Adaptive learning and cross training improves R-wave detection in ECG.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare

Background:

• Automated R-wave detection is crucial for electrocardiography (ECG) and computer-aided diagnosis.
• Traditional methods have limitations, especially with noisy or pathological signals.
• A multi-level 1D deep learning approach shows promise but can be improved.

Purpose of the Study:

• To enhance a multi-level 1D convolutional neural network (CNN) for improved R-wave detection.
• To investigate the impact of adaptive deep learning, cross-database training, and cross-lead training.
• To evaluate the proposed approach on large, diverse ECG datasets.

Main Methods:

• Implemented adaptive deep learning, cross-database training, and cross-lead training on a 1D CNN architecture.
• Utilized four public ECG databases (MIT-BIH, INCART, TELE, SDDB) with over 4.5 million annotated beats.
• Employed a 5-fold cross-validation scheme for robust evaluation.

Main Results:

• The proposed approach achieved state-of-the-art performance, with F-measures of 99.75% (MIT-BIH) and 95.25% (TELE).
• Cross-database training (98.02%) proved more effective than individual database training (97.33%).
• Performance further improved with the inclusion of additional databases and cross-lead training.

Conclusions:

• Adaptive cross-data training optimizes models for robust R-wave detection.
• The enhanced deep learning approach effectively analyzes noisy, pathological, and mobile-recorded ECGs.
• Utilizing multiple datasets and leads enables analysis without ground truths, advancing ECG diagnostics.