Clinical Rehabilitation 2009; 23: 217–228

A proof of concept study for the integration of robot therapy with physiotherapy in the treatment of stroke patients Maura Casadio Neurolab, Department of Informatics, Systems and Telematics (DIST), University of Genova, Psiche Giannoni ART Education and Rehabilitation Center, Genova, Pietro Morasso and Vittorio Sanguineti Neurolab, Department of Informatics, Systems and Telematics (DIST), University of Genova, Italy Received 14th April 2008; returned for revisions 21st June 2008; revised manuscript accepted 21st July 2008.

Objective: To carry out a proof of concept study for integrating robot therapy with physiotherapy in the treatment of stroke patients. Design: A simple and ‘gentle’ paradigm of robot–patient interaction was designed in order to foster the re-emergence of smooth, active control patterns in coordinated shoulder/elbow reaching movements. A haptic robot was programmed according to a strategy of minimal, progressively reduced assistance, with a double representation of targets: (i) visual (circles on a screen) and (ii) haptic (robot-generated force fields). The protocol included trials with and without vision, in order to emphasize the role of proprioceptive feedback. The training paradigm included 10 sessions and more than 5000 movements. Subjects: Ten chronic, hemiparetic subjects; four controls provided reference values for the performance measurements. Outcome measures: Four performance indicators (derived from the analysis of the reaching trajectories); clinical/functional measures (Fugl-Meyer and Ashworth scales). Results: After robot therapy reaching movements became faster and smoother. The performance in the no-vision trials was at least as good as in the vision trials. The Fugl-Meyer arm scores also increased significantly and remained approximately constant at follow-up; the Ashworth scores did not change. Conclusion: In spite of its simplicity, a limited number of ‘gentle’ robot therapy sessions appear to be beneficial, even for severely impaired patients, although no firm conclusion can be drawn at this point. However, the study provides support material for the careful design of controlled clinical trials and for a better integration with physiotherapy.

Introduction

Address for correspondence: Pietro Morasso, Department of Informatics, Systems and Telematics (DIST), University of Genova, Via Opera Pia 13, 16145 Genova, Italy. e-mail: [email protected]

In recent years evidence has mounted about the capacity of the central nervous system to alter its structure and function throughout all sorts of life experiences, including following injury, through a complex network of interacting processes.1–4

ß SAGE Publications 2009 Los Angeles, London, New Delhi and Singapore Downloaded from http://cre.sagepub.com by Pietro Morasso on April 16, 2009

10.1177/0269215508096759

218

M Casadio et al.

Animal models of focal brain injuries suggest that behaviour is probably the most powerful modulator of post-injury recovery.5–7 Beyond the timedependent spontaneous neurological recovery,8 the principal process responsible for functional recovery is the use-dependent reorganization of neural mechanisms made possible by neural plasticity. In view of this, it seems likely that practice is required to aid recovery from neurological damage, and that guided practice may be the most effective form. Usually such practice is given by therapists, but they have limited time and may not be consistent. In contrast, the use of robots offers a way of giving consistent ongoing opportunities to practise. Conventional physiotherapy and robot therapy should not be viewed as alternatives but could complement each other in the rehabilitation process. However, the optimal type of physiotherapy for patients with stroke is unclear, and only a few studies have been carried out to compare in a systematic way the different methods available. A clear consensus on the experimental procedures is still missing.9–14 Nonetheless, in spite of differences we think that there is a de facto convergence, at least for Bobath15 and MRP16 approaches, towards a system-oriented approach, which emphasizes problem-solving and skill learning, aiming at the identification for each patient of specific milestones that stress the underlying neural plasticity. This approach to therapy is clearly consistent with ‘gentle’ robot-therapy systems, an example of which is reported in this work. Although the application of robot technologies to the rehabilitation of neurological patients is about two decades old, the number of clinical studies is still limited, as documented in recent reviews.17,18. A crucial point, in our opinion, is that (robot) rehabilitation techniques can be efficient if they can promote increasing levels of motor skill: repetitive use alone, without learning-inducing variability, is unlikely to foster the largescale, long-lasting changes in cortical networks that are necessary for recovery of function. A relevant consequence is that the haptic interaction between the patient and the therapist, whether human or robotic, must not be invasive and unidirectional but rather should aim at supporting the emergence of functional control patterns by means of a minimal degree of assistance when

