratio 0.82 0.69 2.36 0.03 0.00 36.00 0.49 0.00 1.06 0.96 0.82 6.40 1.00 0.69 2.22 0.70 1.00 2.36 0.93 0.61 1.31 0.73 0.69 1.96 0.99 0.97 23.67 0.92 1.00 11.33 0.05 0.23 2.52 0.89 0.83 3.28
Subject #3 Muscle Name
Latissimus Dorsi P. Maj., Stern. P. Maj., Clav. Lower Trapezius Upper Trapezius Middle Deltoid Posterior Deltoid Anterior Deltoid Infraspinatus Biceps Brachii Triceps Brachii Prime Movers
Concordance Matrix On C
On D
0.50 0.00 0.00 0.99 0.75 1.00 0.99 0.49 1.00 1.00 0.25 0.88
0.50 1.00 1.00 0.01 0.25 0.00 0.01 0.51 0.00 0.00 0.75 0.12
Off D Off C 0.02 0.00 0.00 0.80 0.22 1.00 1.00 0.25 1.00 1.00 0.26 0.85
0.98 1.00 1.00 0.20 0.78 0.00 0.00 0.75 0.00 0.00 0.74 0.15
202
EMG MFP On Concordance/ On Fraction Discordance Fraction Ratio 0.49 0.26 2.90 0.15 0.00 5.83 0.63 0.00 0.58 0.94 0.98 15.40 0.89 0.70 3.10 0.96 1.00 26.33 0.89 0.99 7.20 0.95 0.48 1.00 0.99 1.00 81.00 0.48 1.00 0.91 0.59 0.26 0.82 0.87 0.88 3.67
Subject #4 Muscle Name
Latissimus Dorsi P. Maj., Stern. P. Maj., Clav. Lower Trapezius Upper Trapezius Middle Deltoid Posterior Deltoid Anterior Deltoid Infraspinatus Biceps Brachii Triceps Brachii Prime Movers
Concordance Matrix On C
On D
0.64 0.00 0.00 1.00 0.80 1.00 0.49 0.75 1.00 1.00 0.06 0.86
0.36 1.00 1.00 0.00 0.20 0.00 0.51 0.25 0.00 0.00 0.94 0.14
Off D On C 0.38 0.00 0.00 1.00 0.25 0.00 1.00 1.00 0.19 0.74
0.62 1.00 1.00 0.00 0.75 1.00 0.00 0.00 0.81 0.26
EMG MFP On Concordance/ On Fraction Discordance Fraction Ratio 0.46 0.50 1.69 0.47 0.00 1.11 0.51 0.00 0.95 0.90 1.00 8.75 0.95 0.77 3.88 1.00 1.00 ∞ 0.91 0.45 1.17 0.99 0.76 2.90 1.00 1.00 ∞ 0.76 1.00 3.11 0.65 0.10 0.47 0.93 0.85 4.52
Subject #5 Muscle Name
Latissimus Dorsi P. Maj., Stern. P. Maj., Clav. Lower Trapezius Upper Trapezius Middle Deltoid Posterior Deltoid Anterior Deltoid Infraspinatus Biceps Brachii Triceps Brachii Prime Movers
Concordance Matrix On C
On D
0.92 0.00 0.00 0.95 0.81 1.00 0.94 0.89 1.00 1.00 0.33 0.94
0.08 1.00 1.00 0.05 0.19 0.00 0.06 0.11 0.00 0.00 0.67 0.06
Off D Off C 0.63 0.00 0.00 1.00 0.57 1.00 0.74 0.50 1.00 1.00 0.38 0.82
0.38 1.00 1.00 0.00 0.43 0.00 0.26 0.50 0.00 0.00 0.63 0.18
203
EMG MFP On Concordance/ On Fraction Discordance Fraction Ratio 0.27 0.70 1.10 0.73 0.00 0.38 0.39 0.00 1.59 0.98 0.95 13.67 0.84 0.77 3.00 0.95 1.00 21.00 0.39 0.82 1.10 0.82 0.82 4.50 0.80 1.00 3.89 0.48 1.00 0.91 0.82 0.34 0.63 0.75 0.91 3.00
Subject #6 Muscle Name
Latissimus Dorsi P. Maj., Stern. P. Maj., Clav. Lower Trapezius Upper Trapezius Middle Deltoid Posterior Deltoid Anterior Deltoid Infraspinatus Biceps Brachii Triceps Brachii Prime Movers
Concordance Matrix On C
On D
0.24 0.04 0.06 0.71 0.97 1.00 0.82 0.78 1.00 1.00 0.52 0.89
0.76 0.96 0.94 0.29 0.03 0.00 0.18 0.22 0.00 0.00 0.48 0.11
Off D Off C 0.00 0.03 0.03 0.92 1.00 0.83 0.50 1.00 1.00 0.52 0.91
1.00 0.97 0.97 0.08 0.00 0.17 0.50 0.00 0.00 0.48 0.09
EMG MFP On Concordance/ On Fraction Discordance Fraction Ratio 0.94 0.23 0.41 0.28 0.04 2.46 0.59 0.05 0.77 1.00 0.71 2.46 0.84 0.96 4.93 0.99 1.00 82.00 0.86 0.82 2.61 0.98 0.77 3.37 0.88 1.00 7.30 0.90 1.00 9.38 0.75 0.52 1.02 0.92 0.90 4.87
Subject #7 Muscle Name
