Deep Learning-Based Electrocardiogram Signal Noise Detection and Screening Model.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Signal Processing

Background:

• Noisy biosignal data, particularly electrocardiograms (ECGs), from patient monitoring systems can impede the detection and prediction of critical clinical events.
• Deep learning algorithms offer potential for automated feature extraction, bypassing the need for manual annotation in complex biosignal analysis.

Purpose of the Study:

• To develop and evaluate a deep learning-based model for the automated screening of unacceptable, noise-containing electrocardiograms (ECGs).
• To assess the efficacy of the deep learning model in identifying noisy ECGs compared to expert medical interpretation.

Main Methods:

• A convolutional neural network (CNN) model was developed using a large dataset of 165,142,920 ECG II recordings from a trauma intensive-care unit.
• The model was trained and validated using 2,700 ECGs reviewed by a medical expert, with 9-fold cross-validation employed.
• Performance evaluation was conducted on a separate test dataset of 300 ECGs.

Main Results:

• The deep learning model achieved an area under the receiver operating characteristic curve (AUC) of 0.93.
• The model demonstrated a high F1-score of 0.80, with a sensitivity of 0.88, specificity of 0.89, positive predictive value of 0.74, and negative predictive value of 0.96.

Conclusions:

• The developed deep learning model efficiently detects and screens unacceptable ECGs, addressing a key challenge in biosignal data quality.
• This AI-driven approach shows significant promise for improving the reliability of patient monitoring data for clinical event prediction.