Agents acting and moving in healthcare scenario: a paradigm for telemedical collaboration

Vincenzo Della Mea

Institute of Pathology, University of Udine; p.le S.Maria della Misericordia, 33100 Udine Italy;

Address for correspondence: Vincenzo Della Mea Institute of Pathology, University of Udine; p.le S.Maria della Misericordia, 33100 Udine Italy Fax: +39-0432-559420 Email: [email protected]

IEEE Transaction on Information Technology in Biomedicine 2001: 5; 10-13

Agents acting and moving in healthcare scenario: a paradigm for telemedical collaboration Abstract The present paper describes a novel approach to the analysis and development of telemedicine systems, based on the multi-agent paradigm. An agent is an autonomous, social, reactive and proactive entity, sometimes also mobile. Since telemedicine is grounded on communication and sharing of resources, agents are suitable for its analysis and implementation, and we adopted them for developing a prototype telemedical agent.

Introduction Telemedicine is the practice of medicine at distance, by means of telematic tools. Such a definition includes a wide variety of tasks, ranging from telediagnosis, to distant teaching and learning, to several applications of distributed databases. All these tasks involve the sharing of knowledge, data, expertise, services, among healthcare professionals. In the last years, the multi-agent paradigm faced the information technology area, proposing a novel way of seeing software programs as autonomous, social, reactive and proactive entities, being sometimes also mobile. Each agent has some design aims, to be fulfilled either by operating alone or collaboratively with other agents. The overall behaviour of a multi-agent system is given by the resulting complex interaction among agents. Telemedicine systems may be described not only as single workstations eventually able to intercommunicate, but also as communities of interacting entities, aimed at supporting collaboration and resource sharing in the medical field. When adopting this point of view, a set of telemedicine facilities may be described using the multi-agent paradigm, which can then be embraced for its analysis and implementation. This is particularly true for mobile telemedicine systems, which introduce further complexity and dynamism. Aim of this paper is to introduce the multi-agent paradigm, and then to describe a prototype telemedicine application based on it.

The multi-agent paradigm One main concept for defining agents is the intentional system, developed by Dennett [1], which identifies an entity with a behaviour predictable by attributing belief, desires and rational acumen to it. Agent theories usually adopt these or even other anthropomorphic attitudes (e.g. intention, obligation, commitment, choice, etc) as abstractions to describe artificial agents and their behaviour.

Although there is no widely agreed definition of agenthood, some properties are commonly accepted: -

autonomy: an agent has control over its own actions, and maintains an internal state;

-

sociality: an agent interacts with others by means of an agent communication language (ACL) [2];

-

reactivity: an agent may act as a consequence of changes in the environment;

-

proactivity: an agent acts to pursue its design aims, thus showing an opportunistic behaviour.

In an agent-based development system, these properties correspond to specific primitives available to the developer. Usually, the multi-agent paradigm is compared to the distributed objects paradigm, because both involve entities having an internal state, with an interface based on message passing, and both allow for similar abstraction and modularity. However, there are some differences that make agents unique: first of all, their autonomy means also that they maintain a complete control over their actions, while objects’ remote method invocation does not allow the same level of control. In fact, between agents there occur requests for actions, instead of method invocations. As a second main difference, agent communication languages are independent from applications. As reported in [3], agents provide useful metaphors for describing artificial systems, such as: -

Open systems, which are dynamically changing because they are based on heterogeneous components, appearing, disappearing and changing behaviour;

-

Complex systems: the agent paradigm provides a way of abstracting and analysing them;

-

Systems with distributed data, control or resources, where solutions are given for their design and implementation;

-

Legacy systems, where solutions are given to enable interoperability of old software.

Several initiatives are devoted to the multi-agent paradigm. Among them, KSE (Knowledge Sharing Effort) [4], provides means for knowledge interchange, including a knowledge representation language (KIF, Knowledge Interchange Format), and a communication language (KQML, Knowledge Query and Manipulation Language), which are of great interest for agent-based development. Recently, a Consortium has been established, aimed at delivering a set of standards and recommendations for the development of agent-based applications (FIPA, Foundation for Intelligent Physical Agents) [5]. The KSE approach and FIPA represent up to now the most relevant approaches to multi-agent systems, but new initiatives are starting, including the MASIF proposal of the Object Management Group [6].

