SE-ResNet: Recognition of ECG Anomalies Based on Convolutional Neural Network with Fused Attention

As the most widely used approaches for observing cardiovascular disease in clinical medicine, electrocardiogram (ECG) signal can accurately reflect the physiological state of the heart. However, the unified processing of one-dimensional ECG signal is limited by the different supplies and communication protocols of ECG equipment. In this paper, we propose an image-based ECG anomaly detection method that achieves a unified processing mode through signal to image conversion. To this end, we first convert the one-dimension 12-lead ECG signal into the two-dimension image type. Then, an ECG anomaly classification mode named SE-ResNet is built by the ResNet architecture and a channel attention mechanism is utilized to enhance the feature extraction ability. Lastly, extensive experiments are conducted on a simulated clinical 12-lead ECG image dataset. The classification accuracy and F1 score of the proposed method reach 85.69% and 84.51%. Compared with the baseline method, the classification accuracy and F1 score are improved by 1.08% and 1.03% respectively. Experimental results demonstrate the effectiveness of the proposed method.