helping the patient to carry out a movement such as reaching.19 Therefore, in order to emulate the inherent compliance of human–human interaction, a robot therapist should be highly compliant, allowing a bidirectional interaction with the patient.20–22 This suggests that certain robotic devices, such as those frequently derived from the technology of industrial robots that enforce passive movements without requiring any active intervention of the patient at the neuromuscular and cognitive levels, should not be used. The robot assistance must match the degree of voluntary control in order to foster its improvement. This paper reports a pilot clinical study, in which the paretic arm of chronic hemiparetic patients was treated with a protocol of robot therapy that assisted the patients in performing reaching movements by means of a smooth robot-activated force field. An element of novelty is that target representation is both visual (on a computer screen) and haptic (by the assistive force field), thus allowing alternate trials with force field plus vision, or with only force field; the latter, in particular, are meant to focus the patient’s attention on the proprioceptive aspects of movements. The results of the pilot study show the emergence of features that are typical of normal reaching movements (approximately straight paths with bell-shaped velocity peaks23), suggesting that the patients have recovered at least part of their ability to generate coherent active patterns. Moreover, the ability to perform the task in the absence of vision supports the claim that the subjects may have improved the perception of their paretic limb. We believe that this study is relevant for physiotherapists and rehabilitation teams because it helps them understand what can be expected of a robot, which features are under control, which measurements can be carried out, and which general considerations are appropriate for the design of optimal rehabilitation protocols that combine human and robot intervention. In the end, this is the prerequisite for going from a proof of concept study to a controlled clinical trial.

Materials and methods Subjects Ten hemiparetic subjects (3 men, 7 women, age 53  13 years) participated in this study (Table 1).

Downloaded from http://cre.sagepub.com by Pietro Morasso on April 16, 2009

Robot therapy with physiotherapy Table 1 Subject

219

Demographic and clinical data of subjects Age (years)

S1 S2 S3c S4 S5 S6 S7 S8 S9 S10

72 59 53 69 57 34 30 46 53 55

Mean SD

52.8 13.4

Disease duration (months)a

Ashworth

Sex

Aetiologyb

Paretic hand

6 5 8 12 17 13 6 6 41 36

3 3 3 1þ 3 1þ 2 2 1 1

M F M F M F F F F F

I I I I I I I H H H

L R L R L R L R R L

15.0 13.0

2.1 0.8

FMA

28 39 154 25 39 24 12 26 39 76 46.2 41.5

a

Refers to the initiation of the robot therapy. Either ischemic (I) or haemorrhagic (H). Subject 3 quit after the sixth session. FMA, arm portion of Fugl-Meyer score (0–66); Ashworth, scale of muscle spasticity (0–4), at the beginning of robot therapy. b c

Subjects were recruited among those followed as outpatients of the ART Rehabilitation and Educational Center in Genova. When they entered the robot therapy protocol they had been attending regular rehabilitation sessions, according to the Bobath concept (on average once per week), for at least six months. This routine was continued throughout the study. The inclusion criteria were chronic (at least one year after stroke) and stable (at least one month before entering robot therapy) clinical conditions. The exclusion criteria were inability to understand instructions about the exercise protocol and other neurocognitive problems. Preference was given to patients with a high degree of motor impairment. In the population of patients disease duration was 46.2  41.5 months, with a majority of ischaemic aetiology (7/10). The level of impairment was evaluated by means of the Fugl-Meyer score, limited to the arm section (FMA).24 Two therapists with more than 20 years of experience administered the clinical evaluations; one of them was blinded to the type of robot therapy provided to ensure the consistency of the testing procedure. Five patients had a severe impairment (FMA510/66); 3 patients had an impairment of intermediate level (105FMA520); 2 patients had a mild impairment (FMA420). The average FMA

score was 15  13. The average Ashworth Scale score of muscle spasticity25 was 2.1  0.8. We also carried out a preliminary test of the assistance protocol with four control subjects aged 29.8  2.5 years. The purpose of this test was double: (1) to verify that the experimental setup, for the range of force fields used with the patients, did not alter the general features of reaching movements reported in the literature23; (2) to provide reference values for the performance indicators described in the following. The research conforms to the ethical standards laid down in the 1964 Declaration of Helsinki, which protect research subjects, and to ethical bylaws of the International Association of Bobath Instructors (IBITA: art. IV of the statute). Each subject signed a consent form that conforms to these guidelines.

Experimental apparatus The robot (Braccio di Ferro) is a planar manipulandum with 2 degrees of freedom, which has been fully described elsewhere.22 It has a rigid parallelogram structure with two direct-drive motors that provide a low intrinsic mechanical impedance at the end-effector and a full backdriveability. The robot is impedance-controlled in order to transmit

Downloaded from http://cre.sagepub.com by Pietro Morasso on April 16, 2009

220

M Casadio et al.

smoothly modulated force fields to the hand of the patient. The subject sits in a chair, with chest and wrist restrained by means of suitable holders, and grasps the Braccio di Ferro handle. A light, soft support is connected to the forearm in order to allow low-friction sliding on the horizontal surface of a table. Only the shoulder and elbow can move and motion is restricted to the horizontal plane, with no influence of gravity. The height of the seat is adjusted so that the arm can be kept approximately horizontal. A 19-inch LCD screen is positioned in front of the patients at a distance of about 1 m in order to display the positions of hand and target in the form of 2-cm-diameter circles of different colours (Figure 1).

