Arrhythmia Classification from ECG Signals Using LSTM Neural Networks

Le résumé fourni par la source

Automated analysis of electrocardiographic (ECG) signals is crucial for timely identification of rhythm disorders in clinical and resource-limited settings. This work presents a deep learning framework based on Long Short-Term Memory (LSTM) neural networks for five-class arrhythmia classification from the MIT-BIH Arrhythmia Database. ECG recordings (two leads, 360 Hz) were preprocessed via baseline wander removal and bandpass filtering, normalized, and segmented into fixed-length heartbeat windows centered on annotated R-peaks. Following the AAMI EC57 grouping, heartbeats were mapped to the superclasses N, S, V, F, Q. The proposed model comprises stacked LSTM layers with dropout and L2 regularization, followed by a dense output layer with softmax activation. To avoid patientspecific information leakage, evaluation was conducted with patient-wise cross-validation (GroupKFold), reporting accuracy, precision, recall, F1-score per class, macro-F1, and one-vs-rest ROC-AUC (mean$\pm$standard deviation across folds). Results show strong discrimination across classes, particularly for ventricular ectopic beats, highlighting the suitability of LSTM-based sequence models for arrhythmia analysis in ECG time series and their potential to support clinical decision-making.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.