GLF-Net: A quasi-periodic prior-guided waveform segmentation network for ECG delineation.

Area of Science:

• Cardiology
• Biomedical Engineering
• Signal Processing

Background:

• Electrocardiogram (ECG) signals are crucial for diagnosing cardiovascular diseases.
• Accurate delineation of ECG waveform boundaries (P, QRS, T waves) is challenging due to signal overlap and morphological variability.
• Existing methods often struggle with precise waveform boundary localization.

Purpose of the Study:

• To investigate the role of R-peak prior information in ECG delineation.
• To propose a stepwise segmentation strategy for enhanced waveform boundary localization.
• To develop an advanced encoder-decoder model for improved ECG analysis.

Main Methods:

• A stepwise segmentation strategy leveraging R-peak prior information and the quasi-periodic structure of cardiac cycles.
• Classification of each wave point into P, QRS, T, or background categories.
• An encoder-decoder model fusing local morphological features and global temporal dependencies for sub-wave association.

Main Results:

• The proposed method achieved superior ECG delineation performance on LUDB and QTDB datasets.
• Average F1-scores of 97.60% (LUDB) and 94.59% (QTDB) were obtained.
• The method surpassed the performance of existing state-of-the-art approaches.

Conclusions:

• Incorporating structural cardiac priors, specifically R-peak information, significantly enhances ECG delineation accuracy.
• The developed encoder-decoder model effectively captures associations between ECG sub-waves.
• The findings suggest a promising approach for more reliable and robust automated ECG analysis in clinical settings.