A Multi-Scale CNN-ViT Network with Squeeze-and-Excitation Block for ECG Signal Classification

Electrocardiogram (ECG) is a widely utilized diagnostic tool for detecting different cardiac irregularities, including atrial fibrillation, bradycardia, and other anomalies in the heart’s electrical activity. Menwhile, CNN-ViT networks have recently demonstrated strong performance in various computer vision tasks. In this paper, we propose a CNN-ViT network specifically designed for the ECG classification task. Our method combines the strengths of convolutional neural networks (CNNs) and Vision Transformer (ViT) networks. Convolutional block is employed to capture local ECG features, while the ViT block extracts global features. By integrating CNN and ViT, our network benefits from leveraging both local and global information for the ECG classification task. Experimental results demonstrate the superiority of our network compared to previous baseline approachs on the CPSC dataset and the Chapman dataset in terms of performance evaluation.