A Refined ECG Delineation Framework Incorporating Single-Beat Mode and Conditional Random Field

Electrocardiogram (ECG) signals reflect heart electrical activity and are essential for diagnosing cardiovascular diseases. The typical ECG waveform consists of P-waves, QRS complexes, and T-waves, whose morphological variations reflect cardiac function. However, it is challenging to delineate the boundaries of these waveforms accurately due to signal overlap between consecutive heartbeats and the periodic complexity of the waveforms. Therefore, this study introduces a novel delineation method that integrates sample point pre-classification with a conditional random field (CRF) model, enabling precise identification of ECG waveform onset and offset points. The method begins with R-peak-based segmentation to isolate individual cardiac cycles and reduce inter-beat interference. An encoder-decoder structure is then employed to capture both local waveform features and global contextual relationships. Finally, the CRF incorporates segment dependencies to optimize boundary identification. Experimental results demonstrate that the approach achieves an average accuracy of 95.47%, outperforming state-of-the-art methods, particularly in detecting QRS complexes and T-waves.