Physics-Informed Neural Networks for Cardiac Activation Mapping
A critical procedure in diagnosing atrial fibrillation is the creation of electro-anatomic activation maps. Current methods generate these mappings from interpolation using a few sparse data points recorded inside the atria; they neither include prior knowledge of the underlying physics nor uncertai...
Main Authors: | , , , , |
---|---|
Format: | Article |
Language: | English |
Published: |
Frontiers Media S.A.
2020-02-01
|
Series: | Frontiers in Physics |
Subjects: | |
Online Access: | https://www.frontiersin.org/article/10.3389/fphy.2020.00042/full |
id |
doaj-f96249a175624e17bd50ccb5ae280d8d |
---|---|
record_format |
Article |
spelling |
doaj-f96249a175624e17bd50ccb5ae280d8d2020-11-25T02:42:45ZengFrontiers Media S.A.Frontiers in Physics2296-424X2020-02-01810.3389/fphy.2020.00042510944Physics-Informed Neural Networks for Cardiac Activation MappingFrancisco Sahli Costabal0Francisco Sahli Costabal1Francisco Sahli Costabal2Yibo Yang3Paris Perdikaris4Daniel E. Hurtado5Daniel E. Hurtado6Daniel E. Hurtado7Ellen Kuhl8Department of Structural and Geotechnical Engineering, School of Engineering, Pontificia Universidad Católica de Chile, Santiago, ChileInstitute for Biological and Medical Engineering, Schools of Engineering, Medicine and Biological Sciences, Pontificia Universidad Católica de Chile, Santiago, ChileMillennium Nucleus for Cardiovascular Magnetic Resonance, Santiago, ChileDepartment of Mechanical Engineering and Applied Mechanics, University of Pennsylvania, Philadelphia, PA, United StatesDepartment of Mechanical Engineering and Applied Mechanics, University of Pennsylvania, Philadelphia, PA, United StatesDepartment of Structural and Geotechnical Engineering, School of Engineering, Pontificia Universidad Católica de Chile, Santiago, ChileInstitute for Biological and Medical Engineering, Schools of Engineering, Medicine and Biological Sciences, Pontificia Universidad Católica de Chile, Santiago, ChileMillennium Nucleus for Cardiovascular Magnetic Resonance, Santiago, ChileDepartment of Mechanical Engineering and Bioengineering, Stanford University, Stanford, CA, United StatesA critical procedure in diagnosing atrial fibrillation is the creation of electro-anatomic activation maps. Current methods generate these mappings from interpolation using a few sparse data points recorded inside the atria; they neither include prior knowledge of the underlying physics nor uncertainty of these recordings. Here we propose a physics-informed neural network for cardiac activation mapping that accounts for the underlying wave propagation dynamics and we quantify the epistemic uncertainty associated with these predictions. These uncertainty estimates not only allow us to quantify the predictive error of the neural network, but also help to reduce it by judiciously selecting new informative measurement locations via active learning. We illustrate the potential of our approach using a synthetic benchmark problem and a personalized electrophysiology model of the left atrium. We show that our new method outperforms linear interpolation and Gaussian process regression for the benchmark problem and linear interpolation at clinical densities for the left atrium. In both cases, the active learning algorithm achieves lower error levels than random allocation. Our findings open the door toward physics-based electro-anatomic mapping with the ultimate goals to reduce procedural time and improve diagnostic predictability for patients affected by atrial fibrillation. Open source code is available at https://github.com/fsahli/EikonalNet.https://www.frontiersin.org/article/10.3389/fphy.2020.00042/fullmachine learningcardiac electrophysiologyEikonal equationelectro-anatomic mappingatrial fibrillationphysics-informed neural networks |
collection |
DOAJ |
language |
English |
format |
Article |
sources |
DOAJ |
author |
Francisco Sahli Costabal Francisco Sahli Costabal Francisco Sahli Costabal Yibo Yang Paris Perdikaris Daniel E. Hurtado Daniel E. Hurtado Daniel E. Hurtado Ellen Kuhl |
spellingShingle |
Francisco Sahli Costabal Francisco Sahli Costabal Francisco Sahli Costabal Yibo Yang Paris Perdikaris Daniel E. Hurtado Daniel E. Hurtado Daniel E. Hurtado Ellen Kuhl Physics-Informed Neural Networks for Cardiac Activation Mapping Frontiers in Physics machine learning cardiac electrophysiology Eikonal equation electro-anatomic mapping atrial fibrillation physics-informed neural networks |
author_facet |
Francisco Sahli Costabal Francisco Sahli Costabal Francisco Sahli Costabal Yibo Yang Paris Perdikaris Daniel E. Hurtado Daniel E. Hurtado Daniel E. Hurtado Ellen Kuhl |
author_sort |
Francisco Sahli Costabal |
title |
Physics-Informed Neural Networks for Cardiac Activation Mapping |
title_short |
Physics-Informed Neural Networks for Cardiac Activation Mapping |
title_full |
Physics-Informed Neural Networks for Cardiac Activation Mapping |
title_fullStr |
Physics-Informed Neural Networks for Cardiac Activation Mapping |
title_full_unstemmed |
Physics-Informed Neural Networks for Cardiac Activation Mapping |
title_sort |
physics-informed neural networks for cardiac activation mapping |
publisher |
Frontiers Media S.A. |
series |
Frontiers in Physics |
issn |
2296-424X |
publishDate |
2020-02-01 |
description |
A critical procedure in diagnosing atrial fibrillation is the creation of electro-anatomic activation maps. Current methods generate these mappings from interpolation using a few sparse data points recorded inside the atria; they neither include prior knowledge of the underlying physics nor uncertainty of these recordings. Here we propose a physics-informed neural network for cardiac activation mapping that accounts for the underlying wave propagation dynamics and we quantify the epistemic uncertainty associated with these predictions. These uncertainty estimates not only allow us to quantify the predictive error of the neural network, but also help to reduce it by judiciously selecting new informative measurement locations via active learning. We illustrate the potential of our approach using a synthetic benchmark problem and a personalized electrophysiology model of the left atrium. We show that our new method outperforms linear interpolation and Gaussian process regression for the benchmark problem and linear interpolation at clinical densities for the left atrium. In both cases, the active learning algorithm achieves lower error levels than random allocation. Our findings open the door toward physics-based electro-anatomic mapping with the ultimate goals to reduce procedural time and improve diagnostic predictability for patients affected by atrial fibrillation. Open source code is available at https://github.com/fsahli/EikonalNet. |
topic |
machine learning cardiac electrophysiology Eikonal equation electro-anatomic mapping atrial fibrillation physics-informed neural networks |
url |
https://www.frontiersin.org/article/10.3389/fphy.2020.00042/full |
work_keys_str_mv |
AT franciscosahlicostabal physicsinformedneuralnetworksforcardiacactivationmapping AT franciscosahlicostabal physicsinformedneuralnetworksforcardiacactivationmapping AT franciscosahlicostabal physicsinformedneuralnetworksforcardiacactivationmapping AT yiboyang physicsinformedneuralnetworksforcardiacactivationmapping AT parisperdikaris physicsinformedneuralnetworksforcardiacactivationmapping AT danielehurtado physicsinformedneuralnetworksforcardiacactivationmapping AT danielehurtado physicsinformedneuralnetworksforcardiacactivationmapping AT danielehurtado physicsinformedneuralnetworksforcardiacactivationmapping AT ellenkuhl physicsinformedneuralnetworksforcardiacactivationmapping |
_version_ |
1724771704154095616 |