Automatic Sleep Staging Based On EEG

The automated scoring of sleep stages, using Electroencephalograph (EEG), is introduced inthis thesis. Sleep states are classified into five stages, i.e., awake, sleep stage1,2, slow wavesleep (SWS, sleep stage3and4) and rapid eye movement (REM). The frequency spectrum ofEEG changes with the transition of sleep stages. Based on this, we applied the time-frequencyanalysis on EEG using wavelet transform. Through the analysis on wavelet energy in differentfrequency bands, we find that the ratio of alpha band and delta band can clearly distinguish eachsleep stage. Further, we employ Neuro-Fuzzy Network based on ANFIS model, using relativewavelet energy in different frequency bands as the input parameters, to automatic scoring thesleep stages. We use the physionet sleep eeg database to test and verify the accuracy of automaticclassification system, by comparing the manually scored hypnogram with automatic generatedresults. Because of its higher recognition rate, lower computing costs, higher speed and hardwareimplemented feasibility, the system could be applied for the real time implementation. The resultsin this thesis offer a methodological framework which could be used for the assistedcomputational sleep scoring systems designing.