Learning Cardiac Electrophysiology Digital Twins Through Agentic Discovery of Hybrid Structure

The article introduces LEADS, a framework for creating personalized cardiac electrophysiology (EP) digital twins through an LLM agent that discovers hybrid model structures. It employs a structured action space to iteratively select and refine models while using gradient descent for parameter fitting, resulting in models that are physically grounded, interpretable, and numerically stable. LEADS has been validated on both synthetic and real cardiac EP data, outperforming traditional human-designed models and existing LLM-based approaches, highlighting its potential for improving personalized medicine in cardiac care.