Multi-Agent AI System Improves Safe, Explainable Fracture Diagnosis

This research introduces FRAC-MAS, an innovative multi-agent AI system designed to enhance the safety and explainability of bone fracture diagnosis. While deep learning models excel in diagnostic performance, their "black-box" nature often hinders clinical adoption. FRAC-MAS addresses this by integrating a stacked ensemble of four vision models with conformal prediction, providing statistically grounded differential diagnoses. The system employs a multi-agent workflow where independent agents perform verification, retrieve clinical guidelines, and generate patient-friendly reports. This pipeline effectively triages 86.6% of cases into a high-confidence auto-confirmed cohort, while escalating uncertain cases for human review, significantly outperforming single-agent baselines. Patient preference studies also confirm that FRAC-MAS produces more comprehensible clinical reports than other LLMs. This framework demonstrates how cooperative agentic architectures can serve as auditable, human-in-the-loop decision support systems in safety-critical healthcare.