Wavelet Decomposition Based Automatic Sleep Stage Classification Using EEG

The diagnosis of sleep related disorders like sleep apnea, insomnia, restless legs syndrome, begins with grading sleep into stages to analyze the problem. The R&K rules recommend dividing the polysomnographic record of sleep consisting of EEG, EOG and EMG into 30 s epochs and classifying them as Stage 1, 2, 3, 4, Rapid Eye Movement (REM) and Wake state. In this paper, data from a single EEG electrode are decomposed into its wavelet coefficients (Daubechies wavelet from 2 to 6). Instead of using statistical parameters like entropy, energy, etc. of the coefficients as features, the coefficients are directly applied as input to a neural network for classification. Prior to training the neural network, the high dimensional input data are reduced to its principal components. The proposed method helps in isolating Stage 3 and 4, rather than identifying them as a combined Slow Wave Stage (SWS). Best results were obtained using Daubechies 2 wavelet, with an overall accuracy of 86%.