Improving ECG Arrhythmia Diagnosis with Hybrid Deep Learning Models and Cloud-Based Support

Arrhythmias, irregularities in heart rhythm, are critical indicators of heart disease and can be effectively diagnosed using electrocardiograms (ECGs), a primary tool in cardiac care. This study proposes and evaluates three hybrid deep learning models: CNN-LSTM, CNN-GRU, and Bidirectional GRU with attention (BiGRU-Att) for the classification of ECG signals into five distinct arrhythmia categories. CNN-LSTM and CNN-GRU integrate convolutional layers for spatial feature extraction with recurrent layers for temporal pattern modeling, while the BiGRU-Att model incorporates an attention mechanism to enhance focus on critical features. The preprocessing method employs the synthetic minority oversampling technique (SMOTE) to address the class imbalance, ensuring robust model performance. Results demonstrate that BiGRU-Att outperforms other models, achieving an overall accuracy of 99.4%, sensitivity of 0.987, and the lowest false positive rate of 0.002. To facilitate practical application, a user-friendly cloud-based platform was developed, enabling doctors to upload ECG signals, visualize results, and classify arrhythmias. The platform incorporates a recommendation system powered by a custom-built database, offering personalized recommendations for each arrhythmia type. These findings highlight the potential of hybrid deep learning to enhance arrhythmia detection, ultimately improving diagnostic accuracy.