Computer Vision News - September 2022

48 AI Research paper Hello again! Some news to share for all, this is possibly the last paper review I will do for the magazine, at least for some time. I will be still nevertheless sharing code with you and with that, I am waiting for all your input. Last month I got a great communication from Jeff B. who offered an alternative solution to mine (and which I am going to include in next month’s article). This paper is titled “ Deep Learning for Model Correction in Cardiac Electrophysiological Imaging ” and was accepted in this year’s Medical Imaging with Deep Learning conference ( MIDL 2022 ) written by Victoriya Kashtanova et al at Inria . I would like to thank the authors for sharing comments about the article and offering permission to use them in this review Your kindness is always helping achieve better insights. Introduction This paper introduces a framework of mathematical modelling of cardiac electrophysiology which can help the analysis of this physiology and give greater insights. Specifically, the aim here is to model the dynamics of cardiac electrophysiology at lower cost. Low-fidelity physical model and a learning component are implemented here via neural networks. The model acts as a complement to the physical part and handles all quantities and dynamics that the simplified physical model neglects. Modelling of cardiac physiology Mathematical modelling of the heart is an active research area that is now increasingly coupled with artificial intelligence approaches. These models can accurately reproduce electrical behavior of cardiac cells. The Ten Tusscher-Panfilov model is often used to describe the details of transmembrane voltage, current, and other ionic concentrations when modelling the cardiac cell and is computationally expensive and has many hidden variables. Another type of model is the phenomenological model which has fewer variables and more parameters making it more useful for rapid modelling of wave propagation. Machine learning and deep learning approaches could help provide a correction mechanism to the model. The combination of rapid phenomenological models and machine learning could facilitate developing rapid and accurate models of transmembrane dynamics. CARDIAC ELECTROPHYSIOLOGICAL IMAGING