Quantifying CineECG Output for Enhancing Electrocardiography Signals Classification

Abstract

CineECG, a vectorcardiography-based method, uses standard 12-lead electrocardiography and 3D heart and torso models to depict the electrical activation path during the heart cycle, offering detailed visualization of cardiac electrical activity without numerical quantification. Our research aims to quantify CineECG outputs by defining 54 features that describe the route, shape, and direction of electrical activation. These features were used to develop a multinomial regression model classifying electrocardiography signals into normal sinus rhythm, left bundle branch block, right bundle branch block, and undetermined abnormalities. Trained and tested on 6,860 signals from the PhysioNet/Computing in Cardiology Challenge 2020 and THEW project, the model achieved an F1 score over 84% (normal sinus rhythm: 93%, left bundle branch block: 93%, right bundle branch block: 90%, undetermined abnormalities: 84%). The results suggest CineECG's potential in enhancing electrocardiography interpretation and aiding in the accurate diagnosis of various abnormalities.