ECG Signal Delineation Based on Multi-scale Channel Attention Convolutional Neural Network

Accurate delineation of ECG signals is of paramount importance for cardiovascular disease diagnosis. Existing machine learning and deep learning methods often yield suboptimal results in waveform delineation in the presence of noise. In this study, we propose a novel ECG signal delineation model, namely the Multi-Scale Channel Attention Convolutional Neural Network (MSCA-CNN). MSCA-CNN effectively captures multi-scale features using the MSCA-Block and enhances the discrimination of relevant components of ECG waveforms through channel attention mechanisms. Additionally, we introduce a domain-knowledge-based post-processing algorithm to eliminate misclassified data points. Experimental evaluation conducted on the publicly available database demonstrates the superior performance and robustness of MSCA-CNN compared to state-of-the-art methods. The reported average sensitivity and accuracy for various waveform onsets are 99.84% and 98.35%, respectively. Notably, MSCA-CNN accurately detects and delineates important features such as P-wave, QRS complex, and T-wave onsets even in the presence of noise. This research holds significant promise for advancing ECG diagnosis and improving the performance of ECG analysis tasks.