Faster than real time predictions of cardiac excitation patterns can open up new ways to warn about impending formation of arrhythmias and their prevention. Recently, we have introduced a novel method to create data-driven models for cardiac electrophysiology from 2D spatio-temporal recordings such as optical voltage mapping data. The models obtained from this fully automatic model creation pipeline are encoded as simple polynomials enabling computational simulations of cardiac tissue at high speed.
Directly utilising hardware capabilities and using optimised code, a cardiac emulator using our data-driven low-order predictive models can predict excitation waves in cardiac monolayers faster than real time. We propose a decision making algorithm managing the cardiac emulator in four phases: fitting the model, predicting, warning and preventing cardiac arrhythmias. Such a non-static model can be used to learn properties specific to each monolayer on the fly to then yield more relevant predictions, even if such properties change over the course of the lifetime of a tissue sample.
Extensions of this cardiac emulator to full heart geometry could be the next step towards a true personalised digital twin of the heart, holding immense potential to diagnose and treat cardiac rhythm disorders, bridging the gap between cardiac simulation of the heart and clinical practice.