Multi-Agent Clinical Diagnosis

Overview

Multi-agent clinical systems simulate Multi-Disciplinary Team (MDT) discussions, where specialist agents collaborate on complex cases. Research shows these systems outperform single models, particularly for rare diseases and complex diagnoses.

Architecture (MAC Framework)

Patient Data → Intake Agent → Structured Case
                    ↓
        ┌───────────┴───────────┐
        ↓           ↓           ↓
  Cardiologist  Neurologist  Oncologist
     Agent        Agent        Agent
        └───────────┬───────────┘
                    ↓
             Supervisor Agent → Diagnosis Consensus
                    ↓
           Treatment Planning Agent
                    ↓
             Patient Communication

Agent Roles

Intake Agent

• Collects patient history
• Structures symptoms and findings
• Identifies relevant specialists

Specialist Doctor Agents

• Apply domain expertise
• Provide specialty-specific interpretations
• Suggest relevant tests

Diagnostic Agent

• Synthesizes specialist inputs
• Generates differential diagnoses
• Ranks by probability

Supervisor Agent

• Moderates specialist discussion
• Resolves disagreements
• Ensures comprehensive consideration

Treatment Planning Agent

• Develops treatment recommendations
• Considers patient factors
• Balances efficacy and side effects

Research Results

MAC Framework: Using 302 rare disease cases, Multi-Agent Conversation outperformed single models in both primary and follow-up consultations, achieving higher accuracy in diagnoses and suggested tests.

Optimal Configuration: Four doctor agents plus one supervisor agent using GPT-4 achieved best performance.

Performance Comparison: MAC outperformed Chain of Thoughts, Self-Refine, and Self-Consistency methods.

Upcoming Research (2025-2026)

• LungNoduleAgent (AAAI 2026): Collaborative multi-agent system for lung nodule diagnosis
• MedAgentSim: Self-evolving multi-agent simulations for clinical interactions
• MDTeamGPT: Self-evolving framework for multi-disciplinary consultation

Critical Considerations

Concerns include:

• Ensuring data quality and mitigating bias
• Integrating into existing clinical workflows
• Navigating ethical considerations for autonomous medical systems
• Accountability when decisions are distributed across agents