Non-monotonic Reasoning Supporting Wireless Sensor Networks for Intelligent Monitoring: The SINDI System Alessandra Mileo, Davide Merico, and Roberto Bisiani Department of Informatics, Systems and Communication, University of Milan-Bicocca, viale Sarca 336/14, I–20126 Milan

Abstract. In recent years there has been growing interest in solutions for the delivery of clinical care for the elderly, due to the large increase in aging population. Monitoring a patient in his home environment is necessary to ensure continuity of care in home settings, but this activity must not be too invasive and a burden for clinicians. We prototyped a system called SINDI (Secure and INDependent lIving), focused on i) collecting data about the person and the environment through Wireless Sensor Networks (WSN), and ii) reasoning about these data both to contextualize them and to support clinicians in understanding patients’ well being as well as in predicting possible evolutions of their health.

1

Overview

The life-expectancy in several countries continues to grow, but people who live longer tend to be in a state in which chronic conditions substantially cripple their quality of life and their autonomy. Future Independent Living systems should go beyond data collection to focus on assisting caregivers in understanding health evolution and enhancing autonomy of monitored patients. We refer to an intelligent monitoring system as a monitoring system that is able to reason about gathered data and support decisions. Most of the pervasive systems for healthcare proposed so far use a probabilistic approach to behaviour analysis and activity recognition aimed at enhancing autonomy [1]. These approaches are sometimes coupled with logicbased planning techniques. When we talk about intelligent monitoring, though, we refer to a different (potentially complementary) view of artificial intelligence applied to home healthcare. In our view, expressive knowledge representation and reasoning techniques are needed to analyse the context and to understand health evolution. In the SINDI system we address this issue by using non-monotonic logical reasoning in a two-step inference process: i) summarizing and correlating sensor data in a consistent interpretation of the context in which the person lives in terms of clinical profile, environment, movements, and ii) predicting possible evolutions of the person’s health in order to devise effective preventive strategies. E. Erdem, F. Lin, and T. Schaub (Eds.): LPNMR 2009, LNCS 5753, pp. 585–590, 2009. c Springer-Verlag Berlin Heidelberg 2009 !

586

A. Mileo, D. Merico, and R. Bisiani

The logical framework of Answer Set Programming (ASP) [2] is well suited to deal with similar complex knowledge representation and reasoning tasks, in that it overcomes most of the limitation of previous logic programming systems such as Prolog. Compared to pure statistical approaches, logic inference based on ASP is highly expressive and computationally more performant because it can deal with first-order representations, which are much richer than the propositional ones characterizing probabilistic inference. Furthermore, ASP can deal with incomplete information and commonsense reasoning using defaults. Cardinality and weight constraints together with program optimization techniques can also be used to model different degrees of uncertainty [3,4,5]. In the remainder of this paper we will describe how ASP-based Knowledge Representation and Reasoning supports Wireless Sensor Networks technologies in the SINDI Intelligent Monitoring system.

2

Context Data: Acquisition and Interpretation

Wireless Sensor Networks (WSNs) [6] consist of nodes that are capable of interacting with the environment by sensing or controlling physical parameters. These networks are the enabling technology for acquiring all possible information about the context in which the user lives. Data gathered by the sensors are processed and aggregated according to specific algorithms for feature analysis, features are then processed by the ASP program and made available at the upper levels. The SINDI’s context model aims at being generic and computationally rich at the same time, being based on a high level description of the home environment in terms of rooms, areas, objects, properties, relations and observations mapped into a set of logic predicates to be manipulated by the inference engine for health assessment [7]. The information flow is illustrated in Figure 1.

Studyy

Data aggregation Context interpretation Localization

Bath

Feature generation Livingroom Li i

Bedroom B d

Kitchen Balcony

Inference Engine (ASP)

Database (data from monitoring and knowledge representation)

