Development of Deep Learning Method to Predict the Extend of Atrial Remodeling using Computational Modeling

Dimitris Filos1, Dimitris Tachmatzidis2, Vassilios Vassilikos2, Ioanna Chouvarda1
1Lab of Computing, Medical Informatics and Biomedical Imaging Technologies, School of Medicine, Aristotle University of Thessaloniki, 223rd Cardiology Department, Aristotle University of Thessaloniki


Atrial Fibrillation (AF), the most common cardiac arrhythmia, leads to modifications on the electrical and/or the structural characteristics of the atrial tissue, increasing the susceptibility for AF initiation. Personalized atrial models have been proposed as a mean to guide therapeutics approaches as the extent and the distribution of the atrial remodeling differs among the patients, even in paroxysmal AF. Electrograms reflect the electrical activation of the atrial tissue in a specific area. The goal of this study is the use of simulated electrograms using computational modeling in order to extract information regarding the underline tissue characteristics. In this study, a simplified 2D model of both atria, including main anatomical structures was developed. The Countenance et al. computation model was used for the simulation of the electrical activation of both atria through the CHASTE platform. Different scenarios regarding electrophysiological properties of the atrial substrate were considered, where the characteristics of the atrial tissue were modified both in terms of atrial remodeling extent and the range of specific ionic conduction properties with and without the coexistence of fibrosis. Each tissue was assigned into one of the following classes, (i) healthy, (ii) low electrical remodeling, (iii) moderate electrical remodeling (iv) coexistence of moderate electrical and structural remodeling. In addition, several stimulation protocols were applied, based on the Heart Rate. The electrograms in several areas on the tissue were computed and a deep learning methodology was developed in order to predict the local atrial tissue characteristics. According to the results, the electrograms were accurately classified into the 5 classes with high sensitivity rates. The results denote that the analysis of the electrograms, as recorded in specific areas on the atrial tissue, can be used for the characterization of the underline atrial properties and the extent of remodeling.