A without-stage-annotation insomnia assessment using single-channel electroencephalography

Polysomnography (PSG) is a gold standard for diagnosing insomnia and monitoring the diverse sleep parameters, although there are wide-spreading challenges due to its complexity and time-consuming usage. To primarily assess the insomnia of the subjects, an apparatus based on fewer-electrode electroencephalography (EEG) could be more convenient and self-satisfactory prior to the clinical PSG intervention. To develop the technique, exploring EEG-based objective features between insomnia and the healthy subjects is the first vital step to ensure its realization. This study therefore introduced an analytic method for extracting the convincing EEG features to assess insomnia effectively without additional sleep stage annotation. To deal with this, a signal processing procedure was introduced to clean the signals from the candidate's EEG channels, and a feature selection procedure was then imposed on a set of spectral/temporal features which were extracted from the EEG signals throughout sleep to dig out the adequate information for the insomnia assessment. Validating with a 1-dimensional convolutional neural networks (1D-CNN) model, the analytic steps in the procedures guaranteed the obtained features, including the zero-crossing rate, the absolute slow-wave power, the relative θ power, and the optimal assessing performance, in which the best accuracy (95.00 %) and 90 % Cohen's kappa value were achieved while validating with 100 clinic Fp2 channel records.