A Comparative Study of ECG Leads in Predicting Cardiac Arrhythmias Using Deep Learning Models

Abstract

Cardiac arrhythmias are a group of conditions that have a high incidence and prevalence worldwide, and receive considerable attention from the medical community because they are associated with several risk factors and can cause serious impairment of the individual's cardiac function in more critical cases. The electrocardiogram is the main tool for the diagnosis of cardiac arrhythmias because it is considered flexible, non-invasive, and low-cost. The so-called 12-lead system is the most widely used ECG configuration in clinical practice and has been considered for several years as the gold standard for detecting cardiac arrhythmias. Although this configuration is widely popular, there are situations in which it may be more interesting to use simpler ECG configurations to expand the tool to scenarios other than traditional healthcare environments, such as using mobile devices for cardiac monitoring. These scenarios require using simplified ECG configurations, using a single lead or a subset of leads, due to technical restrictions of the devices or limitations of the scenario itself. Knowing the performance of each lead when considered individually is important for defining which leads are most suitable for use in each scenario. This study presents a comparative analysis of the leads of the 12-lead system for predicting cardiac arrhythmias employing a deep learning-based approach and a large dataset containing diagnoses of 32 types of arrhythmias. A large public dataset well-annotated according to international standards for arrhythmia diagnosis was used. Both individual results on the performance of each lead and patterns involving groups of leads that share common characteristics were highlighted. The results presented allow healthcare professionals to be equipped with quantitative data that can provide a robust basis for decision-making and overall improvement of medical processes. The results demonstrate the feasibility of using technologies based on Artificial Intelligence as tools to support cardiology practice and the expansion of cardiac monitoring practices to environments outside clinics and hospitals.