Automated detection of sleep spindles

Görür, Dilan
Sleep spindles are one of the rhythmic activities observed in sleep electroencephalogram (EEG). As they are well defined and functional, sleep spindle analysis is significant for brain research. Identifying the characteristics of sleep spindles may lead to an understanding of the functions of sleep. Furthermore, understanding the sleep spindle generation mechanisms can explain the other rhythmical activity occurring in other brain regions. The detection process of the sleep spindle data of a whole night sleep EEG prepared by an expert would be too time consuming and it may be not objective for some spindle regions. Therefore, an automated detection system would assist the expert. In this thesis, a system for automated detection of sleep spindles in the EEG has been developed and tested on the recorded data of normal and insomniac subjects. Different methods for the automated detection of sleep spindles in EEG recordings are investigated. Features from the data are extracted by using two different approaches, short time Fourier transform (STFT) and autoregressive (AR) modeling. Multilayer perceptron (MLP) and also support vector machines (SVM) are utilized as classifiers for comparison. The best classification performances of MLP are found to be 97.5% and 93.6% for STFT and AR model features respectively. The best performances of SVM are found to be 97.5% for STFT and 94.4% for AR model. It is demonstrated that the classifiers trained by a healthy subject's EEG could perform well on another healthy subject's EEG but poorly on an insomniac subject's
Citation Formats
D. Görür, “Automated detection of sleep spindles,” M.S. - Master of Science, Middle East Technical University, 2003.