Explainable deep learning models for ECG classification using CNN, LSTM, and BiLSTM architectures

Cardiovascular diseases (CVD) are one of the main causes of premature death worldwide, early and exact diagnosis is crucial. Identify arrhythmias & other cardiac disorders, electrocardiograms (ECGs), which capture the electrical activity of the heart isvital. However, manually interpreting ECGs requires lot of time and is prone to errors. This work explores the use of CNN, CNN-LSTM, & CNN-BiLSTM architectures, which are deep learning (DL) models, for automatic ECG classification using spectrogram representations extracted from the MIT-BIH arrhythmia. While LSTM & BiLSTM layers record temporal dependencies within beat sequences, CNN layers extract spatial features from spectrograms. With a high classification accuracy of 96.8%, the CNN-BiLSTM model performed the best. These findings demonstrate how deep learning can reduce manual labor and minimize diagnostic errors while offering quick, dependable, and consistent assistance with clinical ECG diagnosis.