Related Experiment Videos
Automatic scoring of non-apnoea arousals using hand-crafted features from the polysomnogram
1Charles Perkins Centre, Faculty of Engineering, School of Biomedical Engineering, The University of Sydney, NSW 2006, Australia.
Physiological Measurement
|December 5, 2019
Summary
This study developed an automated system for annotating non-apnoea arousals using polysomnogram (PSG) signals. The system reliably distinguishes arousal events, improving sleep study analysis.
Area of Science:
- Biomedical Engineering
- Sleep Medicine
- Machine Learning
Background:
- Automated sleep stage scoring and event detection are crucial for clinical diagnosis.
- Non-apnoea arousals, subtle sleep disruptions, require accurate identification for comprehensive sleep analysis.
Purpose of the Study:
- To develop and validate a reliable automated system for annotating non-apnoea arousals using polysomnogram (PSG) data.
- To leverage machine learning, specifically feed-forward neural networks, for enhanced sleep event detection.
Main Methods:
- Extracted 59 time- and frequency-domain features from 12 PSG signals (airflow, EEG, EOG, EMG, oximetry, respiratory effort).
- Utilized a bank of 30 feed-forward neural networks, processing combined features from adjacent 15-second epochs.
- Trained and tested the system on data from the 2018 PhysioNet/Computing in Cardiology Challenge, employing 10-fold cross-validation.
Main Results:
- The top-performing model achieved an Area Under the Receiver Operator Curve (AUC) of 0.848.
- The system demonstrated an Area Under the Precision-Recall Curve (AUPRC) of 0.325 for discriminating non-apnoea arousals.
- Performance was evaluated on an independent test set of 989 PSGs.
Conclusions:
- The developed system achieves a high degree of reliability in the automated annotation of non-apnoea arousals.
- This automated approach shows significant potential for improving the efficiency and accuracy of sleep study interpretation.