SER inspired deep learning approach to detect cardiac arrhythmias in electrocardiogram signals using Temporal...

Electrocardiogram (ECG) signals play a critical role in diagnosing cardiovascular diseases (CVDs), yet automated ECG classification remains challenging due to inter-patient variability, signal noise, and heart rhythm complexity. To address these challenges, we propose a novel deep learning approach that combines Temporal Convolutional Networks (TCN) with Graph Convolutional Networks (GCN) for robust cardiac arrhythmia detection. Unlike conventional methods, our approach constructs cycle graphs from ECG signals, leveraging graph signal processing to enhance data representation and improve classification accuracy. The proposed model processes raw ECG signals using a TCN-based feature extractor, followed by spectral graph convolutions via a GCN. We evaluate our model using the Chapman and Shaoxing 12-lead ECG database, consolidating 11 rhythm classes into four superclasses for improved classification reliability. Our model achieves a high classification accuracy of 95.5%, surpassing standard GCNs and other deep learning architectures while significantly reducing graph construction complexity. These results demonstrate the potential of graph-based deep learning for efficient and accurate ECG analysis in resource-constrained medical settings.