Experimental design and training protocol The experimental design was specifically focused on facilitating active execution of large, outward-reaching movements and thus the chosen task consisted of reaching a set of targets,

Figure 1 A view from above of a subject holding the manipulandum of the Braccio di Ferro haptic robot. The subject’s shoulders are strapped to a chair; the forearm slides on the table surface; the wrist is stabilized by means of a skateboard wrist brace and the hand grasp by means of a Velcro holder. The targets (circles with a 2 cm diameter) are arranged on three shell layers: A, B, C. The distance between adjacent shells is 10 cm; the distance between targets on the same shell is 6.26 cm (shell A), 8.77 cm (shell B), 5.65 cm (shell C). The C shell is in front of a virtual wall. The basic sequence of target activation is A ! C ! B ! A and it is repeated 3  7  3 ¼ 63 times in a random order.

arranged in the horizontal plane (Figure 1) according to three shell layers: inner (A, 3 targets), middle (B, 3 targets) and outer (C, 7 targets), which required an almost full arm extension. Target sequences were generated according to the following scheme: A ! C ! B ! A. When a target was selected, it was presented to the subject in two modalities: visual (as a circle on the computer screen, with a 2-cm diameter) and haptic (as a robot-generated force field, directed toward the target whichever the position of the hand). This control scheme does not impose any constraint on either the timing of the reaching movement or the trajectory that subjects have to follow in order to reach the target. Two types of trials were alternated in the treatment protocol: (1) open-eyes trials and (2) closed-eyes trials. In the latter case the subject could perceive the direction of the target from the direction of the force vector applied to the hand. In this way, the robot-generated force field carried out two complementary functions: (1) it provided motor assistance and (2) it induced the patient to focus his or her attention on the proprioceptive feedback. For this reason, the force field was activated immediately at the target onset, but not in a step-like manner, in order to avoid a jerky pull: we used instead a smooth activation profile, with a rise-time of 1 second. This duration was chosen to minimize the influence of the robot as reaching patterns approached the typical performance of normal subjects: approx. 0.25 s of reaction time þ approx. 0.75 s of reaching time. The value of the field intensity achieved at the end of activation profile was chosen according to the protocol explained below. The force was switched off as soon as the subject hit the target. The next target was presented after a pause of 1 s. This pause was introduced because whereas a normal subject is likely to stay on target after the force field is turned off, we may expect patients to perform a ‘rebound movement’, with a velocity peak that is higher, the higher their hypertonus. This gives us an opportunity to detect if the hypertonus of the patients goes up or down as a sideeffect of the treatment, by looking at the velocity peak of the rebound movement. In addition to the assistive force field described above, the haptic environment implemented by the robot included two additional components: a mild

Downloaded from http://cre.sagepub.com by Pietro Morasso on April 16, 2009

Robot therapy with physiotherapy viscous field, which was intended to damp occasional hand oscillations, and a stiff virtual wall, in order to prevent going beyond the C shell of targets and, at the same time, provide a proprioceptive feedback about the successful achievement of the movement to the outward target. The parameters of this haptic control scheme were chosen during preliminary experiments with normal subjects: rise time of the assistive field ¼ 1 s; viscous coefficient ¼ 12 Ns/m, and elastic coefficient of the wall ¼ 1000 N/m. Each treatment session consisted of several blocks of trials that included repetitions of the A ! C ! B ! A sequence with different targets in random order, for a total of 3  3  7 ¼ 63 movements. The protocol started with a test phase, in which the patient was familiarized with the apparatus and with the range of assistive forces. This phase was supervised by a physical therapist, who observed the response to the different force levels and selected the minimum level capable of inducing at least a partial movement in the direction of the target. In this manner, the patients could indeed reach the targets but usually after a sequence of several submovements. Open-eyes and closed-eyes blocks of trials were alternated in the same session. The first training session used the same level of force determined in the test session for a total of 126 movements (63 with open eyes and 63 with closed eyes). After a little rest, the therapist considered the level of performance and asked the subject about fatigue. The decision could be (1) to terminate the session, (2) to continue with the same force level, (3) to continue with a reduced force that was 10–20% smaller than the previously used force. The procedure was iterated, with possibly reduced assistive force levels until the decision to stop was agreed by the patient and the therapist and in any case no later than 75 minutes after the initiation of the session. In following sessions, the training always started with two blocks of the initially selected force level and continued with decreasing levels of force assistance until one of the previously defined termination conditions occurred. In other words, the strategy of adaptive reduction of the assistance level is non-monotonic in the sense that is characterized by a saw-tooth pattern: it decreases within a session but it regains the original level at the

221

beginning of the next session. If a patient reached a level of assistance with a force below 4 N, the closed-eye block was eliminated because that level of force is very close to the perceptual threshold26 and thus is insufficient to allow the patient to perceive the direction of the target without vision. Summing up, the training procedure is tailored to the individual patients in two respects: (1) the initial level of the assistance force is tuned according to the therapist judgement; (2) the number of blocks of trials depends on the performance of the patient and the speed of learning from session to session. Overall, the protocol included 10 sessions (one session per week, with a duration ranging between 60 and 75 minutes), plus the initial test session. During that time, all subjects continued their physiotherapy programme with the same schedule and the same routine. The FMA and Ashworth scores were evaluated at the beginning of the robot therapy protocol and immediately after its completion. A follow-up evaluation was carried out three months later. The robot training sessions were carried out at Neurolab-DIST, University of Genova, under the supervision of a physical therapist (PG). The physiotherapy sessions were carried out at the ART premises.

