Related Experiment Video
Updated: Aug 5, 2026
04:54
IntelliSleepScorer, a Software Package with a Graphic User Interface for Mice Automated Sleep Stage Scoring
Published on: November 8, 2024
Accuracy of Machine Learning Algorithms Based on Electroencephalogram in Sleep Apnea Detection: Systematic Review and
Xiangshuo Li1, Lulu Wang1, Ting Tang2
1School of Nursing, Nanjing Medical University, 101 Longmian Avenue, Jiangning District, Nanjing, Jiangsu, 211166, China, 86 13851646546.
Journal of Medical Internet Research
|July 31, 2026
Summary
Machine learning (ML) models using electroencephalogram (EEG) data show high accuracy for detecting sleep apnea (SA) at the segment level. Further prospective studies are needed for real-world clinical integration.
Area of Science:
- Biomedical Engineering
- Neuroscience
- Artificial Intelligence in Medicine
Background:
- Sleep apnea (SA) diagnosis relies on polysomnography, a costly and time-intensive method.
- Electroencephalogram (EEG) signals offer a promising, easily extractable alternative for SA detection.
- Consistent evaluation of machine learning (ML) model performance for EEG-based SA detection is lacking.
Purpose of the Study:
- To systematically review and evaluate the diagnostic accuracy of ML algorithms for SA detection using EEG data.
- To provide an evidence base for the clinical application and future research of EEG-based SA detection.
- To assess the performance of ML models in identifying sleep apnea from neural activity patterns.
Main Methods:
- Systematic search of multiple databases (PubMed, Embase, Web of Science, etc.) following PRISMA-DTA and PRISMA 2020 guidelines.
- Inclusion of studies evaluating ML algorithms for SA detection solely based on EEG data.
- Risk of bias assessment using QUADAS-2 and P-RM-AI tools; statistical analysis in R and Meta-DiSc; GRADE for evidence certainty.
Main Results:
- Twenty-seven retrospective studies were included, demonstrating high segment-level diagnostic performance (pooled sensitivity: 0.90, specificity: 0.92).
- The pooled area under the ROC curve was 0.95, indicating strong diagnostic capability.
- Meta-regression revealed EEG channel configuration, region, and validation strategy as sources of heterogeneity; multichannel EEG and deep learning showed better performance.
Conclusions:
- EEG-based ML models show significant diagnostic accuracy for SA detection at the segment level, promising for screening and decision support.
- Current evidence primarily consists of retrospective, segment-level analyses, potentially overestimating real-world utility.
- Prospective, patient-level validation studies with full-night monitoring are essential for clinical integration.