Automatic sleep classification according to Rechtschaffen and Kales

Area of Science:

• Sleep Medicine
• Biomedical Engineering
• Artificial Intelligence in Healthcare

Background:

• Traditional polysomnography relies on 1968 Rechtschaffen and Kales (R&K) visual scoring rules.
• Manual sleep scoring is time-consuming and subject to inter-rater variability.
• There is a need for automated, reliable sleep analysis tools in clinical practice.

Purpose of the Study:

• To develop and validate an automated sleep classification system, Somnolyzer 24x7.
• To assess the system's adherence to R&K scoring rules and its performance against human experts.
• To demonstrate the system's reliability and validity for clinical and research applications.

Main Methods:

• Developed an e-health solution, Somnolyzer 24x7, incorporating quality checks, feature extraction (sleep spindles, delta waves, eye movements), and an expert system classifier.
• Implemented rule-based smoothing for REM and stage 2 sleep transitions.
• Validated the system on 286 recordings from healthy individuals and patients with sleep disorders, comparing results with human expert scoring.

Main Results:

• Somnolyzer 24x7 achieved 80% epoch-by-epoch agreement (Cohen's Kappa: 0.72) with human experts, exceeding inter-rater reliability (77%, Kappa: 0.68).
• The system demonstrated excellent inter-rater reliability (Kappa: 0.991) when used with human quality control.
• Correlation analyses showed comparable or superior relationships between Somnolyzer 24x7 and expert evaluations for R&K target variables.

Conclusions:

• The Somnolyzer 24x7 system demonstrates high reliability and validity in automatic sleep stage classification.
• Its performance is comparable, and in some aspects superior, to human expert scoring.
• The system is suitable for integration into clinical routine and sleep studies, offering efficient and accurate sleep analysis.