Target applications developed so far belong to the areas of telecommunications, electronic commerce, manufacturing, industrial process control, air traffic control, entertainment, and also healthcare [3].

Mobile agents add a new property to the ones previously described for generic agents: they can migrate from a system to another, and then operate on the target machine. There are several reasons to do so: distributing the computational load among several nodes; reducing the communication bandwidth; reducing the server load in a client/server environment; extending dynamically -and often temporarily- the functionalities of programs; customising the communication protocols. All these topics have been regarded as crucial for the development of mobile communications systems [7], including those derived from the recent wireless communication media, which are of great interest in mobile telemedicine.

Agent-based medical applications and telemedicine Telemedicine seems to be an artificial system suitable for the multi-agent systems: it is composed by complex systems with heterogeneous components, managing distributed data and resources, and often needing integration with legacy systems (such as hospital/laboratory information systems). Mobile telemedicine systems in particular may be considered as crucially based on continuously appearing and disappearing components, with distributed features. In the last years, some agent-based application has been developed in the medical field. The most important is perhaps the GUARDIAN system [8], devoted to patient monitoring in a Surgical Intensive Care Unit. In such a system, support is provided for collaboration among specialists, each expert in a specific domain but committed to share data and knowledge with the others, and nurses that continuously monitor the patient under the physician’s control. Lanzola et al. presented a system devoted to diabetes care, where cooperation among the different specialists, even outside the medical field (e.g. administration) is supported by an agency [9]. A KQML-like language has been adopted for inter-agent communication. Huang et al. presented an agent-based system for the collaboration among general practitioners and specialists about the patient healthcare [10]. We experimented a framework for telemedicine services through KQML-based Internet agents [11], where the agents are organised into federations, each one providing a particular service to other agents. The next section describes another approach to telemedicine agents, based on the FIPA specifications, which have been chosen because they are an emerging initiative aimed at proposing an industrial standard for agent-based applications. Although up to now medical applications are not yet considered among those described in the

specifications, it can be supposed that in the future also this field will be covered, thus leading to a interoperability framework of interest for telemedicine.

A telemedicine-oriented Medical Assistant A scenario A scenario has been analysed in order to identify some of the relevant issues for a telemedicine tool integrated into existing systems. The scenario regards a medical specialist (such as a dermatologist or a radiologist) dealing with routine visits, but having sometimes the need of consulting distant experts on difficult cases. Let's imagine the specialist during his/her routine work; he/she is going to examine patients having previously taken an appointment, which is reported on an agenda together with some data about the patient (coming from the electronic patient record). Sometimes, the case is difficult, and thus it will need a second opinion diagnosis from another expert specialist, available for giving such a service. It can also happen that the specialist will give a second opinion on somebody else's case. In both cases, it is necessary to have at disposal the relevant data, which means that an abstract of the clinical record with appropriate images (e.g., radiographs or skin photographs) have to be shared between the referring and consultant specialists. Sharing may occur in a synchronous way, when some videoconferencing-like tool is available, or asynchronously, in a way similar to usual mailing. In the former case, an appointment should be taken with the expert, in a date and time agreed upon by both parties. Finally, when the second opinion diagnosis is obtained, it should be inserted into the electronic patient record. In this scenario, some topic adequate for a multi-agent solution can be recognised. First, there is the need of interfacing legacy systems, which are the information systems of the hospital and of the laboratories. Establishing an appointment between two specialists is matter of negotiation and collaboration, and can be taken through a collaborative effort of the two user agents. The process of discussing a case at the same time is strictly related to the capability of communicating complex multimedia data with specialised protocols. New specialists may become available for consultation, and possible users should be made aware of that. Finally, new kinds of services may also appear, which need to be advertised and provided in some seamless way. The agent community Figure 1 shows a sketch of an agent community, which deals with the issues presented here above. It can be seen that the main component is the Telemedicine-Oriented Medical ASsistant (TOMAS), which is the agent used by each specialist. As a medical assistant, it has two generic features: an agenda for managing

