Ensemble Classifier based on Linear Discriminant Analysis for distinguishing Brugada Syndrome Patients according to Symptomatology
Abstract
Identifying high-risk patients requiring an ICD among asymptomatic Brugada patients is nowadays a bit challenging. In this study, 62 patients suffering from Brugada syndrome (14 symptomatic) were studied by analyzing the 12-lead ECG recording acquired during a physical exercise test. For each patient, conventional HRV indices from time-frequency analysis and heart rate recovery (HRV features), as well as several morphological depolarization indices (QRS features), were evaluated at relevant periods of the test. Most discriminant features from both the HRV and QRS sets were selected using a two-stage feature selection algorithm and used for model classification building. For the detection step, an ensemble classifier using stacking approach plus a fixed combiner was designed, using linear discriminant analysis as the base classification algorithm. Best features from each model were then used for building the final individual and combined classification models. Detection performance using the symptomatic group as the target class, was as follows: HRV-based model: Se= 1, Sp= 0.67, AUC= 0.87; QRS-based model: Se= 75, Sp= 0.67 AUC= 0.73. When joining best features of both models (HRV-QRS-based model), the performance increased up to Se= 1, Sp= 0.83, AUC= 0.90. The study showed that by combining both HRV and depolarization analysis, a better risk stratification can be performed. This could be useful for the identification of Brugada patients with previous symptoms, and it may help to the decision making process of asymptomatic patients needing an ICD.