Design of Smart Agents for e-Health Services in the Context of Communicating Objects (Conception d’Agents Intelligents pour les Services d’e-Santé dans le Contexte des Objets Communicants)

The context of this thesis fits into the realm of "e-health" that we keep developing for many years now in the OCIF team (Communicating Objects for the Internet of the Future) of the D2 department (Networks, Telecommunications, and Services) at the IRISA laboratory. For many years now, we continue to carry on this important activity and promoted this research at national and international levels. For instance, over the past 7 years, we organized and chaired the international IEEE e-HPWAS (e-Health Pervasive Wireless Applications and Services) event on e-health applications, networks, and services. This important event takes place every year in conjunction with the renowned IEEE WiMob conference (rank B) which continues to attract the best researchers and professionals in this field [1].

In this thesis, by exploiting the ecosystem of communicating objects (objects, networks, data, architectures) coupled with artificial intelligence, we aim to develop intelligent services focused on the needs of humans (user-centric approach). In particular, we aim to resolve the shortcomings of the current systems proposed to assist/monitor dependent and vulnerable people such as elderly and isolated persons [2]. Therefore, this work aims to design a global ecosystem that provides adaptable services that assist dependent people in a given space (i.e. a smart space, e.g. smart homes/buildings/cities) [3].

The proposed approaches should be based on the use of communicating objects which will be operated by smart algorithms that foster a better understanding of the user situations (e.g. risky situations for an elderly person). Hence, the proposed system can trigger the appropriate services according to the real needs that arise (e.g. alarms, need for remote consultation, etc.). The desired intelligence aims to understand the human behavior and also to anticipate (predict) the risks of the daily life of users and optimize the allocation of material and human resources (e.g. the solicitation of health stakeholders, power consumption, network exchanges, etc.) [4]. For home care services, the proposed approaches should consider the integration and collaboration between available data sources and objects (e.g. local sensors) which until now have rendered isolated functionalities such as presence detection, remote measurements, smoke alarms, etc.) [5].

Another targeted objective, that could be a consequence of the previous approaches, is to improve the quality and performance of decision support systems used by healthcare professionals. This contribution consists of proposing innovative and potential solutions based on the processing of the person's context and his environment (context awareness approach), the personalization of provided services (based on the person's profiles), the design of smart algorithms for learning, recognizing and predicting the real needs. The proposed algorithms should also have a positive impact on energy consumption involved by the used objects.

To mitigate the impact of heterogeneity of user profiles, it is necessary to consider a new structuring of the different existing models of activities of daily living (ADL) of monitored people. The main existing models of ADL are the Katz, Barthel scales and the Lawton scale for instrumental activities (IADL) [6]. Optimal structuring such activities should enable providing assistance and comfort services that are strongly linked to the context of the assisted person.

The thesis work includes the following aspects:

1- Refine the understanding of the context of users who evolve in a smart space. For example, the proposed approach must refine the abstract models used in the classification of the person's dependency levels (e.g. levels assessed by the AGGIR or SMAF models [6]) and the measurement/prevention scales of activity risks. (e.g. the Norton, Braden, and Doloplus scales).

2- Study and evaluation of the application of existing AI approaches for understanding, learning and recognizing the user's context. The developed system should propose a "global reasoning" based on elementary information provided by different objects/sensors/data sources and use prediction techniques with AI algorithms [4,7,8].

3- Optimization of the network traffic and energy consumption of the various components [4,5,6]. This optimization must concern the following levels: 1- the residential level (exchanges between the terminal nodes (e.g. sensors) on one hand and intermediate nodes such as a central node (e.g. a home-box) on the other hand; 2- inter-users level (e.g. between different spaces) [3,9].