Frequency domain analysis of sleep EEG for visualization and automated state detection

Conventional analysis of EEG signals for sleep scoring is based on the time domain assessment of wave patterns. Human experts carry out this task relying on the direct visualization of EEG epochs. Techniques that enhance an intuitive visualization may encourage a wider use of more abstract descriptors, such as frequency domain features. This paper presents a feature extraction method for EEG signals based on FFT and principal component analysis. The result of the method is a characterization of EEG epochs with only two variables. Density plots of this 2D projection show compact clusters that correspond to sleep behavioral states. The distance to the centroid of a cluster is a reliable scoring criterion which is both easy to visualize and easy to automate. The techniques presented here have been shown to work reliably for both human and rat sleep studies.