Body Sensors

Sensor

Fig. 1. Flow of Data in the SINDI System

Internet

Caregiver and patient interfaces

Non-monotonic Reasoning Supporting WSN for Intelligent Monitoring

587

In the context of home monitoring, the more intuitively relevant aspects of a context are where you are, who you are (clinical profile and personal information), which resources you are using, what you are doing and when [8,9]. In order to represent this information, SINDI’s context model considers four entities on the first level: the Person entity, the Room entity, the Area entity and the Object entity. A small subset of generic1 spatial relations among entities are also defined, such as in, under, on, near. All the other pieces of information are at a second level and can be indexed by the primary context because they are attributes of the entities at the primary level. Values of both attributes and spatial relations (except the inclusion of an area in a room which is static) are dynamic and need to be associated to an interval of time. In this way, the reasoning system can take into account their dynamic evolution in context interpretation. Reasoning about the collected data is used to better characterize movements and to localize the person in the areas of the house. If we consider SINDI’s localization component , given proximity values with a certain accuracy P and defined over (possibly overlapping) time intervals Ti , Tj , the ASP program takes all available sensor data as input and identifies all possible consistent sequences of moves across rooms and areas. Logical rules for disambiguation state that, by default, proximity to an area A of a room R in a temporal segment T1,T2 is given by the fact that a signal has been received from the corresponding node in that temporal segment. This holds unless there is a more reliable signal received in the same interval from another node. This other signal determines proximity unless additional contextual data make it invalid (e.g. a mat sensor indicating pressure in a different area A1 of room R1) and in case several of these data are available, a measure of reliability can be used to identify the best solution. The high level representation of the collected data is automatically mapped into logic predicates resulting in a set of facts which is combined with the ASP logic program. Besides disambiguation, context interpretation is also a crucial phase in understanding basic behaviours that may be important for health assessment. Results are used to evaluate meaningful aspects of the elderly care as detailed in Section 3.

3

Reasoning Support for Intelligent Monitoring

In order to support caregivers in understanding and controlling the patient’s health-evolution in his home environment, the reasoning component of the SINDI system uses the results of context interpretation in two reasoning steps: static and dynamic evaluations of significant aspects of the patient’s quality of life (referred to as indicators) and prediction of possible evolutions of the patient’s health state to plan appropriate preventive strategies. To do this, SINDI uses a knowledge representation model of health described in Section 3.1. 1

The term “generic” here refers to the fact that they refer to relative spatial position of two entities, no matter where they are in the physical space. This results in greater generality in that we do not need an absolute physical description of the environment.

588

3.1

A. Mileo, D. Merico, and R. Bisiani

Knowledge Representation: SINDI’s Model of Health

SINDI’s Model of Health formalizes medical knowledge and correlations between indicators and health-related factors (referred to as items)2 . A careful analysis of health care in home settings suggests that items can be classified into three classes: i) functionalities, evaluated in terms of functional disabilities; ii) daily activities, evaluated in terms of level of dependence in performing daily activities and iii) risks, representing complex aspects of the elderly care. The relation between items and indicators is the following: each indicator can contribute to the evaluation of one or more items; items are correlated by dependency relations3 indicating how the value of an item may impact values of other items and how. The reasoning tasks detailed in Section 3.2 are performed by analyzing the graph of dependencies connecting an item I with related indicators as well as with other items. 3.2

Reasoning Tasks: Evaluation, Prediction and Prevention

Evaluation. At each reasoning cycle, results of the context interpretation process are used to infer consistent absolute and differential evaluations of indicators and items. In general, absolute evaluation of items is available only as inputs from caregivers according to results of specific tests. As for indicators, their absolute evaluations can be based on i) results of specific evaluation by clinicians (e.g. hearing functionalities), results of data aggregation (e.g. quality of movement), results of ad-hoc logic rules (e.g. quality of sleep). Differential evaluations are obtained, when possible, as a measure of the value increase or decrease derived by comparing values at the current inference cycle with values at the previous inference cycle and it has four possible outcomes: worsening, improvement, no substantial change, undefined. Given that different kinds of potentially contradictory dependencies are allowed in SINDI’s knowledge model, the combination of such influences of several indicators Indi on item I so as to provide a coherent differential evaluation for I is not always possible and the evaluation process returns a partially labelled graph as output. The following reasoning step of SINDI starts from this incomplete information and uses the computational power of the ASP framework in order to predict possible evolutions in terms of differential evaluations of items that have not been labeled as a result of the evaluation task and provide a qualitative analysis of results. Prediction. is identified as the identification of plausible effects of certain changes in items’ values on values of unlabeled items; intuitively, this is done by considering all possible consistent values for the missing information according to SINDI’s logic model of dependencies between items; prediction makes it possible to act before major symptoms and to plan appropriate short- and long-term interventions, thus reducing risks. Prediction involves 2

3

Indicators and Items have been identified according to the medical practice in health assessment of the elderly [10] and encoded in our declarative framework [11]. Dependency relations are specified by knowledge engineers and medical experts.

Non-monotonic Reasoning Supporting WSN for Intelligent Monitoring

589

several subtasks: i) contextualization, referred to as the correlation of context evaluation results with the clinical profile of the person; ii) local prediction, combining results of contextualization with medical dependencies to generate all total consistent labeling of the dependency graph resulting from evaluation and iii) local explanation referred to as the identification of all possible minimal set of dependency arcs that may justify the worsening of a person’s health (provided as a result of the prediction task). These tasks are local processes rather than a case-based ones, in that they take into account results of reasoning under particular clinical and environmental conditions. Prevention. Prevention [12] is referred to as all those interventions (feedback) that may keep health changes within safe boundaries. The combination of inference results (prediction) and context-related knowledge about the person and the environment is used to determine i) what should be provided as feedback, ii) in which form and iii) when. The content of the feedback is determined according to the medical literature (evidence-based studies) and encoded in the system. When the system determines a set of feedback actions that should be performed at a given time, they are qualitatively analyzed in order to infer which action is more urgent. A reaction to a feedback, when detected, is logged to be used at a later time. Exploring this history, caregivers can improve the way feedback actions are performed and identify the most effective communication patterns.