appointments, and methods for access to patient records. Support to telemedicine is given by software features for remote exchange of patient data, co-operative annotation of cases and negotiation of appointments. An appointment can be entered directly by the specialist, or obtained by remote negotiation with another agent (e.g., the assistant of another specialist, a secretary’s agent, or a patient’s agent). If the appointment is a conventional one, the data related to the patient are retrieved from the appropriate information system, which is wrapped by an agent, and presented to the specialist at the specified time and date. If the appointment is a request for a second opinion, first of all a specialist of the needed field is chosen among those given by the specialists broker agency. Then, the appointment is negotiated with his/her agent, and finally patient data - together with the appropriate images (radiographs, photographs, microscope images...) - are sent to the remote specialist before the appointment, in order to have them ready when needed. Together with the data, a mobile agent is sent containing the ontology needed for managing them, along with the procedural knowledge for its graphical representation. In this way, each agent has its individual data interfaces, and it can temporarily pass part of them to the agent of the remote consultant. Data and images can be either studied off-line by the remote consultant, or collaboratively browsed through an interface which allows for cursor sharing and text annotation. Data resulting from such collaborative work may be then inserted automatically into the patient record. The DBAgent connected to the Hospital Information System (HIS), together with as many TOMAS instances as are the specialists inside the hospital (and even other agents, not considered here for the sake of simplicity), belong to the hospital agency. Following the FIPA recommendations, they are reachable from outside through the hospital communication broker, which acts as a broker of the internal expertise and capabilities. Further capabilities can be given to the user agent by accessing services provided by other agents: as an example, access to bibliographic information may be obtained through a Medline wrapper agent, which connects to the Internet to retrieve bibliographic references. The community of agents has been developed in Java using an already existing FIPA-compliant platform, FIPA_SMART 3.0 (FIPA-based Stationary and Mobile Agent Resource Toolkit, version 3.0) [12]. The basic feature of the agent, i.e., the agenda, has been implemented following the FIPA recommendations for the development of personal assistants. Preliminary results The prototype has been preliminarily tested between the Institute of Pathology and the Department of Informatics, University of Udine, Italy. The aim of the overall system was to support collaborative diagnosis in pathology, following the principles already studied in the field of static telepathology. Thus, the main activity was the

exchange of microscope images together with patient's data coming from a sample database. First tests regarded only the technical functioning of the system, and will be followed by thorough evaluations to be carried out inside the newly launched Italian Telepathology Network for Research, Education and Quality Control. Although at a preliminary stage, the prototype developed so far allowed us to explore the ways telemedical collaboration may occur using agents. In particular, agents let design the software in a direct correspondence to individuals or organisations being users of the system. Thus, there will be a set of medical assistant agents directly related to specialists, other agents providing specific services, agents providing single access points to entire organisations (such as hospitals), etc. All agents communicate through a common language, and this allows for further service extensions through a process of registration of new agents inside the hospital agency. A main technical pitfall have been recognised at this stage: the hospital communication broker, which is a required agent inside a FIPA platform, acts as a bottleneck in the system, when many exchanges with external agents are carried out. However, it is the same kind of problem occurring when local area networks are isolated through firewalls and proxies. Apart from this, there is also a generic overhead due to the agent messaging system: the use of a generic language allows for great flexibility but requires also greater resources. Further work is needed for evaluating, in a real environment, the behaviour of a society of intercommunicating agents that exchange multimedia data, such as those involved in telemedicine tasks. Furthermore, the security features considered in FIPA should be adopted in order to have a practically useful system. Discussion The multi-agent paradigm is somehow similar to an evolution of the object-oriented paradigm, being situated at a higher abstraction level, which allows to integrate both approaches, if needed. The features of agents are aimed at distributing the task of solving problems, by allowing different software components to cooperate, each one with its own expertise. Patient management has been so far the best experimented application of agents in healthcare, because of the inherently distributed nature of the expertise needed for that problem, as occurs in telemedicine also. Agents are not the panacea for all problems. In fact, they raise new security problems, which are even more crucial when using mobile agents. All these problems result in an increased need of control when approaching telemedicine, because of the involvement of sensible patient’s information. However, security is present among the main research lines in the agent field, and both FIPA specifications and KQML provide for security and privacy solutions, although we did not test them yet.

