Heartfelt congratulations to Felix Meister and team for winning the best oral presentation award at the STACOM 2020 workshop, a satellite event of MICCAI 2020. The award was given for his work on “Graph convolutional regression of cardiac depolarization from sparse endocardial maps”.
In this paper, we show how geometric deep learning can be used to extrapolate electrophysiology signals throughout the myocardium given sparse and noisy endocardial measurements. A big thank you to our collaborators from JHU (Baltimore, MD), Dr. Ashikaga and Dr. Halperin, as well as the Pattern Recognition Lab at FAU (Erlangen, Germany). Congratulations to the entire team!
Buliard et al., Intra-procedural evaluation of a computational modelling method for cardiac resynchronization therapy, EP Europace, Volume 22, Issue 4, April 2020, Page 656, https://doi.org/10.1093/europace/euz239
With the team we developed a way to measure response to cardiac resynchronization therapy (CRT). CRT is a common tool used in cardiac electrophysiology (CEP), but it relies on perfect placement of the left ventricular (LV) lead. Our research team has been working on a project that evaluates how simulations can be incorporated into CEP for better lead implantation. Our case report, published in EP Europace, explores the guidance of LV lead implantation using intra-procedural CEP simulations.
To do this, we first made a computational model of CEP based on each individual patient’s cardiac MRI and ECG about one day before the CRT. During the procedure, we implanted a CRT system with a quadripolar LV lead. Using each patient’s individual data, we updated the model. We were able to sense ventricular delays and epicardial activation. We changed LV lead positions and AV delays to evaluate 6 different CRT protocols. We then simulated CRT.
We found that there was excellent agreement between the actual QRS duration and the QRS duration we had predicted. These results show that it could be possible to use computational simulations of CEP intraprocedurally to guide CRT!
Designing an inverse modeling approach to estimate the parameters of a computational model of cardiac function can be challenging, in particular when available clinical data is sparse and noisy. Instead, we propose to train an artificial agent exactly like a new user is trained: first learn how the computational model behaves by random and playful experiments. Then let it devise the optimal strategy for parameter identification through reinforcement learning. Applied on several problems, the agent could learn its new task and perform it much more efficiently and accurately than available inverse modeling strategies!