A neural network approach for the prediction of arrhythmic events in patients with Brugada syndrome via ECG features analysis

Brugada syndrome is an inherited cardiac electrical disorder associated with elevated risks of ventricular fibrillation and sudden cardiac death. Precise evaluation of the risk of arrhythmic events could prevent premature deaths and unnecessary interventions. This research uses neural networks to analyze electrocardiograms (ECGs) to assess their effectiveness in predicting whether or not a patient could experience fatal cardiac episodes by differentiating between subjects with and without documented arrhythmic events. The study involved 265 ECGs on which fourteen ECG features were independently measured by three cardiologists. The classification performances were evaluated by considering: accuracy, positive and negative predictive values, specificity, sensitivity, and AUC. Also, univariate statistical tests were performed to select the most significant features. Results show the capability of neural networks for risk stratification in patients with Brugada syndrome, achieving satisfactory performance in both validation (93% accuracy, 93% AUC) and test (88% accuracy, 90% AUC) sets, proving the ability of AI-based ECG analysis in assisting in clinical decision-making.