Data analysis and performance evaluation Hand trajectories and the forces generated by the robot were sampled at 1000 Hz and stored at 100 Hz. Hand speed was estimated by using a fourth-order Savitzky–Golay smoothing filter. The analysis focused on the outward movements (A ! C) and, in particular, we defined the following performance indicators, which were chosen because it is easy to identify the normal reference values:  Mean speed (I1): average hand speed, computed from the time of target presentation to the time at which the subject reaches the target. This indicator is expected to increase as training proceeds, approaching a normal value of about 20 cm/s, which is the average value of control subjects.  Number of submovements (I2), identified by the number of peaks in the speed profile, after we

Downloaded from http://cre.sagepub.com by Pietro Morasso on April 16, 2009

222

M Casadio et al.

eliminated ‘spurious’ peaks according to a simple threshold mechanism. (Note: ‘Spurious’ velocity peaks were eliminated by means of two criteria: (1) a threshold on the speed (0.01 m/s), (2) a threshold on the time interval between one peak and the next one (0.3 s).) The reaching movements of the control subjects were characterized by a single-peaked, bell-shaped speed profile23 and thus this indicator should approach 1, as training proceeds.  Endpoint error after the first submovement (I3), defined as the distance between hand position and the position of the target at the end of the first submovement. As training proceeds, this indicator should go down to 0.  T-ratio (I4), defined as the ratio between the duration of the first submovement and the total time required for reaching the target. As training proceeds, this indicator should increase, up to the value of 1 (or 100%) that characterizes the performance of control subjects. Statistical analysis The performance improvements of each indicator, which could be attributed to the robot therapy sessions, were evaluated in two different manners:

considering all the different levels of assistive force experienced by each subject. For this purpose, we used a mixed-effect model, with three fixed factors (session, vision and force) and two additional random factors (subject and target) to properly account for the correlations among repeated measures from the same subject: 0

Indicator0 ¼ b0 þ b1  ð0 session0  1Þ þ b2  0 force0 þ b3  0 vision0

ð1Þ

where ‘indicator’ is one of the four parameters defined above, ‘session’ is an integer (from 1 to 10), ‘force’ is the field intensity (in N), and ‘vision’ is a binary variable that designates whether the exercise was carried out with open or close eyes. Model coefficients can be interpreted as follows: b0 ¼ ‘baseline performance level’; b1 ¼ ‘rate of improvement session by session’; b2 ¼ ‘rate of improvement dependent on the assistance level’; b3 ¼ ‘performance bias introduced by presence of vision’.

Results Performance improvements at the beginning of each session Recalling that each session started with the assistance level selected in the test session, the performance improvements were evaluated by means of a two-way, repeated measures ANOVA, with two factors: session (1–10) and vision (yes/no). To assess whether the session effects actually corresponded to an increase of performance over sessions, for each indicator we ran a contrast analysis (linear contrast) of the session effect and of the session  vision interaction.

Test with control subjects The control subjects were exposed to the training protocol with the full range of forces required by the population of patients, from 0 N to 25 N. In all cases, the subjects could reach the targets with a single movement, characterized by a velocity peak, which in most cases occurred before the end of the rise-time of the force field. This means that robot assistance automatically becomes less and less relevant as the performance level of a patient approaches the level of the control subjects.

Overall performance The analysis above evaluates session and condition effects for a fixed level of robot assistance. However, all indicators (in particular, the mean speed) are clearly dependent on the assistive force value. Therefore, a more complete picture of training performance can be obtained by

Improvement of performance indicators for the assistance force selected in the test session Figure 2 shows the evolution of I1 during the training sessions. The level of assistance is different for the different patients, according to the initial test session (e.g. F ¼ 25 N for S1, F ¼ 20 N for S2, etc.). The graphs clearly exhibit a positive

Downloaded from http://cre.sagepub.com by Pietro Morasso on April 16, 2009

Robot therapy with physiotherapy

223

Mean speed (cm/s) vs. session 18 16

18

18 S1 F = 25N

16

S2 F = 20N

16

18

18 S3 F = 18N

16

S4 F = 12N

16

14

14

14

14

14

12

12

12

12

12

10

10

10

10

10

8

8

8

8

8

6

6

6

6

6

4

4

4

4

4

2

2

2

2

2

4

6

8 10

2

4

6

2

8 10

4

2 2

6

4

6

8 10

2

18

18

18

18

14

S7 16 F = 6N 14

12

12

12

12

12

10

10

10

10

10

8

8

8

8

8

6

6

6

6

6

4

4

4

4

4

2

2

2

2

18 16

S6 F = 9N

2

4

6

8 10

Vision

2

4

S8 16 F = 5N 14

6

8 10

2

4

S9 16 F = 5N 14

6

8 10

S5 F = 9N

4

6

8 10

6

8 10

S10 16 F = 4N 14

2 2

4

6

8 10

2

4

No vision

Figure 2 Each panel shows the evolution of the I1 indicator (mean speed of the outward reaching movements) over the 10 training sessions for the different subjects at a level of assistance selected in the initial session. Bold lines refer to average values and dashed lines to standard errors. Subjects 8 and 9 initiate no-vision blocks in the second session with the same force level of the vision ones; subject 3 quit after the sixth session. The reference value (average speed of the normal population) is Vref ¼ 19.3  4.9 cm/s.

