Issue No.06 - November/December (2006 vol.21)
Published by the IEEE Computer Society
Antonio Moreno , University Rovira i Virgili
DOI Bookmark: http://doi.ieeecomputersociety.org/10.1109/MIS.2006.111
This article is part of a special issue on Intelligent Agents in Healthcare. Agent technology is currently being applied to solve different kinds of problems in the healthcare domain. This special issue presents six articles that discuss different applications, in areas as diverse as automatic e-health service discovery, use of argumentation techniques in organ transplants, and the implementation of agent-based mechanisms that ensure the confidentiality and privacy of sensitive medical data.
A few years ago, some researchers began to argue that agent technology could provide good tools for solving healthcare problems. 1 Researchers based this opinion on a straightforward matching between intelligent agents' properties and healthcare problems' characteristics. Nowadays, the evolution of information and communication technologies and agent technology strengthens that argument.
Researchers believe that agent technology is suitable for addressing healthcare problems for seven main reasons.
First of all, multiagent systems can successfully solve problems in which the required knowledge is physically distributed in several places (for example, they can gather patient data from different medical institutions or discover distributed e-health services). Some agents are even mobile, moving physically within an electronic network to obtain, analyze, filter, personalize, and present the data the user requires.
Second, agents can model autonomous entities (for example, different possible recipients of an organ transplant or different units of a hospital).
Third, agent-based systems can introduce security mechanisms that ensure the confidentiality and privacy of sensitive data, such as electronic medical records.
Fourth, you can usually divide healthcare's complex problems into (more or less independent) subproblems. Agent technology provides distributed problem-solving techniques in which agents communicate to coordinate their activities and collectively solve a problem that they couldn't tackle alone.
Fifth, all fields, including e-health, are seeing a steady increase in the number of electronic services available through the Internet. In recent years, agent researchers have developed mechanisms for automatically discovering and composing e-services.
Sixth, most healthcare problems require AI techniques (for example, machine learning, inference, or planning). Intelligent agents are built on top of the AI techniques, which are arguably the best tools to show a deliberative, reactive, flexible, and learning behavior.
Finally, agents' autonomous, reactive, and flexible character make them ideal for implementing ambient intelligence applications, in which computing is pervasive and embedded in the system. Hospitals would make an ideal setting for this kind of application because medical staff could access any kind of medical information at any point of the medical center (such as looking up a patient's health record bedside during an examination).
This argument, along with e-health's strategic importance to the European Union, 2 convinced the European AgentCities and AgentLink research networks to fund working groups on agents applied in healthcare. The working groups helped initiate regular meetings of (primarily European) research groups interested in agents' application in healthcare. The meetings took place in first-class forums such as the 15th European Conference on Artificial Intelligence (ECAI 2002) in France, ECAI 2004 in Spain, the 19th International Joint Conference on Artificial Intelligence (IJCAI 2005) in Scotland, and ECAI 2006 in Italy. You can find the works presented at these gatherings in their associated proceedings and in select publications. 3-5 They provide an excellent survey of the evolution of agent technology's uses in healthcare over the past five years.
In this issue
The six contributions in this special issue cover many of the aspects I've already discussed. Some are evolutions of work presented at the specialized workshops I mentioned earlier. Many resulted from EU-funded e-health research projects. This demonstrates the importance the EU is giving to e-health, 2 but it also shows that most of these experiences remain in the academic field and haven't yet been deployed for daily use in real clinical environments. Nonetheless, all the articles make sound, significant new contributions to agent technology (contributions that, in most cases, could also be used in other fields).
César Céceres, Alberto Fernández, Sascha Ossowski, and Matteo Vasirani study a basic problem of agent technology in open environments: how to automatically discover, invoke, and coordinate e-services (which was also the original AgentCities European network's focus). Their article, "Agent-Based Semantic Service Discovery for Healthcare: An Organizational Approach," discusses their main contribution—the design and development of a service matchmaker. Their approach's novelty resides in their use of role ontologies and interaction ontologies in the matchmaking process. The authors use a healthcare scenario related to managing medical emergencies to illustrate their technique, but it's certainly general enough to apply to other scenarios.
