ECG Images Analysis for the Prediction of Fatal Arrhythmic Events in Patients with Brugada Syndrome Based on a Vision Transformer Model

Brugada syndrome (BrS) is an inherited arrhythmogenic disease that is distinguished by an increased risk of sudden cardiac death due to the appearance of ventricular arrhythmias without any structural heart diseases. There are distinct types of BrS electrocardiogram (ECG) patterns observed in the right precordial leads, but only Type 1, characterized by a coved shape, is considered a diagnostic indicator of Brugada syndrome. Patients with spontaneous Type 1 who have experienced ventricular fibrillation are generally recommended to undergo implantable cardioverter-defibrillator (ICD) placement to prevent sudden death, despite the associated complications. Nevertheless, risk stratification in Brugada syndrome remains challenging, particularly in cases where patients do not exhibit symptoms. Due to limitations in experimental research involving human cardiac tissue, there is growing interest in alternative methods such as artificial intelligence (AI) based on deep learning. In this study, we present a Vision Transformer model that utilizes 12-lead ECG images as input to predict arrhythmic fatal events in patients with Brugada syndrome. The employed dataset is composed of 278 ECGs from the Piedmont Brugada Register, where 94 ECGs are connected to patients who have developed significant arrhythmic events and the remaining portion belongs to patients without any recorded arrhythmic events. The promising results obtained in negative predictive value and sensitivity may reveal the AI-based ECG analysis as an encouraging tool for risk stratification of Brugada Syndrome patients, offering helpful support for clinical decision-making, and possibly preventing sudden cardiac death (SCD).