trend for all the subjects. Similar curves were obtained for the other indicators. Statistical analysis showed that the session effect is strongly significant for all indicators. In particular, it provided the following estimates of the indicator changes from the initial to the final session: I1 (mean speed) almost doubles from 5.61  0.94 cm/s at session 1, to 10.00  0.78 cm/s at session 10 (F(1,8) ¼ 15.47, P ¼ 0.004); I2 (number of peaks) is reduced from 5.71  1.04 to 1.69  0.21 (F(1,8) ¼ 16.01, P ¼ 0.004); I3 (endpoint error after the first submovement) is reduced from 5.05  1.22 cm to 3.06  0.33 cm (F(1,8) ¼ 16.98, P ¼ 0.003); I4 (the T-ratio) is increased from 47.1  8.9% to 83.8  5.4% (F(1,8) ¼ 29.28, P ¼ 0.0006). All together, the analysis of the performance indicators shows a linear improvement trend,

thus supporting the choice of a non-monotonic reduction of the assistance profile: the linear trend means that the patients remain equally responsive throughout the training protocol to the patterns of robotic assistance. With regard to vision, different subjects exhibit different behaviours, as exemplified in Figure 2 for I1: (a) in the majority of subjects there is no clear difference between the vision and no-vision situations; (b) in some subjects there is a tendency for the no-vision paradigm to prevail; (c) in other subjects the vision case is clearly better. Therefore, it is not surprising that, at the population level, no vision effect or vision  session interaction appears to be statistically significant. However, what is significant from the point of view of motor recovery is that in most subjects (8 out of 10) the deprivation of an importance sensory channel such as vision does not

Downloaded from http://cre.sagepub.com by Pietro Morasso on April 16, 2009

224

M Casadio et al. complementary information: the number of blocks of trials to which each patient was exposed in the sequence of training sessions. For example, subject 1 started with three blocks (for a total of 756 movements) and at the end he could go through 12 blocks (for a total of 3024 movements). On average, each subject performed more than 5000 reaching movements all together. The profile of progressive reduction of the assistance level was not always monotonically decreasing. In two cases (subject 7, session 6 and subject 8, session 3) the minimum level of assistance was slightly greater than in the previous session. This is due to the fact that according to the protocol during each session the patient went through the

necessarily imply a deterioration of the reaching performance; this further supports the working hypothesis that an experimental design that emphasizes the role of the proprioceptive channel has a relevant rehabilitation value.

Evolution of the assistance level Table 2 shows the progressive reduction of the assistance level after the initial test session for each patient. In the overall population of subjects, the initial level of assistance ranged between 5 N and 25 N and it was generally higher for patients with lower initial levels of the FMA score. Table 3 shows

Table 2 Assistance force (in N) selected in the ‘test’ session and then the progressive reduction of the minimum force adopted in the therapy sessions (SS1–SS10) Subjects

S1 S2 S3a S4 S5 S6 S7 S8 S9 S10 a

Tests

25 22 18 15 13 13 9 9 5 5

Sessions SS1

SS2

SS3

SS4

SS5

SS6

SS7

SS8

SS9

SS10

25 20 18 13 10 8 6 5 3 2

18 18 14 12 10 8 6 4 2 1

18 16 14 12 9 8 6 5 1 0

15 16 12 10 8 7 5 3 0 0

15 14 10 9 8 7 5 3 0 0

15 12 10 9 8 9 7 3 0 0

15 12

13 12

13 12

10 8

6 7 6 5 3 0 0

6 6 5 4 3 0 0

6 5 4 2 3 0 0

6 4 4 2 2 0 0

Subject 3 quit after the sixth session.

Table 3

Total number of training blocks used in each therapy sessions

Subjects

S1 S2 S3a S4 S5 S6 S7 S8 S9 S10

Sessions SS1

SS2

SS3

SS4

SS5

SS6

SS7

SS8

SS9

SS10

3 6 4 5 4 4 6 8 7 8

7 7 7 9 4 4 6 8 8 8

8 7 8 9 5 4 6 10 9 10

10 9 9 10 6 5 6 10 10 10

10 9 11 11 6 6 7 10 11 10

10 11 11 11 6 4 7 11 12 12

10 11

10 13

11 11

12 13

12 8 7 8 10 11 13

14 8 7 8 11 12 12

15 8 8 9 11 12 12

16 8 8 9 12 12 11

In each session, each subject was exposed to all the previously experienced force levels, in decreasing order, before attempting a further reduction of the force. a Subject 3 quit after the sixth session.

Downloaded from http://cre.sagepub.com by Pietro Morasso on April 16, 2009

Robot therapy with physiotherapy sequence of assistance levels of the previous session until he or she was willing to perform: in the two cases above the patient stopped before the end of the sequence.

