This research introduces a novel framework that integrates large language models with small specialized models to improve visual inspection of medical time series data.

• ConMIL framework combines the strengths of broad LLMs and specialized models for superior performance
• Utilizes conformal prediction to provide statistical guarantees on model outputs
• Enhances interpretability through model collaboration, improving trust in clinical settings
• Demonstrates effectiveness in critical applications including arrhythmia detection and sleep staging

This approach addresses a critical need in healthcare for AI systems that balance broad reasoning capabilities with specialized medical expertise, potentially accelerating clinical adoption and improving patient outcomes.

Enhancing Visual Inspection Capability of Multi-Modal Large Language Models on Medical Time Series with Supportive Conformalized and Interpretable Small Specialized Models