Automatic scoring of sleep stages and cortical arousals using two electrodes on the forehead: validation in healthy adults

Accuracy and limitations of automatic scoring of sleep stages and electroencephalogram arousals from a single derivation (F-p1-F-p2) were studied in 29 healthy adults using a portable wireless polysomnographic recorder. All recordings were scored five times: twice by a referent scorer who viewed the standard polysomnographic montage and observed the American Academy of Sleep Medicine rules (referent scoring and blind rescoring); and once by the same scorer who viewed only the F-p1-F-p2 signal (alternative scoring), by another expert from the same institution, and by the algorithm. Automatic, alternative and independent expert scoring were compared with the referent scoring on an epoch-by-epoch basis. The algorithm's agreement with the reference (81.0%, Cohen's =0.75) was comparable to the inter-rater agreement (83.3%, Cohen's =0.78) or agreement between the referent scoring and manual scoring of the frontopolar derivation (80.7%, Cohen's =0.75). Most misclassifications by the algorithm occurred during uneventful wake/sleep transitions, whereas cortical arousals, rapid eye movement and stable non-rapid eye movement sleep were detected accurately. The algorithm yielded accurate estimates of total sleep time, sleep efficiency, sleep latency, arousal indices and times spent in different stages. The findings affirm the utility of automatic scoring of stages and arousals from a single frontopolar derivation as a method for assessment of sleep architecture in healthy adults.