A deep learning-based hybrid computational approach to cardiac electrophysiology

Date

2023-9-11

Author

Kuloğlu, Ali Fatih

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

197
views

117
downloads

Cite This

Electrophysiological modeling of the heart has witnessed significant progress with the increase of available computational power. Realistic electrophysiology models often require the solution of highly nonlinear differential equation systems. The complex nature of the electrodynamic activity of a cell limits the applicability of simplistic numerical techniques and necessitates the utilization of more advanced and demanding techniques. Deep learning has emerged as a promising tool for predicting the solution of highly nonlinear problems and has shown tremendous success in differential equation-based phenomena of biological systems over recent years. In this work, a deep learning-based algorithm is proposed for the accurate and time-efficient solution of the electrophysiology problem of the heart. A deep learning-based model is developed for forecasting transmembrane voltage at the cellular level. For this purpose, the biophysically detailed ten Tusscher-Panfilov model is used for the generation of the training data and performance measurements. Training data are acquired by solving ten Tusscher-Panfilov model as an ordinary differential equation system. The resulting deep learning-based model incorporates the external stimulus information and past potential values while making predictions. An important novelty of this work is extending a model trained with ordinary differential equations to the realm of partial differential equations by associating the external stimuli with the conductivity term of the partial differential equation. This approach facilitates the application of more conventional partial differential equation solvers. Therefore, the classical way of solving partial differential equations is combined with deep learning in the proposed approach. This hybrid approach has successfully been applied to solve multiple problems and has been evaluated in different settings.

Subject Keywords

Deep learning, Cardiac electrophysiology, Finite element modeling

URI

https://hdl.handle.net/11511/105616

Collections

Graduate School of Natural and Applied Sciences, Thesis