Classification of healthy subjects and insomniac patients based on automated sleep onset detection

This work aims to investigate new indexes quantitatively differentiate sleep insomnia patients from healthy subjects, in the context of sleep onset fluctuations. Our study included the use of existing PSG dataset, of 20 healthy subjects and 20 insomniac subjects. The differences between normal sleepers and insomniacs was investigated, in terms of dynamics and content of Sleep Onset (SO) process. An automated system was created to achieve this and it consists of six steps: 1) preprocessing of signals 2) feature extraction 3) classification 4) automatic scoring 5) sleep onset detection 6) identification of subject groups. The pre-processing step consisted of the removal of noise and movement artifacts from the signals. The feature extracting step consists of extracting time, frequency and non-linear features of Electroencephalogram (EEG) and Electromyogram (EMG) signals. In the third step, classification was done using ANN (Artificial Neural Networks) classifier. The fourth step consisted of scoring sleep stages (wake, S1, S2, S3 and REM) and produced a hypnogram. In the fifth step, we are detecting sleep onset from our automatic detected hypnogram and identified time of SO reference point and the combination of stages. In the final step we differentiated healthy subjects from insomniac patients based on the parameters calculated in the fifth step