Non-pharmacological early detection of concealed Brugada syndrome via basal ECG analysis

Abstract Introduction We present a novel, non-pharmacological method for the early detection of concealed Brugada syndrome (BrS) through basal electrocardiogram (ECG) analysis, offering a potentially more accessible and cost-effective alternative to pharmacological provocation testing. Methods A total of 196 consecutive patients presenting to our center with suspected BrS were enrolled, all of whom exhibited clinical features suggestive of BrS (arrhythmic syncope, nocturnal agonical respiration) and/or a family history of BrS or sudden arrhythmic death syndrome, however without a spontaneous type 1 Brugada ECG pattern. A standard 12-lead ECG was recorded for each patient before any pharmacological intervention. Subsequently, in accordance with current guidelines, a sodium channel blocker challenge with ajmaline was performed and independently interpreted as either positive (n=111) or negative (n=85) by three certified electrophysiologists at our Institution. After data collection, advanced analysis techniques were employed, including delayed cross-Poincaré plot analysis, topological data analysis, and machine learning techniques. Results All 196 basal 12-lead ECG recordings were subsequently preprocessed to remove artifacts and a balanced subset (n=50) was analyzed, with a focus on precordial leads V1 and V2. An advanced delayed cross-Poincaré plot analysis was developed to assess the interactions between these leads, capturing the dynamic interplay between the right and left ventricles. The findings revealed distinct patterns of ventricular electrical activity in BrS patients compared to healthy controls. Specifically, subtle ST-segment changes were identified as more pronounced in the anatomical region corresponding to the right ventricular outflow tract, consistent with known electrical imbalances associated with sodium channel dysfunction at this level. Thereafter, topological data analysis and machine learning techniques for nonlinear dynamical systems were utilized. A variational autoencoder (VAE) was employed to transform the point clouds derived from the cross- Poincaré plots into a compact, unsupervised representation, enabling clear separation of positive and negative samples through t-distributed stochastic neighbor embedding (t-SNE) clustering. This approach achieved an area under the receiver-operating characteristic (ROC) curve of 0.93 (95% CI: 0.90-0.95), demonstrating high accuracy in differentiating BrS cases. Conclusion Our results show a robust ability to differentiate BrS patients from healthy individuals, underscoring the promise of this method as a convenient non-pharmacological tool for early detection of BrS. To our knowledge, this is the first application of topology-based analysis combined with unsupervised learning for the electrocardiographic detection of BrS. However, further refinement and larger cohort studies are necessary to enhance the diagnostic specificity and validate the clinical utility.Cross-Poincaré plots Delayed cross-Poincaré plots