CARREL: The Medical large scale demonstrator
Carrel: The Medical Large Scale Demonstrator
The objectives intended for the ASPIC Medical Large scale demonstrator were to introduce argumentation processes in a scenario as complex and sensitive as human organ transplantation. The starting point was the CARREL system [1], an agent-based system intended to support the management of information related to organ distribution and allocation processes, and so increase the efficiency, safety, rapidity, and quality of organ distribution for transplantation, thereby helping the allocation process. In collaboration with transplant professionals at the Sant Pau Hospital in Barcelona we proposed a scenario application that was both potentially useful for the transplant domain and theoretically challenging from the Artificial Intelligence point of view.
The chosen scenario proposed changes in the current policies of human organ selection and allocation. These changes would help to reduce the number of organs that are discarded despite being available for transplantation. In particular, the current human organ selection process is defined such that the assessment over the viability of an organ is made exclusively by professionals located at the donor site. Thus, if these professionals believe the organ to be non-viable it is discarded, i.e. never offered for transplantation. However, it may be the case that professionals located at a potential recipient site may deem the organ as viable for their patient, and given the opportunity may be able to successfully defend their position. Thus the proposed scenario involves coordinating joint deliberation between agents at the donor site with agents at the recipient site in order to decide whether the available organ is viable.
The proposed scenario was extensively explored and validated in many publications, both in the artificial intelligence (e.g. [2-8]) and transplantation domain (e.g. [9,10]).
At the same time we developed prototypes to test the scenario and its argumentation aspects. The success of the first small scale prototype, built in COGENT[11], lead to the development of the second prototype, also in COGENT and the canonical agents’ work in [12]. This time the intention was that this will be the medical large scale demonstrator. Although promising progress was initially made, as soon as the scale of the agents’ reasoning grew, the performance of the prototype became very poor, making it impossible to demonstrate the ASPIC components’ performance. This led to a third version of the medical large scale demonstrator, this time built in java, with JADE [13] agents. Thus agents interact in a JADE platform. This work became the ASPIC medical large scale demonstrator, that is, CARREL.
Figure 1 Architecture of the Medical Large Scale Demonstrator
CARREL makes use of both the ASPIC Inference component and the Dialogue component. The former gave CARREL agents the possibility to adapt to changing and conflicting knowledge by constructing arguments for and against the agents’ queries. Another important feature is the possibility to present to a human user the reasoning behind a decision in the form of an argument, especially interesting when users must take decisions in safety critical domains as the transplant domain.
Figure 2. Visualization of an agent reasoning. (why hepatitis C on the donor is not a contraindication for transplanting a lung on a recipient that already has hepatitis C)
Figure 3. CARREL’s GUI for the professionals at the site of the potential recipient
While the inference component was relatively easy to make use of, the requirements given by the transplant scenario made the use of a dialogue component more challenging. The dialogue component allows agents to argue over whether an organ is viable or not. The component keeps track of all the made moves, providing at each time the legal moves to each participant. It allows flexibility in the participants turn taking. That is, agents can submit more than one move per turn. More over, if a participant agent loses connection with the dialogue and it’s interlocutor made several moves, as soon as this agent re-establishes communication, it can retrieve all the submitted moves made in its absence. Finally, agents can backtrack in the dialogue, which as discussed earlier in the description of the ASPIC dialogue model and component, is especially important in safety critical domains.
References:
[1] J. Vázquez-Salceda et al., “The Organ Allocation Process: A Natural Extension of the Carrel Agent-Mediated Electronic Institution,” AI Comm., vol. 16, no. 3, 2003, pp. 153–165.
[2] Pancho Tolchinsky, Ulises Cortés, Sanjay Modgil, Francisco Caballero and Antonio López-Navidad. Increasing Human-Organ Transplant Availability: Argumentation-Based Agent Deliberation. In IEEE Intelligent Systems Journal: Special Issue on Intelligent Agents in Healthcare, vol. 21, no. 6, pp. 30-37, Nov/Dec, 2006.
[3] Pancho Tolchinsky, Sanjay Modgil, Ulises Cortés and Miquel Sŕnchez-Marrč. Using CARREL+ to Increase Availability of Human Organs for Transplantation, in IWWAN 2007, LNCS, pages 1082-1091.
[4] Pancho Tolchinsky, Katie Atkinson, Peter McBurney, Sanjay Modgil and Ulises Cortés. Agents Deliberating Over Action Proposals Using the ProCLIAM Model. Accepted in The 5th International Central and Eastern European Conference on Multi-Agent Systems (CEEMAS07); September 2007, Leipzig, Germany.
[5] Pancho Tolchinsky and Ulises Cortés. Arguing Agents for Fairness in the Allocation of Human Organs for Transplantation. In 4th Workshop on Agents Applied in Health Care (ECAI-06), 2006.
[6] Pancho Tolchinsky, Sanjay Modgil and Ulises Cortés. Argument schemes and critical questions for heterogeneous agents to argue over the viability of a human organ. AAAI 2006 Spring Symposium Series; Argumentation for Consumers of Healthcare. Stanford, USA. March 2006.
[7] Sanjay Modgil, Pancho Tolchinsky and Ulises Cortés. Towards formalising agent argumentation over the viability of human organs for transplantation. In Mexican International Conference on Artificial Intelligence (MICAI), pages 928–938, 2005.
[8] Pancho Tolchinsky, Ulises Cortés, Juan Carlos Nieves, Antonio López-Navidad, and Francisco Caballero. Using arguing agents to increase the human organ pool for transplantation. In 3rd Workshop on Agents Applied in Health Care (IJCAI-05), 2005.
[9] P. Tolchinsky, U Cortés, F. Caballero and A. López-Navidad. A Novel Organ Selection Procedure that Uses Artificial Intelligence to Increase the Organ Pool for Transplantation. In XI Congreso Societat Catalana de Trasplantament. Barcelona 2007.
[10] P. Tolchinsky, U Cortés, F. Caballero and A. López-Navidad. CARREL+, Intelligent Electronic Support for Distribution and Allocation of Organ Transplantation. XXII Reunión Nacional de Coordinadores de Trasplantes,May 2007, Mallorca, Spain.
[11] R. Cooper, Modeling High-Level Cognitive Process, Lawrence Erlbaum Associates, 2002.
[12] John Fox, David Glasspool, Sanjay Modgil, "A Canonical Agent Model for Healthcare Applications," IEEE Intelligent Systems, vol. 21, no. 6, pp. 21-28, Nov/Dec, 2006
[13] http://jade.tilab.com/