Performance indicators: interaction between session and force The dependence of the performance indicators upon assistance level and session is shown in Figure 3 for subject 4: the subject initiates the therapy with a level of assistance of 15 N; this is reduced to 9 N at session 5, and is further reduced to 5 N at session 8. The figure shows that all indicators improve as a function of both the session and the level of assistance. This is confirmed by the regression analysis of all the indicators. More specifically, we found highly significant effects of session for I1, I2 and I4, but not for I3. As regards the level of assistance, 12 (cm/s)

225

all indicators display a significant effect (I1: P50.0001; I2: P ¼ 0.0006; I3: P ¼ 0.0009; I4: P50.0001). The b1 coefficients are positive for the two indicators (I1, I4) for which the improvement is indicated by an increase of the indicator (I1: b1 ¼ 0.36  0.097 cm/s per session; I4: b1 ¼ 2.98  0.77% per session), and negative for the two indicators (I2, I3) for which improvement is indicated by a decrement (I2: b1 ¼ 0.3  0.09 peaks per session; I3: b1 ¼ 0.29  0.08 cm per session). We also used the model of Equation (1) in order to assess the overall effect of therapy for each subject. For each indicator we computed the quantity
Ik ¼ b1  (‘session’–1), where ‘session’ ¼ 10 and k ¼ 1,2,3,4, which estimates the magnitude of the session effect at the end of training, extrapolated for the case of no assistance (‘force’ ¼ 0): this is a measure of the portion of the observed variation of the indicators that can be attributed to training. The results are reported in Table 4. 14 12

Mean speed

10

Number of sub-movements

10 8

8

6

6

F = 15N F = 9N

F = 5N

4 4 2 90

F = 5N

F = 15N 1

2

2

F = 9N 3

4

6

7

8

9

1

10

(%)

6 T-Ratio

2

3

7

8

9

10

Endpoint error after the 1st sub-movement

70

F =5N

4

F = 9N

3 F = 5N

2

30 F = 15N 10

6

(cm)

5

50

4

1

2

F = 9N 3 4 6 7 8 Robot therapy session

F =15N

1 9

10

0

1

2

3 4 6 7 8 Robot therapy session

9

10

Figure 3 Interaction of session and assistance level for subject 4. A force of 15 N is used throughout all the sessions; a force of 9 N is introduced at the fifth session; a force of 5 N is introduced at the eighth session. Other intermediate assistive force levels were used with this subject (13 N, 12 N, 6 N, 4 N, respectively) but are not plotted for simplicity.

Downloaded from http://cre.sagepub.com by Pietro Morasso on April 16, 2009

226

M Casadio et al.

Table 4 Variation of movement indicators between the beginning and the end of the robot training
I2 Number
I3 Endpoint Subject
I1 Mean speed of error (cm) (cm/s) submovements


I4 T-ratio (%)

FMAfollowup ¼ 20  13 Modifications of the hypertonus Hypertonus of the flexor muscles of the upper limb is a typical feature of most hemiparetic people. To assess indirectly whether the training protocol induced an increase of the activation of the flexors and, in general, the coactivation level of shoulder and elbow muscles, we looked at the bouncing movements that occur immediately after a C-target is reached. At that time the force field is switched off in a step-like fashion and the patient remains without robot assistance for 1 second before the next target is selected. A normal subject would stay on target, whereas the hypertonus of hemiparetic patients results in a rebound movement, the velocity peak of which is a good indicator of the level of the hypertonus. Statistical analysis showed that the peak speed of the rebound movements did not significantly change over sessions. This result suggests that there is no hint, even in severely impaired subjects, that the training protocols induces any kind of spasticity or increase of the muscle tonus. We attribute this important result to the fact that the protocol was designed to be as gentle as possible. This was also confirmed by the Ashworth score, which did not increase for any subject over the robot training period.

S1 S2 S3a S4 S5 S6 S7 S8 S9 S10

þ6.7 0.1 þ7.0 þ5.8 þ4.0 þ1.2 þ6.4 þ1.2 þ2.4 þ0.9

3.1 0.1 7.7 3.3 6.5 2.4 4.4 1.2 1.4 2.9

2.6 0.1 1.8 2.3 6.8 5.1 3.6 0.5 1.2 1.0

þ50.2 1.0 þ50.0 þ41.9 þ32.9 þ23.4 þ50.7 þ10.5 þ13.9 þ6.4

Mean SD

þ3.55 2.36

3.30 2.36

2.50 2.13

þ27.9 19.84

Computed according to the following
Ik ¼ b1  (‘session’–1). a Subject 3 quit after the sixth session.

equation:

Table 5 Arm portion of Fugl-Meyer score (FMA, max 66) before initiation of the therapy, at the end and three months afterwards Subject

FMA pre

FMA post

S1 S2 S3 S4 S5 S6 S7 S8 S9 S10

6 5 8 12 17 13 6 6 41 36

8 8 9 18 21 23 9 13 45 41

Mean SD

15.0 13.0

19.5 13.6


FMA 2 3 1 6 4 10 3 7 4 5


FMA% 33 60 13 50 24 77 50 117 10 14

4.5 2.6

45 34

resulted in a substantial preservation of the improvement:

FMA follow-up 7 7 9 22 18 24 11 16 42 41 19.7 12.9

a

Subject 3 quit after the sixth session.

