Agentic Brain: Agentic AI System for Autonomous Diagnosis of Functional Connectivity Network

Artificial intelligence (AI) has demonstrated high classification performance in the analysis of brain functional connectivity (FC) data; however, its clinical utility is fundamentally limited by the opaque “black-box” nature of its decisionmaking processes. To overcome this critical challenge, this study introduces the “Agentic Brain,” a novel multi-agent framework inspired by the brain’s own distributed processing architecture. The Agentic Brain is composed of deterministic “Analyst Agents,” providing both anatomical (Lobe) and functional (Network) perspectives through local connectivity analysis, and a “Central Executive Agent.” This agent, powered by deep reinforcement learning, integrates these distributed findings to autonomously form and verify diagnostic hypotheses through an iterative process of self-directed inquiry. Crucially, its policy is optimized via Reinforcement Learning (RL) to learn the most efficient and accurate diagnostic pathways from simulated experience. Experimental results validate this dual-objective approach, demonstrating that our framework not only achieves high diagnostic accuracy but also provides a transparent, step-by-step reasoning trace for each conclusion. The Agentic Brain offers a new path toward enhancing explainability (XAI) and clinical reliability, representing a significant step towards trustworthy AI partners in clinical decision support.