Robust R-Peak Detection in Low-Quality Holter ECGs Using 1D Convolutional Neural Network.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Artificial Intelligence

Background:

• Electrocardiogram (ECG) signal quality from Holter and wearable devices often hinders accurate R-peak detection.
• Existing R-peak detection algorithms struggle with the noise and low fidelity characteristic of such ECG records.

Purpose of the Study:

• To develop a generic and robust system for R-peak detection in Holter ECG signals.
• To improve the accuracy and reduce false alarms in R-peak detection, particularly for low-quality ECG data.

Main Methods:

• A novel 1D Convolutional Neural Network (CNN) architecture with an encoder-decoder structure was implemented.
• A verification model was integrated to minimize false positives.
• The system was trained and tested on large, open-access ECG databases (CPSC-DB and MIT-BIH Arrhythmia Database).

Main Results:

• Achieved state-of-the-art performance with 99.30% F1-score on the CPSC-DB and 99.83% F1-score on the MIT-BIH Arrhythmia Database.
• Reduced false positives by over 54% and false negatives by over 82% compared to competing methods.
• Demonstrated high precision (98.91% on CPSC-DB, 99.82% on MIT-DB) and recall (99.69% on CPSC-DB, 99.85% on MIT-DB).

Conclusions:

• The proposed system offers a generic and robust solution for R-peak detection applicable to any ECG dataset.
• The CNN-based approach effectively handles noisy and low-quality ECG signals.
• The system's simple and invariant parameters facilitate its application in real-time monitoring on portable devices.