Computer Vision News - September 2022

51 Cardiac Electrophysiological Imaging To examine the performance of the model, they first mathematically calculate the mean squared error for the different time horizons by feeding the model with only one initial measurement. They find that, while the model struggles to learn the phenomena with the pure physical and data-driven models, the APHYN-EP model captures themwell at different time horizons. The physical model created the transmembrane potentials of the cell, and the data-driven model generated the transmembrane potentials and evaluated them. The model is very sensitive to self-generated noise and performs poorly on the temporal dynamics. These results shown in the following figure prove that the APHYN-EP can capture the first 9ms of cardiac dynamics in a high-precision manner. They also show that the correction term brought by Fa (posteriorly pointed line) is clearly visible. As a conclusion and summary, in the paper a surrogate model was used on APHYN-EP to model cardiac electrophysiology .  It does prove able to reproduce the dynamics of Ten Tusscher-Noble-Noble-Panfilov model, even using a simplified electrophysiology model as a physical component of the framework. It seems that the framework shows promise for giving physics-knowledge, specifically introducing prior knowledge in deep learning approaches, such as explicit equations, as well as correcting model errors from data, which is a current field of research. It only remains to see it appliedonmore challenging settings andvirtual 3Dheart simulations! Exciting news for the future! Next month Thank you for reading the article this month and accept my apologies for not continuing the review section. I do hope that you got something out of the paper selection and for this month but also the next coding sessions keep on sending and talking about it and don’t hesitate to send any corrections, suggestions, or ideas for next month’s issue! Take care and as always have a great time and always be curious! 