Ex-vivo - to - In-vivo Learning in Cardiology

Alexander M. Zolotarev, Oleg Y. Rogov, Aleksei V. Mikhailov, John D. Hummel, Vadim V Fedorov, Dmitry V. Dylov

Keywords: Atrial fibrillation, Image-to-image translation, CycleGAN, Cardiac imaging

TL;DR: Training on explanted hearts improves the visualization of Atrial Fibrillation sources in clinic

Abstract: The clinical Atrial Fibrillation (AF) visualization method, multi-electrode mapping (MEM), delivers electrode grid $\textit{in-vivo}$ to the heart muscle and is known for its low resolution. A more cutting-edge imaging modality, near-infrared optical mapping (NIOM), allows seeing the AF sources in high resolution; however, it is currently $\textit{ex-vivo}$ only (i.e., designed for explanted organs only). In this work, we present the $\textit{ex-vivo}$ to the $\textit{in-vivo}$ learning paradigm, where the former serves the purpose of improving the latter. Specifically, the NIOM improves the detection of AF sources in MEM data via an image-to-image model. We validate the idea on 7 explanted human hearts and test the models on 2 clinical cases.

Paper Type: methodological development

Primary Subject Area: Transfer Learning and Domain Adaptation

Secondary Subject Area: Integration of Imaging and Clinical Data

Paper Status: original work, not submitted yet

Source Code Url: to be prepared

Data Set Url: restricted. compliance with the proper review board.

Registration: I acknowledge that publication of this at MIDL and in the proceedings requires at least one of the authors to register and present the work during the conference.

Authorship: I confirm that I am the author of this work and that it has not been submitted to another publication before.