Health care involves several healthcare professionals, with different roles. The patient meets general practitioners, specialists of many branches, nurses, administrative employees, each one with his/her own expertise and with the need of accessing specific patient data, graded to their knowledge; such groups have continuous contacts. In our opinion, all such information exchanges can be supported by the purposive use of intercommunicating agents. This needs to take into account, but also to exploit as much as possible, the results of the other related research fields, i.e. electronic patient records and medical terminology and standards (HL7, GALEN). The use of FIPA recommendations allows to exploit the ongoing industrial efforts towards agent-based interoperability. However, adopting the multi-agent paradigm will involve a thorough analysis of the medical domain, in order to identify actors to be agentified, and relationships to be put forward as explicit communication acts.

References 1. D.C. Dennett, 1987, The intentional stance, Cambridge, USA, MIT Press, 1987. 2. M.R.Genesereth, S.P.Ketchpel, Software Agents, Communications of ACM 37, 48-53, 1994. 3. N.R.Jennings, M.J.Wooldridge, Agent Technology Foundations, Applications, and Markets, Berlin, Germany: Springer-Verlag, 1998. 4. R.S.Patil, R.E.Fikes, P.F.Patel-Schneider, D.McKay, T.Finin, T.Gruber, R.Neches, The DARPA Knowledge Sharing Effort: Progress Report, Proc. of Knowledge Representation and Reasoning, 777-788, 1992. 5. FIPA - Foundation for Intelligent Physical Agents, URL:http://www.fipa.org/, 1999. 6. D.Milojicic, M.Breugst, I.Busse, J.Campbell, S.Covaci, B.Friedman, K.Kosaka, D.Lange, K.Ono, M.Oshima, C.Tham, S.Virdhagriswaran, J.White, MASIF: The OMG Mobile Agent System Interoperability Facility, Proc. 2nd Int. Workshop on Mobile Agents, Lecture Notes in Computer Science, 1477, 50-67, 1998. 7. M.Breugst, L.Hagen, T.Magedanz, Impacts of Mobile Agent Technology on Mobile Communications System Evolution, IEEE Personal Communications Magazine, 5, 1998. 8. B.Hayes-Roth, J.E.Larsson, A domain-specific software architecture for a class of intelligent patient monitoring systems, Journal of Experimental and Theoretical Artificial Intelligence, 8, 149-171, 1996. 9. G.Lanzola, S.Falasconi, M.Stefanelli, Cooperative software agents for patient management, Lecture Notes in Artificial Intelligence, 934, 173-184, 1995. 10. I.Huang, N.R.Jennings, J.Fox, An agent-based approach to healthcare management, International Journal of Applied Artificial Intelligence, 9, 401-420, 1995. 11. V.Della Mea, V.Roberto, A.Conti, L.Di Gaspero, C.A.Beltrami, Internet Agents for Telemedicine Services, Medical Informatics, 24, 179-186, 1999. 12. R.Crepeau, FIPA_SMART v3.0, URL:http://dirac.nosc.mil:6996/FIPA_SMART/FipaSmartIntro.html, 1999.

TOMAS TOMAS TOMAS TOMAS Telemedicine-Oriented Medical ASsistant Telemedicine-Oriented Medical ASsistant Telemedicine-Oriented Medical ASsistant Telemedicine-Oriented Medical ASsistant

HIS

Hospital Agency

DBAGENT

Hospital Communication Broker

Specialists broker agency

Medline wrapper

TOMAS WWW access to Medline

Figure 1: Sketch of an agent community for telemedical collaboration.