Myocardial infarction (MI) is a life-threatening condition that requires accurate and rapid detection through electrocardiogram (ECG) signals for effective clinical management. Tra ditional methods, such as rule-based approaches and shallow learning models, often struggle to capture the complex, non-linear patterns associated with MI, limiting diagnostic accu racy and computational efficiency. This study presents an efficient hybrid deep learning model (EHDL) for improved MI detection efficiency, leveraging a CNN-BILSTM architec ture integrated with Depthwise Separable Convolution (DSC). CNN-BILSTM effectively captures intricate spatial and temporal dependencies in ECG signals, while DSC optimizes processing time and memory usage. The proposed model demonstrates reliability and ef f iciency compared to alternatives such as CNN-BILSTM, CNN-LSTM, and CNN-RNN. The method enables efficient and accurate classification of normal and nine MI subtypes, and achieves higher accuracy and F1 scores. Evaluations were performed on the PTB-XL dataset, comprising 37,685 ECG recordings. Feature extraction using Discrete Wavelet Transform (DWT) produced 36-dimensional feature vectors. Two configurations were an alyzed: a baseline model and an optimized version with hyperparameter tuning. The tuned EHDL model achieved an accuracy of 94.6%, a sensitivity of 94.2%, a specificity of 99.4%, precision of 94.5% and an F1 score of 94.8%. It also demonstrated efficient resource uti lization, requiring only 41.0 MB of memory, a 2.3-second inference time, and 370.1 seconds of training time. Consistent accuracy and loss trends during the training and validation phases confirm the robustness of the model. These findings position EHDL as a compu tationally efficient and highly effective alternative to traditional HDL models, showcasing significant potential for medical signal analysis. Future work will focus on improving data quality and exploring advanced architectures to enhance detection capabilities.