4

Preliminary Evaluation

So far we have run the full system for short periods of time (days) in a mock-up environment without real users. Our preliminary evaluation took into account how SINDI addresses most of the expected requirements [11]. The combination, in the reasoning process, of different sources of information (sensors, medical knowledge, clinical profile, user defined constraints) that change over time makes the system more reliable (i.e. much better able to disambiguate situations, thus reducing false positives) and adaptive (e.g. easily extended on the face of new available information). The modularity, espressivity and computational efficiency of the ASP framework make us strongly believe it is a good logic programming paradigm for automated reasoning in knowledgeintensive domains. In the first implementation of our system we evaluated ASP programs by using Gringo as grounder [13] and the Clasp solver [5] as inference engine4 . Security and privacy are guaranteed by the use of security standards and techniques and the use of off-the-shelf components in SINDI considerably reduces overall costs. Research-wise we are considering the problems of the encoding and integration of the appropriate medical knowledge and the use of apparently inconsistent prediction results to discover missing dependencies in the initial medical knowledge. 4

http://potassco.sourceforge.net/

590

A. Mileo, D. Merico, and R. Bisiani

Application-wise, given the high variability among trials and studies addressing prediction and prevention issues, it is still difficult to extract a coherent picture of what leads to disability and to develop coherent prevention strategies. In this respect, our system has the potential to automatically collect a massive amount of data in oder to evaluate context-related prediction patterns and effective communication strategies for prevention. These aspects are being concretely taken into account in the context of a real deployment of SINDI in a geriatrics hospital.

References 1. Haigh, K.Z., Yanco, H.: Automation as caregiver: A survey of issues and technologies. In: Proc. of AAAI 2002 Workshop on Automation as Caregiver, pp. 39–53 (2002) 2. Gelfond, M., Lifschitz, V.: The stable model semantics for logic programming. In: Proc. of ICLP 1988, pp. 1070–1080. MIT Press, Massachussets (1988) 3. Simons, P., Niemel¨ a, I., Soininen, T.: Extending and implementing the stable model semantics. Artificial Intelligence Journal 138(1-2), 181–234 (2002) 4. Leone, N., Pfeifer, G., Faber, W., Eiter, T., Gottlob, G., Perri, S., Scarcello, F.: The dlv system for knowledge representation and reasoning. ACM Transactions on Computational Logic 7(3), 499–562 (2006) 5. Gebser, M., Kaufmann, B., Neumann, A., Schaub, T.: Conflict-driven answer set solving. In: Proc. of IJCAI 2007, pp. 386–392. AAAI Press, Massachussets (2007) 6. Akyildiz, I.F., Su, W., Sankarasubramaniam, Y., Cayirci, E.: A survey on sensor networks. IEEE Communications Magazine 40(8), 102–114 (2002) 7. Mileo, A., Merico, D., Bisiani, R.: Support for context-aware monitoring in home healthcare. In: Proc. of IE 2009 (to appear, 2009) 8. Ryan, N.S., Pascoe, J., Morse, D.R.: Enhanced reality fieldwork: the context-aware archaeological assistant. In: Gaffney, V., van Leusen, M., Exxon, S. (eds.) Computer Applications in Archaeology 1997. British Archaeological Reports, Tempus Reparatum (1998) 9. Schilit, B., Adams, N., Want, R.: Context-aware computing applications. In: Proc. of the Workshop on Mobile Computing Systems and Applications, pp. 85–90. IEEE Computer Society, Los Alamitos (1994) 10. Fleming, K.C., Evans, J.M., Weber, D.C., Chutka, D.S.: Practical functional assessment of elderly persons: A primary-care approach. Mayo Clinic 70, 890–910 (1995) 11. Merico, D., Mileo, A., Pinardi, S., Bisiani, R.: A Logical Approach to Home Healthcare with Intelligent Sensor-Network Support. The Computer Journal (2009) bxn074 12. Mileo, A., Bisiani, R.: Context-aware prediction and prevention to extend healthy life years:preventing falls. In: Proc. of the IJCAI Workshop on Intelligent Systems for Assisted Cognition (to appear, 2009) 13. Gebser, M., Schaub, T., Thiele, S.: Gringo: A new grounder for answer set programming. In: Baral, C., Brewka, G., Schlipf, J. (eds.) LPNMR 2007. LNCS (LNAI), vol. 4483, pp. 266–271. Springer, Heidelberg (2007)