A Hybrid Deep Learning Model Framework for Multi-Label ECG Classification

Abstract

Automated interpretation of multi-label electrocardiograms (ECGs) is an essential tool for diagnosing cardiovascular diseases, as multiple concurrent cardiac events can occur in a single record. Modern deep learning techniques enable rapid and accurate automated analysis of these multi-label records. In this paper, the current study proposes a hybrid model that combines a deep convolutional neural network (CNN) and bidirectional long short-term memory (BiLSTM) layers with an attention mechanism, with demographic features (age, sex) added. This study applied advanced data augmentation techniques (time warping, gaussian noise, and stochastic perturbations), during training and conducted evaluation on the PTB-XL dataset using five-fold patient-level cross-validation (5-fold CV) to ensure that patient samples do not leak between groups. The proposed model demonstrated superior performance across key evaluation metrics achieving an accuracy of 98.8%, precision of 89.4%, recall of 86.63%, F1-score of 88% and an AUC-ROC (Area Under the Receiver Operating Characteristic Curve) of 94.4%. These results confirm the effectiveness of the system in rapidly detecting cardiac abnormalities during clinical ECG screening.