An Efficient Algorithm for Detecting Total Sleep Time from Electroencephalography Signal

Polysomnography (PSG), the gold standard for sleep analysis, is manually scored by sleep experts. The measurement of total sleep time (TST) from PSG is very important for diagnosing sleep related abnormalities. Manual scoring of TST is time consuming and vary among experts. This study aimed to develop an algorithm for automatically detecting the TST from single-channel electroencephalography (EEG) signals collected from overnight PSG recordings. The algorithm was developed and optimized using a development dataset of 41 EEG records and tested using a validation dataset of 46 EEG records. The designed algorithm took the raw EEG as input, and pre-processed it by filtering and removing the phase shift. The feature was extracted from each 30-second EEG epoch and classified each epoch as sleep or wake using an optimized threshold. TST was estimated by summing up these epochs classified as sleep. The estimated TST was compared to the standard annotation for performance evaluation. A good agreement was established between the algorithm and annotation using the Bland-Altman plot that indicated a very small mean bias of -0.16 hours with a narrow limit of agreement of -3.76 to 3.43 hours. Since the designed method outperformed the existing relevant studies, this algorithm can be effectively applied for the early diagnosis of sleep related disorders, especially sleep apnea.