Automatic Blink Detection: A Method for Differentiation of Wake and Sleep of Intellectually Disabled and Healthy Subjects in Long-Term Ambulatory Monitoring

To develop a method for automatic detection of blinks in electrooculograms and to evaluate reliability of blink rate as an indicator of wake and sleep in subjects with developmental brain disorders. Categorization of wake and sleep by blink rate was compared with visual sleep scoring of the polysomnograms. Ambulatory polysomnographic recordings at home or in the sleep laboratory. Nine healthy volunteers for calibration, 10 for validation; 7 intellectually disabled patients for calibration, 10 for validation of the method. N/A Blinks were detected from electrooculograms using a weighted finite impulse response median hybrid filtering and the criterion of minimum duration at amplitude threshold. More than 80% of the visually identified electrooculographic blinks were detected in most subjects. When 30-second epochs of electrooculograms with 1 or more blinks were defined as wake and those without blinks as sleep, the average agreement with polysomnographic scoring was 95% in healthy subjects and 84% in patients. The mismatch was mostly due to the 30-second epochs without blinks during waking. A contextual redefinition of wake and sleep by expanding the inspected electrooculographic span from 1 to 20 epochs (10 minutes) increased the agreement to 93% in patients. The agreement is comparable to that of actigraphy with polysomnography. The linear correlation coefficient of the proportions of sleep epochs between visual scoring and the contextual blink rate method was 0.869. The main sleep periods detected by the blink-rate method were an average of 7 minutes longer than those determined by visual scoring. This was caused by differences in the detection of sleep onsets: blinking ceased before the first stage 1 sleep period was scored. The absolute period lengths obtained by the 2 methods did not differ significantly from each other in unpaired t-tests, and the linear correlation between the values was 0.999. The blinks extracted from the electrooculographic signal can be used to reliably determine the main sleep and wake periods in both healthy subjects and patients with developmental brain disorders.