John Fox, David Glasspool, and Sanjay Modgil have many years of experience in building agent-based systems in the healthcare domain (for example, in the CREDO project and in developing the ProForma guideline representation language 6). In "A Canonical Agent Model for Healthcare Applications," they propose a general autonomous agent model that can also apply to nonhealthcare scenarios. Their agent architecture consists of 12 modules defined by software signatures that specify their input-output patterns. These components relate to classical AI tasks such as inference, problem solving, plan enactment, and learning. The authors use complex data types (such as plans, options, or arguments) as standard interfaces to facilitate communication between these independent modules. This work has been developed within the European IST project ASPIC.
The Carrel project has been working in recent years on developing an agent-mediated electronic institution that can better manage organ and tissue transplants in Spain. 7 In "Increasing Human-Organ Transplant Availability: Argumentation-Based Agent Deliberation," Pancho Tolchinsky, Ulises Cortés, Sanjay Modgil, Francisco Caballero, and Antonio López-Navidad describe an extension to this system, called Carrel+, which uses new agent-based argumentation techniques to decrease the number of organs rejected for transplantation. Even if hospital experts think an organ isn't viable (for example, because of the donor's smoking history), potential recipients can argue in favor of transplantation anyway (say, because the donor didn't have a particular smoking-related disease).
Tamás Kifor, László Varga, Javier Vázquez-Salceda, Sergio Álvarez, Steven Willmott, Simon Miles, and Luc Moreau based their work on Provenance, another EU-funded project. They developed a system for keeping track of a complex distributed process's events. One field they chose as an excellent testbed for their ideas was healthcare systems. In "Provenance in Agent-Mediated Healthcare Systems," the authors show how an agent-based healthcare system can appropriately document all the decisions that take place in a distributed setting so that they can be recovered later (such as for auditing purposes or to revise the aspects that were considered when a certain decision was made). The authors also ensure that their system maintains the privacy of medical data.
Security is also the main focus of "Secure Integration of Distributed Medical Data Using Mobile Agents" by Pedro Manuel Vieira-Marques, Sergi Robles, Jordi Cucurull, Ricardo João Cruz-Correia, Guillermo Navarro, and Ramon Martí. Previously, the authors developed MAID (Multiagent System for Integration of Data), an agent-based system that provides access to patient data that's scattered between different units of a hospital. 8 MAID has already been running for two years in a major Portuguese hospital. In this article, the authors propose a mobile-agent-based extension that retrieves pieces of a virtual electronic health record from diverse medical institutions. The article focuses on the security mechanisms that ensure the confidentiality, integrity, and privacy of patient records. Its main contribution to the healthcare field is the use of novel agent-based self-protection mechanisms.
Finally, security is a main concern of "Privacy-Aware Autonomous Agents for Pervasive Healthcare," by Monica Tentori, Jesus Favela, and Marcela D. Rodríguez. The authors have experience in implementing agent-based pervasive computing systems in hospitals, using their own agent framework, SALSA (the Simple Agent Library for SOPE Applications). 9 Their article describes their extensions to SALSA to build pervasive computing systems that ensure appropriate privacy measures in medical data management.
All these works show the vitality of the field of agents applied in healthcare. They also show that there is still much work to be done before agent technology is routinely used in healthcare, and that many open issues still must be addressed (such as medical practitioners' acceptance of the technology, end users' confidence and trust in agents that help manage their care, or researchers' development of commonly accepted standards).
I thank all the scientists that submitted to this special issue. Many of the submissions were high quality, but we couldn't accept all because of space restrictions. I warmly thank the invaluable international reviewers, who made careful reviews and re-reviews this summer—in most cases, during their holiday periods.
Antonio Moreno is a lecturer at the University Rovira i Virgili's Computer Science and Mathematics Department. He is the founder and head of the Multi-Agent System Group (GruSMA). His research interests include the application of agent technology to healthcare problems and ontology learning from the Web. He received his PhD in AI from UPC (Technical University of Catalonia). Contact him at Department d'Enginyeria Informàtica i Matemàtiques, Escola Tècnica Superior d'Enginyeria, Universitat Rovira i Virgili, Av.Paisos Catalans, 26, 43007-Tarragona, Spain; firstname.lastname@example.org.