Insomnia disorder detection using EEG sleep trajectories

In this paper, we present a novel objective method for insomnia disorder detection based on sleep trajectories, without sleep stage scoring. The sleep trajectories are computed from EEG recordings, generated by BrainTrak, which estimates the physiological parameters of the brain activity during sleep using a neural-field brain model. This method also allows combining multiple different EEG datasets, as sleep trajectories are not affected by systemic differences in EEG collection. We then propose a data-driven semi-supervised approach based on multi-class conditional deep convolutional GAN (CDCGAN) to distinguish between people with insomnia and normal sleepers. Our method uses CDCGAN as a semi-supervised classifier on 20-min subtrajectories of sleep to learn and identify the characteristics of insomnia disorder compared to normal sleepers. We conducted an evaluation using two datasets: Insomnia-100 and MASS. CDCGAN achieved an accuracy of 74.5%, substantially outperforming CNN, kNN and SVM approaches used for comparison. More generally, our work demonstrates the potential of GANs for medical informatics applications.