Physically motivated projection of the electrocardiogram—A feasibility study

Abstract

We present PhysECG: a physically motivated projection of the 12 lead electrocardiogram, supported by a deep learning model trained on 21,799 recordings from the PTB-XL database and discuss its feasibility. The method allows to evaluate the epicardial activity (inverse problem of ECG imaging) and, in particular, to distinguish left and right ventricular activity, with statistical spread related to localization of the septum. The observed dyssynchrony resembles other experimental results. The foundations of the method are based on the molecular theory of biopotentials. The heart’s activity in view of the method is decomposed into two processes: the passage of the electric activation wavefront and the response of cardiomyocytes. We introduce the idea of the electrode-resolved activity function, which represents the mass of the ventricle in Phase 0 of action potential within the lead field of each electrode. The computations are fast and robust, with excellent convergence. We present the quality metrics for the reconstruction based on the model on the testing set selected from the PTB database. In order to prove feasibility, we present and discuss two healthy controls: male and female, and two pathologies: right bundle branch block, and anterior myocardial infarction. The results obtained using PhysECG seem to be in accordance with the changes evoked by pathology, which has to be confirmed by subsequent clinical studies. The method is based on ECG, and does not require reconstruction of body geometry, which presents an affordable solution for low and middle-income countries where access to imaging is limited.