Fugl-Meyer score Table 5 reports the FMA scores evaluated at three phases of the protocol: (1) before beginning the robot therapy protocol, (2) just after its termination, and (3) three months later. We found a statistically significant change (P ¼ 0.004) from FMAPRE ¼ 15  13 to FMAPOST ¼ 20  14, corresponding to an average
FMA ¼ 4.5  2.6 improvement. This is in line with previous studies,17 which report an average improvement of 3.7  0.5. Evaluation of the FMA at follow-up

Discussion The first conclusion of this proof of concept study is that a gentle approach to robot therapy is well tolerated and is also effective for severely impaired subjects. When patients are put in front of a machine or robot it is important to avoid making them feel that they are forced to perform an endless repetition of uninteresting actions. This was avoided in our case by giving the subjects freedom with regard to time and path, introducing enough variability in the exercise to make it interesting. The Braccio di Ferro was indeed perceived by our subjects as a benign and gentle agent that did not press them overly.

Downloaded from http://cre.sagepub.com by Pietro Morasso on April 16, 2009

Robot therapy with physiotherapy The protocol also included elements of challenge, for example facing reduced levels of assistance, that motivated the repetitions in order to improve performance. A negative side-effect of robot–human interaction could be to induce spasticity, which is velocity- and acceleration-dependent. This was avoided by using minimal assistance, which kept the movement velocity at relatively low levels, without any sharp acceleration peaks, and by the fact that each session started with a ‘warm-up’, using the assistance levels mastered in the previous sessions, before attempting a more challenging task. A related risk is fatigue, both muscular and cognitive. This study suggests that the challenge of the task was well tolerated, even by the most severely impaired patients, and the endurance of the patients increased with practice. In general, we believe it is important to monitor the willingness of the patients to continue the session without forcing them to a predetermined sequence. The results show significant improvements in smoothness and motor coordination for the reaching task: we observed a remarkable reduction in the degree of segmentation and the emergence of quasi-normal reaching patterns, with bell-shaped velocity profiles. It may be asked to what extent such improvements are also indicative of functional recovery. Studies on recovery from neural injury have suggested that smoothness is a result of a learned, coordinative process rather than a consequence of the structure of the neuromuscular system,27 and there is also evidence that the segmented structure of arm movements in stroke patients can be attributed to a deficit of interjoint coordination.28 Thus smooth movements result from improved coordination and this is a conditio sine qua non for functional recovery. In any case, our results show that higher smoothness is associated with better FMA score and this suggests that training general motion control abilities can be as effective as focusing primarily on specific functional movements. A peculiar feature of the proposed approach is that assistance is kept at a minimum level, thus avoiding as much as possible the risk that movements are performed in a passive way: movements must be assisted, not enforced by the robot therapist. Motor improvement is indeed only part of the story: all patients reported a better awareness of

227

their affected limb. This was confirmed by the therapist when delivering the FMA test, in particular with regard to the test items related to coordination (speed, tremor and dysmetria). Training in the closed-eyes condition is likely to be crucial in enhancing such awareness: with eyes closed patients must focus their attention on the purely haptic component of the task and this may help to reinforce the internal representation of the arm that is essential for delivering coordinated active commands. Having discussed the positive elements of the proposed approach to robot therapy it is also fair to draw attention to its intrinsic weaknesses and/or limitations. An obvious limitation of the Braccio di Ferro robot is that it only allows planar movements that involve the shoulder and elbow. Moreover, these movements are not ‘functional’ in a strict sense and one may ask to what extent the improved smoothness of reaching movements is useful for functional movements such as drinking a glass of water. This is indeed the main challenge faced in the design of optimal, integrated protocols of robot–human therapy, aimed at the ‘carryover’ of the observed gains in motor abilities to the activities of daily life. Our experience with this limited study emphasizes the need for multidisciplinary rehabilitation teams that include physiotherapists, rehabilitation medical doctors and rehabilitation engineers for the rational exploitation of robots in the rehabilitation of neurological patients. The tasks of the team are multifaceted and include the following items: (1) quantitative evaluation of the specific deficits/requirements of a patient (for example by using the same robotic system used for treatment), (2) identification of clear functional goals, (3) translation of the goals into robotic interaction paradigms, tailored for the patient and endowed with suitable coordination/ performance evaluation procedures. Defining, choosing and adapting protocols of robot therapy to the individual needs of different patients, in short personalizing robot therapy, is an open field that requires attention and research effort in the future. The purpose of this study was to provide support material and suggestions for the design of controlled clinical trials of integrated human–robot therapy.

Downloaded from http://cre.sagepub.com by Pietro Morasso on April 16, 2009

228

M Casadio et al. Clinical messages

 Design rehabilitation protocols in a patientoriented way, integrating robot therapy and physiotherapy.  Learn to exploit and understand the quantitative measurements coming out of patientrobot interaction during treatment.

