Automatic Sleep Stage Classification Based on Convolutional Neural Networks

This paper proposes a CNN model for automatic sleep stage scoring based on 4 Electroencephalogram (EEG) and 2 Electrooculogram (EOG) channels. Clinically, this task is done manually using polysomnographic (PSG) records by highly trained professional sleep experts, which is slow and labor-intensive. The proposed method was evaluated using PSG data from 184 subjects, recorded in Fukushima Oostsuki Clinic, Fukushima, Japan. The model showed an overall accuracy of 84.13 % and precision, recall and, F ₁ score measures were calculated as ~ 84 %. Clearly, this model is capable of classifying sleep stages without employing any hand engineered features.