A sleep spindle detection algorithm that emulates human expert spindle scoring.

Area of Science:

• Neuroscience
• Sleep Science
• Computational Biology

Background:

• Sleep spindles, characteristic of stage 2 NREM sleep, are vital biomarkers for learning, memory, and neurological disease assessment.
• Current gold standard for sleep spindle detection, visual EEG inspection, is labor-intensive, costly, and prone to inter-scorer variability.

Purpose of the Study:

• To develop and validate a novel, efficient sleep spindle detection algorithm (A7) that mimics human scoring accuracy.
• To provide an open-source tool for reproducible sleep spindle analysis.

Main Methods:

• The A7 algorithm utilizes a single EEG channel and analyzes absolute sigma power, relative sigma power, and sigma band-passed signal correlation/covariance.
• Performance was evaluated against a gold standard dataset established by expert human scorer consensus.

Main Results:

• The A7 detector achieved a by-event performance of 74% precision, 68% recall, and an F1-score of 0.70, comparable to individual human experts (average F1=0.67).
• A7 demonstrated superior performance over existing automated detectors, with an F1-score 0.17 points higher.
• On a by-subject basis, A7's spindle density estimates strongly correlated with human experts (r²=0.82), significantly outperforming other automated methods (average r²=0.27).

Conclusions:

• The A7 algorithm offers a sensitive and precise method for sleep spindle detection, effectively emulating human scoring while minimizing missed detections.
• The open-source availability of A7 facilitates its adoption and further validation in sleep research and clinical applications.