CardiacEP-PINOs

Repository for the work "Physics-Informed Neural Operators for Cardiac Electrophysiology".

This repository provides training and evaluation scripts for running a Physics-Informed Neural Operator (PINO) model for simulating cardiac electrophysiology propagation scenarios using the Aliev–Panfilov (AP) cardiac cell model.
It also includes the results of the experiments presented in the accompanying paper, including animations of model predictions.

---

Repository Overview

Data

Contains scripts for transforming raw simulation data into training, testing, and evaluation datasets, as well as example dataset folders for the planar, centrifugal, spiral, and spiral-break propagation scenarios.

Paper Results

Includes evaluation results for the experiments described in "Physics-Informed Neural Operators for Cardiac Electrophysiology".
Results cover baseline tests, mesh resolution experiments, and zero-shot transfer evaluations.
Each results folder includes side-by-side animations comparing model predictions with ground truth simulations.

Scripts

• PINO_Train.py
  Training script for the PINO model. Supports different model configurations and training parameters.
  Trained models are saved in timestamped results folders.

• Evaluation_P2P.py
  Evaluates the trained model on a point-to-point basis, where ground truth inputs are used for each prediction.

• Evaluation_Rollout.py
  Evaluates the trained model in a recursive (rollout) fashion, using previous model outputs as inputs for subsequent predictions.

• Model_Comparison.py
  Compares the performance of multiple trained models on the same evaluation dataset.

---

Usage

1. Install Dependencies
   Install the neuraloperator library following the instructions provided here:
   https://github.com/neuraloperator/neuraloperator

2. Run Training
   To train the PINO model on one of the provided datasets, run the training script from the command line:

   python PINO_Train.py -d <dataset_path> -<additional arguments>

   To inspect optional arguments, run:

   python PINO_Train.py --help

---

Citation

If you use this repository in your work, please cite:

"Physics-Informed Neural Operators for Cardiac Electrophysiology"
(https://arxiv.org/abs/2511.08418)

Additionally, this repository makes use of the Neural Operator library. Please cite:

"A Library for Learning Neural Operators." Kossaifi, J., Kovachki, N., Li, Z., Pitt, D., Liu-Schiaffini, M., Duruisseaux, V., George, R., Bonev, B., Azizzadenesheli, K., Berner, J., and Anandkumar, A. arXiv, 2025. doi:10.48550/arXiv.2412.10354

---

Contact

For questions or collaboration inquiries, please contact:
[Hannah Lydon] – [hannah.lydon@kcl.ac.uk]