Validation of an EEG-Based Algorithm for Automatic Detection of Sleep Onset in the Multiple Sleep Latency Test

The Multiple Sleep Latency Test (MSLT) is a standard test to objectively evaluate patients with excessive daytime sleepiness. Sleep onset latencies are determined by visual analysis, which is costly and time-consuming. The aim of this study was to implement and test a single automatic algorithm to detect the sleep onset in the MSLT on the basis of electroencephalographic (EEG) signals. The designed algorithm computed the relative EEG spectral powers in the occipital area and detected the sleep onset corresponding to the intersection point between the lower and alpha frequencies. The algorithm performance was evaluated by comparing the sleep latencies computed automatically by the algorithm and by a sleep specialist using MSLT recordings from a total of 19 patients (95 naps). The mean difference in sleep latency between the two methods was 0.025 min and the limits of agreement were +/- 2.46 min (Bland-Altman analysis). Moreover, the intra-class correlation coefficient showed a considerable inter-rater reliability (0.90). The algorithm accurately detected the sleep onset in the MSLT. The devised algorithm can be a useful tool to support and speed up the sleep specialist's work in routine clinical MSLT assessment.