Enhanced sleep staging with artificial intelligence: a validation study of new software for sleep scoring

Area of Science:

• Sleep Medicine
• Artificial Intelligence in Healthcare
• Biomedical Signal Processing

Background:

• Manual sleep staging (MSS) from polysomnography is labor-intensive, requires extensive training, and suffers from inter-scorer variability.
• Automatic sleep staging (ASS) using machine learning aims to provide a more efficient and consistent alternative.

Purpose of the Study:

• To evaluate the agreement of STAGER, a machine learning-based software, with manual sleep staging (MSS) performed in clinical practice and by expert technicians.
• To assess the accuracy and reliability of STAGER's automatic sleep staging (ASS) using only electroencephalography (EEG) signals.

Main Methods:

• Retrospective analysis of 40 polysomnographic recordings from patients referred to US sleep clinics.
• Independent sleep staging by three expert technicians and comparison with STAGER's ASS and clinical MSS.
• Calculation of agreement statistics and bootstrap resampling for confidence intervals and significance testing.

Main Results:

• STAGER's ASS demonstrated high agreement with MSS, often statistically superior to inter-technician agreement.
• Accuracy was comparable to MSS across most sleep stages, with a slight reduction in positive percent agreement for the wake stage.
• The software's performance was validated against established clinical protocols and expert evaluations.

Conclusions:

• STAGER software shows promising accuracy for automatic sleep staging of inpatient polysomnography recordings.
• It offers a reliable alternative to manual sleep staging, potentially improving efficiency and reducing variability in sleep analysis.
• Further validation may enhance its utility in clinical sleep evaluation settings.