Development of a probability model for high-resolution drowsiness detection using electroencephalogram

Capturing the dynamics of sleep onset process is fundamental to sleep medicine and circadian neurobiology. Even though wakefulness/sleep transition is a gradual and continuous process, it has been considered instantaneous and scored subjectively at low resolution. Therefore, a model to capture the dynamics of wakefulness to sleep transition is needed. The purpose of this study is to develop an efficient, high-resolution, and reliable model to quantitatively capture the dynamics of wakefulness/sleep transition using electroencephalogram (EEG).

We collected EEG signals from 53 subjects during an overnight sleep study. We extracted relative power features from EEG to develop a new model that yields the likelihood of wakefulness for each of the 3-s EEG segments. Furthermore, using the model, we identified three clusters, namely wakefulness, drowsiness and sleep, and employed statistical analyses, cluster quality evaluation, and graphical analysis for validation.

The proposed method successfully separated three distinct cases of alertness. The mean silhouette value on the test data was 0.74 and the mean Davies-Bouldin index value was 0.43, which indicated that the three discovered clusters were compact. Based on the silhouette values, the detection accuracy was 93.21 %. One-way repeated measures analysis of variance results suggested that the feature values were significantly different (p < .0001) among the three detected clusters.

The proposed method was able to detect short episodes of wakefulness, drowsiness, and sleep with high accuracy in overnight polysomnography data. This proof-of-concept study suggests potential future applications in drowsiness detection, pending validation in relevant contexts such as driving simulators and workplace environments.