Robust R-peak detection in noisy ECG using deep residual U-Net for enhanced cardiac rhythm analysis

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Signal Processing

Background:

• Accurate R-peak detection in ECG is crucial for diagnosing cardiac arrhythmias.
• Traditional methods struggle with noise and signal variations in clinical ECG data.
• A robust R-peak detection method is needed for reliable automated cardiac analysis.

Purpose of the Study:

• To introduce a novel deep residual U-Net (ResU-Net) architecture for robust R-peak detection in noisy ECG signals.
• To enhance R-peak detection accuracy and reliability across diverse clinical conditions.
• To establish a new benchmark for R-peak detection in challenging ECG recordings.

Main Methods:

• Developed a deep learning framework integrating residual networks and U-Net architecture (ResU-Net).
• Incorporated skip connections, multi-scale feature extraction, and attention mechanisms for enhanced feature representation.
• Trained and validated the model on multiple ECG databases (MIT-BIH, INCART, QT) with extensive noise stress testing.

Main Results:

• ResU-Net achieved superior performance over traditional (Pan-Tompkins, Hamilton-Tompkins) and deep learning (CNN, LSTM-CNN, ResNet-18) methods.
• Achieved high sensitivity (99.76%), positive predictive value (99.82%), and F1-score (99.79%) on the MIT-BIH database.
• Maintained over 98.2% sensitivity under severe noise conditions (-6 dB SNR), demonstrating exceptional robustness.

Conclusions:

• The proposed ResU-Net architecture sets a new standard for robust R-peak detection in noisy ECG signals.
• The integration of residual connections and U-Net enhances feature learning and noise resilience.
• This reliable method provides a strong foundation for automated cardiac rhythm analysis in clinical practice.