| Hybrid Deep Learning structure for ECG signal Classification |
| Paper ID : 1052-ICEEM2025 (R2) |
| Authors |
|
Doaa Mostafa Khattab *1, El-Sayed M. EL Rabaie2, Fathi E. Abd El-Samie2, Heba M. Emara2 1Faculty of Electronic Engnieering 2Faculty of Electronic Engineering, Menoufia University |
| Abstract |
| Abstract— Cardiovascular diseases (CVDs), the leading cause of global mortality, demand highly accurate and efficient diagnostic methodologies. Electrocardiogram (ECG) analysis, while essential for non-invasive cardiac assessment, presents significant interpretation challenges due to signal complexity and variability. This complexity increases clinician workload and diagnostic error potential. To address these limitations, we propose a novel hybrid deep learning framework for automated multi-class ECG classification. The architecture integrates Convolutional Neural Networks (CNNs) for discriminative feature extraction with stacked Long Short-Term Memory (LSTM) networks to model long-range temporal dependencies inherent in ECG sequences. Specifically, the CNN module utilizes three sequential convolutional layers, each employing 64 filters. Temporal dynamics are captured via a dual-layer stacked LSTM structure. Evaluation on the MIT-BIH Arrhythmia and PTB Diagnostic ECG benchmark databases demonstrates high performance, with classification accuracies of 98% and 97.34% achieved on clean and noise-corrupted signals, respectively. The system effectively identifies diverse arrhythmias, including fusion beats (F), premature ventricular contractions (PVC), and atrial premature contractions (APC). These results underscore the framework potential as a robust, scalable solution for real-time ECG analysis, enhancing diagnostic precision and supporting clinical decision making. |
| Keywords |
| Keywords— ECG, LSTM, CNN, deep learning, classification, healthcare. |
| Status: Accepted |