Latissimus Dorsi P. Maj., Stern. P. Maj., Clav. Lower Trapezius Upper Trapezius Middle Deltoid Posterior Deltoid Anterior Deltoid Infraspinatus Biceps Brachii Triceps Brachii Prime Movers
Concordance Matrix On C
On D
0.54 0.00 0.00 1.00 0.78 1.00 0.38 0.74 1.00 1.00 0.14 0.85
0.46 1.00 1.00 0.00 0.22 0.00 0.62 0.26 0.00 0.00 0.86 0.15
Off D Off C 0.42 0.00 0.00 1.00 0.59 0.00 1.00 1.00 0.09 0.67
0.58 1.00 1.00 0.00 0.41 1.00 0.00 0.00 0.91 0.33
204
EMG MFP On Concordance/ On Fraction Discordance Fraction Ratio 0.59 0.49 1.25 0.28 0.00 2.52 0.68 0.00 0.47 0.93 1.00 12.50 0.33 0.65 1.13 1.00 1.00 ∞ 0.94 0.36 0.72 1.00 0.74 2.86 0.96 1.00 26.00 0.84 1.00 5.23 0.43 0.11 1.38 0.86 0.82 3.43
Subject #8 Muscle Name
Latissimus Dorsi P. Maj., Stern. P. Maj., Clav. Lower Trapezius Upper Trapezius Middle Deltoid Posterior Deltoid Anterior Deltoid Infraspinatus Biceps Brachii Triceps Brachii Prime Movers
Concordance Matrix On C
On D
0.11 0.00 0.05 0.71 0.95 1.00 0.82 0.82 1.00 1.00 0.37 0.89
0.89 1.00 0.95 0.29 0.05 0.00 0.18 0.18 0.00 0.00 0.63 0.11
Off D Off C 0.00 0.00 1.00 0.67 1.00 1.00 1.00 1.00 0.29 0.97
1.00 1.00 0.00 0.33 0.00 0.00 0.00 0.00 0.71 0.03
EMG MFP On Concordance/ On Fraction Discordance Fraction Ratio 1.00 0.11 0.13 0.33 0.00 2.05 0.69 0.03 0.53 0.97 0.72 2.21 0.95 0.93 11.20 1.00 1.00 ∞ 0.98 0.82 4.08 0.98 0.82 4.08 0.90 1.00 9.17 0.66 1.00 1.90 0.89 0.36 0.69 0.92 0.90 4.69
Average for all Subjects Muscle Name
Concordance Matrix On C
Latissimus Dorsi Pec Maj, Stern Pec Maj, Clav Lower Trapezius Upper Trapezius Middle Deltoid Posterior Deltoid Anterior Deltoid Infraspinatus Biceps Brachii Triceps Brachii Prime Movers
0.48 0.01 0.01 0.87 0.84 1.00 0.72 0.76 1.00 1.00 0.23 0.88
On D 0.52 0.99 0.99 0.13 0.16 0.00 0.28 0.24 0.00 0.00 0.77 0.12
Off D
Off C
0.27 0.00 0.00 0.69 0.58 1.00 0.58 0.60 1.00 1.00 0.28 0.79
0.73 1.00 1.00 0.31 0.42 0.00 0.42 0.40 0.00 0.00 0.72 0.21
205
EMG MFP Concordance On On Discordance Fraction Fraction Ratio 0.69 0.39 1.26 0.29 0.00 7.58 0.59 0.01 0.78 0.95 0.87 8.13 0.82 0.80 4.08 0.93 1.00 27.58 0.81 0.67 2.56 0.92 0.74 3.45 0.94 1.00 25.17 0.73 1.00 4.59 0.60 0.26 1.01 0.87 0.87 4.02
APPENDIX F LIST OF MUSCLE FORCES PREDICTED BY THE BIOMECHANICAL SHOULDER MODEL
Number 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38
Muscle Modeled (Muscle Portion) Latissimus Dorsi I (Upper) Latissimus Dorsi II (Lower) Levator Scapulae Omohyoid Pectoralis Major I (Sternal Insertion) Pectoralis Major II (Clavicular Insertion) Pectoralis Minor Rhomboid Minor Rhomboid Major Serratus Anterior I (Upper) Serratus Anterior II (Middle) Serratus Anterior III (Lower) Sternocleidomastoid Sternohyoid Subclavius Trapezius I (Intermediate) Trapezius II (Lower) Trapezius III (Upper) Trapezius IV (Clavicular Insertion) Middle Deltoid Posterior Deltoid Anterior Deltoid Coracobrachialis Infraspinatus I (Upper) Infraspinatus II (Lower) Subscapularis I (Upper) Subscapularis II (Middle) Subscapularis III (Lower) Supraspinatus Teres Major Teres Minor Biceps I (Long) Biceps II (Short) Triceps I (Long) Triceps II (Medial) Triceps III (Lateral) Brachialis Brachioradialis
206
![Download [Pdf] Rockwood and Matsen's The Shoulder, 5e Full Pages](https://p.pdfkul.com/img/300x300/download-pdf-rockwood-and-matsens-the-shoulder-5e-_5b39769a1723ddb30f74fc56.jpg)
