ECG Characteristic Detection Using DenseNet based on Attention Mechanism and Feature Pyramid

Abstract

Heartbeat characteristic points are the main features of electrocardiogram (ECG), which can provide important information for ECG-based cardiac diagnosis. In this manuscript, we proposed a modified DenseNet to detect ECG characteristic points, including the onset, peak and termination points of P-wave, QRS complex wave and T-wave. The proposed DenseNet embedded the convolutional block attention module (CBAM) and feature pyramids in convolutional blocks. The model in this manuscript was trained using PhysioNet’s QT database, a dataset containing 105 2-lead dynamic recordings at 250 Hz. The mean± variance of the detection errors measured by our proposed method for the detection of the onset, peak and termination points of P-wave, the onset and termination points of QRS-wave, and the peak and termination points of T-wave were - 0.32 ± 18.08 ms, -0.56 ± 17.6 ms, -5.96 ± 16.84 ms, -5.8 ± 14.12 ms, -6.24 ± 18.76 ms, -0.2 ± 31.36 ms, 0.84 ± 27.24 ms, respectively. The results show that the deep learning network constructed in this manuscript can accurately detect the characteristic points of heartbeat, laying the foundation for automatic extraction of diagnosis-related key information from ECG.