Learning Cardiac Electrophysiology Digital Twins Through Agentic Discovery of Hybrid Structure

2026-06-16

AI Summary

Personalized cardiac electrophysiological modeling requires automatically discovering an appropriate hybrid model architecture for each patient, rather than merely optimizing parameters. Existing approaches rely on manual expert design and exhibit limited generalizability. This work proposes LEADS, a novel framework that introduces large language model (LLM) agents into this domain for the first time. LEADS integrates domain knowledge through a structured action space and employs a reasoning–acting loop to autonomously search for model architectures that are physically plausible, interpretable, and numerically stable, while parameters are optimized via gradient descent. Experiments demonstrate that LEADS outperforms both handcrafted models and other LLM-driven methods on both synthetic and real-world cardiac electrophysiology data, achieving automated, cross-patient model structure discovery.

Abstract

Building personalized cardiac electrophysiology (EP) digital twins requires identifying the appropriate model structure for each patient, not merely fitting parameters. Traditional methods rely on experts to manually prescribe hybrid physics-neural architectures, which requires deep domain expertise and does not transfer across patients. Recent works have applied large language models (LLMs) to generate or act as hybrid models. However, despite their promising generalization capacity, these LLM-based methods lack the structural priors needed for stable cardiac simulations. Hence, we propose LEADS, a framework that formulates cardiac EP domain knowledge as a structured action space and utilizes an LLM agent to discover hybrid models. The agent follows an iterative reasoning-and-action loop to select, combine, and refine hybrid models, whilst gradient descent handles parameter fitting. The proposed LEADS designs every candidate model towards physically grounded, interpretable, and numerically stable, while allowing open-ended architectural discovery. We validate LEADS on synthetic data with three ground-truth reaction models and on real cardiac EP data, demonstrating that it outperforms both human-designed hybrid models and other LLM-based hybrid modeling.

Problem

• cardiac electrophysiology
• digital twins
• hybrid modeling
• model structure discovery
• personalized modeling

Innovation

• digital twins
• cardiac electrophysiology
• hybrid modeling
• LLM agent
• agentic discovery