Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

HyPer-EP: Meta-Learning Hybrid Personalized Models for Cardiac Electrophysiology

Xiajun Jiang, Sumeet Vadhavkar, Yubo Ye, Maryam Toloubidokhti, Ryan Missel, Linwei Wang

TL;DR

HyPer-EP addresses the challenge of personalizing cardiac electrophysiology models by combining a physics-based expression with a neural correction to capture model gaps, enabled by a meta-learning strategy that separately identifies the physics parameters $\boldsymbol{\theta}$ and the neural components $\boldsymbol{\phi}$. The framework formalizes a hybrid model $\mathcal{M}_{\text{Hybrid}} = \mathcal{M}_{\text{PHY}} + \mathcal{M}_{\phi}$ and demonstrates two instantiations: Instantiation 1 uses an Eikonal PDE for rapid wavefront propagation augmented by a spatial-temporal graph neural network to recover full action potentials, while Instantiation 2 adopts a UDE formulation with a partial physics term and a neural correction. Proof-of-concept experiments on synthetic data (Aliev-Panfilov-based) show HyPer-EP can identify parameters with high accuracy and reconstruct spatiotemporal EP dynamics more accurately than pure physics or pure neural baselines, suggesting improved personalization and potential clinical utility. The work provides a scalable, interpretable hybrid paradigm for digital twin cardiac models with efficient test-time personalization and a pathway toward real-data validation.

Abstract

Personalized virtual heart models have demonstrated increasing potential for clinical use, although the estimation of their parameters given patient-specific data remain a challenge. Traditional physics-based modeling approaches are computationally costly and often neglect the inherent structural errors in these models due to model simplifications and assumptions. Modern deep learning approaches, on the other hand, rely heavily on data supervision and lacks interpretability. In this paper, we present a novel hybrid modeling framework to describe a personalized cardiac digital twin as a combination of a physics-based known expression augmented by neural network modeling of its unknown gap to reality. We then present a novel meta-learning framework to enable the separate identification of both the physics-based and neural components in the hybrid model. We demonstrate the feasibility and generality of this hybrid modeling framework with two examples of instantiations and their proof-of-concept in synthetic experiments.

HyPer-EP: Meta-Learning Hybrid Personalized Models for Cardiac Electrophysiology

TL;DR

HyPer-EP addresses the challenge of personalizing cardiac electrophysiology models by combining a physics-based expression with a neural correction to capture model gaps, enabled by a meta-learning strategy that separately identifies the physics parameters and the neural components . The framework formalizes a hybrid model and demonstrates two instantiations: Instantiation 1 uses an Eikonal PDE for rapid wavefront propagation augmented by a spatial-temporal graph neural network to recover full action potentials, while Instantiation 2 adopts a UDE formulation with a partial physics term and a neural correction. Proof-of-concept experiments on synthetic data (Aliev-Panfilov-based) show HyPer-EP can identify parameters with high accuracy and reconstruct spatiotemporal EP dynamics more accurately than pure physics or pure neural baselines, suggesting improved personalization and potential clinical utility. The work provides a scalable, interpretable hybrid paradigm for digital twin cardiac models with efficient test-time personalization and a pathway toward real-data validation.

Abstract

Personalized virtual heart models have demonstrated increasing potential for clinical use, although the estimation of their parameters given patient-specific data remain a challenge. Traditional physics-based modeling approaches are computationally costly and often neglect the inherent structural errors in these models due to model simplifications and assumptions. Modern deep learning approaches, on the other hand, rely heavily on data supervision and lacks interpretability. In this paper, we present a novel hybrid modeling framework to describe a personalized cardiac digital twin as a combination of a physics-based known expression augmented by neural network modeling of its unknown gap to reality. We then present a novel meta-learning framework to enable the separate identification of both the physics-based and neural components in the hybrid model. We demonstrate the feasibility and generality of this hybrid modeling framework with two examples of instantiations and their proof-of-concept in synthetic experiments.
Paper Structure (12 sections, 11 equations, 3 figures)

This paper contains 12 sections, 11 equations, 3 figures.

Table of Contents

  1. Introduction
  2. Problem Formulation
  3. Methodology
  4. Hybrid Modeling of Cardiac EP Process
  5. General HyPer formulation:
  6. Instantiation 1 -- HyPer to bridge simple and data-generating physics:
  7. Instantiation 2 -- HyPer as a universal differential equation (UDE):
  8. Learning to Identify
  9. Experiments and Results
  10. Results on Instantiation 1:
  11. Results on Instantiation 2:
  12. Conclusion

Figures (3)

  • Figure 1: Visual examples of the action potential obtained from $\mathcal{M}_{\text{PHY}}$ versus $\mathcal{M}_{\text{Hybrid}}$ in comparison to the truth, along with the true and estimated values for parameter $a$.
  • Figure 2: Summary of signal reconstruction performance on $\mathcal{M}_{\text{Hybrid}}$, $\mathcal{M}_{\text{NN}}$, and $\mathcal{M}_{\text{PHY}}$.
  • Figure 3: Visual examples of the action potential obtained from $\mathcal{M}_{\text{Hybrid}}$ versus $\mathcal{M}_{\text{NN}}$ and $\mathcal{M}_{\text{PHY}}$.