14

15 16 17

References 1 Nudo RJ. Mechanisms for recovery of motor function following cortical damage. Curr Opin Neurobiol 2006; 16: 638–44. 2 Lee RG, van Donkelaar P. Mechanisms underlying functional recovery following stroke. Can J Neurol Sci 1995; 22: 257–63. 3 Carmichael ST. Plasticity of cortical projections after Stroke. Neuroscientist 2003; 9: 64–75. 4 Dancause N, Barbay S, Frost SB et al. Extensive cortical rewiring after brain injury. J Neurosci 2005; 25: 10167–79. 5 Jones TA, Chu CJ, Grande LA, Gregory AD. Motor skills training enhances lesion-induced structural plasticity in the motor cortex of adult rats. J Neurosci 1999; 19: 10153–63. 6 Biernaskie J, Corbett D. Enriched rehabilitative training promotes improved forelimb motor function and enhanced dendritic growth after focal ischemic injury. J Neurosci 2001; 21: 5272–80. 7 Kopp B, Kunkel A, Muhlnickel W, Villringer K, Taub E, Flor H. Plasticity in the motor system related to therapy–induced improvement of movements after stroke. Neuroreport 1999; 4: 807–10. 8 Kwakkel G, Kollen B, Twisk J. Impact of time on improvement of outcome after stroke. Stroke 2006; 37: 2348–53. 9 Dickstein R, Hocherman S, Pillar T, Shaham R. Stroke rehabilitation: three exercise therapy approaches. Phys Ther 1986; 66: 1233–38. 10 Lord JP, Hall K. Neuromuscular reeducation versus traditional programs for stroke rehabilitation. Arch Phys Med Rehabil 1986; 67: 88–91. 11 Ernst E. A review of stroke rehabilitation and physiotherapy. Stroke 1990; 21: 1081–85. 12 Wagenaar RC, Meijer OG et al. The functional recovery of stroke: a comparison between neurodevelopmental treatment and the Brunstrom method. Scand J Rehabil Med 1990; 22: 1–8. 13 Langhammer B, Sunnaas S. Bobath or Motor Relearning Programme? A comparison of two different approaches of physiotherapy in stroke

18

19

20

21 22

23 24

25 26

27 28

rehabilitation: a randomized controlled study. Clin Rehabil 2000; 14: 361–69. van Vliet PM, Lincoln NB, Foxall A. Comparison of Bobath based and movement science based treatment for stroke: a randomised controlled trial. J Neurol Neurosurg Psychiatry 2005; 76: 503–508. Bobath B. Adult hemiplegia: evaluation and treatment, 3rd edition. Butterworth-Heinemann, 1990. Carr JH, Shepherd RB. A motor relearning programme for stroke, Butterworth-Heinemann, 1987. Prange GB, Jannink MJ et al. Systematic review of the effect of robot-aided therapy on recovery of the hemiparetic arm after stroke. J Rehabil Res Dev 2006; 43: 171–84. Kwakkel G, Kollen BJ, Krebs HI. Effects of robotassisted therapy on upper limb recovery after stroke: a systematic review. Neurorehabil Neural Repair 2008; 22: 111–21. Casadio M, Morasso P, Sanguineti V, Giannoni P. Impedance-controlled, minimally-assistive robotic training of severely impaired hemiparetic patients. BioRob 2006 (the First IEEE/RAS-EMBS Int Conf on Biomedical Robotics and Biomechatronics), Pisa, Italy, 20–22 February 2006. Krebs HI, Hogan N, Volpe BT, Aisen ML, Edelstein L, Diels C. Overview of clinical trials with MIT-MANUS: a robot-aided neuro-rehabilitation facility. Technol Health Care 1999; 7: 419–23. Harwin WS. Robots with a gentle touch: advances in assistive robotics and prosthetics. Technol Health Care 1999; 7: 411–17. Casadio M, Morasso P, Sanguineti V, Arrichiello V. Braccio di Ferro: a new haptic workstation for neuromotor rehabilitation. Technol Health Care 2006; 13: 1–20. Morasso P. Spatial control of arm movements. Exp Brain Res 1981; 42: 223–27. Platz T, Pinkowski C et al. Reliability and validity of arm function assessment with standardized guidelines for the Fugl-Meyer Test. Clin Rehabil 2005; 19: 404–11. Bohannon RW, Smith MB. Interrater reliability of a modified Ashworth scale of muscle spasticity. Phys Ther 1987; 67: 206–207. Boff KR, Kaufman I, eds. Cutaneous sensitivity (chapter 12), Kinesthesia (chapter 13), Tactile perception (chapter 31). In Handbook of human perception and human performance, volume 1. J. Wiley and Sons, 1986. Rohrer B, Fasoli S, Krebs HI et al. Movement smoothness changes during stroke recovery. J Neurosci 2002; 22: 8297–304. Levin MF. Interjoint coordination during pointing movements is disrupted in spastic hemiparesis. Brain 1996; 119: 281–93.

Downloaded from http://cre.sagepub.com by Pietro Morasso on April 16, 2009