Reconstruction of hidden heart behavior possible through machine learning

Heart disease is the leading cause of death worldwide. Though cardiac arrhythmias account for most of the fatalities, it is impossible to directly detect the electrical excitation waves within the heart muscle that coordinate — or in the case of arrhythmia, fail to properly coordinate — the heart’s mechanical contractions. Being able to deduce these excitation waves would provide deeper insight into the causes, behavior, and treatment of cardiac arrhythmia.

Stenger et al. demonstrated the feasibility of reconstructing electrical excitation from surface data. Using a simplified model of chaotic dynamics, the authors measured the ability of different machine learning programs to reproduce the model’s interior.

When the spatio-temporal chaotic dynamics in excitation waves prevent effective pumping, cardiac arrhythmia can be fatal. A 3D Barkley model replicated a simplified version of these electrical dynamics, and three neural networks were trained with a varying number of input time steps. Different approaches proved favorable depending on the amount of required input data, algorithm training time, reconstruction time, and desired depth of information.

“The main message, assuming our abstractions, is that it is possible to reconstruct the dynamics inside the myocardium from surface data using different types of artificial neural networks,” said author Sebastian Herzog.

While the simplified model is only an idealized reproduction of the heart, this study demonstrates the possibility of using measurable data to investigate cardiac arrhythmia. The application of these techniques could also extend to other aspects of medicine.

“Artificial neural networks allow the analysis of large amounts of data; we hope that studies like ours will form the basis for personalized medicine in the future,” said Herzog.

Source: “Reconstructing in-depth activity for chaotic 3D spatio temporal excitable media models based on surface data,” by R. Stenger, S. Herzog, I. Kottlarz, B. Rüchardt, S. Luther, F. Wörgötter, and U. Parlitz, Chaos (2023). The article can be accessed at https://doi.org/10.1063/5.0126824 .