The aim of this study was to develop the first freely available software for automated analysis of sleep electroencephalogram (EEG) microstructures without relying on prior sleep staging. Hereby, we present RAVEN (Rhythmic Analysis of Variations in EEG Neural activity), an automated analysis tool for K-complexes, spindles, delta waves, and cyclic alternating patterns (CAPs). RAVEN is MATLAB-based software for analysing EEG recordings in European Data Format and SleepLab Format. RAVEN relies on American Academy of Sleep Medicine (AASM)-compliant. RAVEN was developed using high-quality polysomnography data from 10 healthy volunteers and validated in three independent datasets: the Montreal Archive of Sleep Studies (MASS; n = 19), the CAP Database (n = 8), and the SmartSleep Lab (SSL; n = 11). In the CAP database precision, sensitivity, and F1-score were 0.53, 0.55, 0.54, respectively. In the MASS dataset, precisions, sensitivities, and F1-scores were, 0.57, 0.29, and 0.38 for K-complexes; and 0.82, 0.39, and 0.53 for spindles, respectively. In the SSL dataset, precisions, sensitivities, and F1-scores were 0.20, 0.14, and 0.16 for K-complexes; 0.70, 0.52, and 0.60 for spindles; and 0.80, 0.66, and 0.73 for delta waves, respectively. RAVEN identified sleep microstructures automatically from EEG recordings based solely on signal characteristics and signal-derived parameters, without prior sleep staging. In addition to detecting the expected microstructures, RAVEN also identified events that were missed during manual scoring. Detected events were consistent with published literature and aligned with AASM guidelines, demonstrating RAVEN’s potential as a robust and objective tool for sleep microstructure analysis.
Arolaakso et al. (Thu,) studied this question.