Abstract
Clinical decision support systems combine knowledge and data from a variety of sources, represented by quantitative models based on stochastic methods, or qualitative based rather on expert heuristics and deductive reasoning. At the same time, case-based reasoning (CBR) memorizes and returns the experience of solving similar problems. The cooperation of heterogeneous clinical knowledge bases is based on medical ontologies. A multi-agent decision support system (MADSS) enables the integration and cooperation of agents specialized in different fields of knowledge (semiology, pharmacology, clinical cases, etc.). Our approach is based on the specialization of agents adapted to the knowledge models used during the clinical steps and ontologies. This modular approach is suitable for the realization of MADSS in many areas.
Introduction
Clinical decision-making is inherently complex due to the fragmented nature of medical knowledge. Clinicians often synthesize statistical trends, expert heuristics, and analogies from past clinical cases. Most existing decision support software struggles to integrate these diverse reasoning styles. Furthermore, clinical work follows a specific temporal flow that must be respected for safety and efficacy. This paper addresses the challenge of creating a system that is both semantically interoperable and procedurally coherent, following the standard clinical démarche (diagnosis → prognosis → therapy).
Methodology and Architecture
The authors propose a hierarchical, supervised Multi-Agent System (MAS) architecture designed to mimic the natural clinical workflow.
The Supervisor Agent
The Supervisor Agent acts as the central controller, utilizing Finite State Automata (FSA) to plan and trigger tasks. It maintains a "working memory" of the current situation, including system time, problem context, and user data. By enforcing a structured progression, it prevents the system from recommending treatments before a diagnosis is established.
Specialized Agents
The system employs "worker" agents specialized by reasoning model and domain. For instance, a Rule-Based Agent might handle antibiotic guidelines, while a Case-Based Reasoning (CBR/RàPC) module stores and retrieves past clinical experiences. This modularity allows the system to be extended with new medical specialties by simply "plugging in" new agents.
Semantic Integration through Ontologies
To enable communication between heterogeneous agents, the system utilizes medical ontologies based on UMLS, MeSH, and the NCI thesaurus. These ontologies provide a standardized vocabulary and define semantic relationships between domain terms.
By using ontologies as a protocol rather than just a dictionary, the system allows a statistical model and a rule-based model to share data seamlessly. This semantic interoperability is crucial for maintaining consistency across the different stages of clinical decision-making.
Evaluation and Results
The framework was evaluated using three detailed clinical scenarios involving acute diarrhea. The system successfully mapped clinical signs to diagnostic categories and combined rule-based logic with case-based reasoning to refine treatment plans.
- Workflow Integration: The Supervisor Agent successfully managed the transition from symptoms to therapy.
- Reasoning Synergy: The system checked for contraindications while simultaneously comparing the patient to similar historical cases.
- Learning Capability: The CBR module demonstrated the ability to index new cases, allowing the knowledge base to grow organically with use.
Conclusion
The proposed MADSS offers a modular and scalable approach to medical decision support. While the prototype demonstrated successful integration of diverse knowledge sources, limitations remain. The user interface can become overloaded in complex cases, and the strict supervision required for clinical safety limits the autonomy of individual agents. Future work will focus on multi-user support and adapting the framework to other complex domains beyond medicine.
Summary updated