The electroencephalogram (EEG) is mostly used as a diagnostic tool for
relatively rare brain diseases, e.g. epilepsy. In this master's thesis, we focused
on the clustering and analysis of EEG microstates, which are recurring
topographies of electric potentials on the surface of the head. Our focus
was on the problem of diagnosing early-onset Alzheimer's disease, the most
common type of dementia that affects 5-10% of people older than 65 years.
We analyzed three clustering algorithms. Based on the acquired microstates,
we analyzed the frequency of EEG microstate occurrences and observed significant
differences in the frequencies of certain microstates between healthy
subjects and subjects with dementia. We then used these microstates in
a classification. Using a 10-fold cross-validation approach, we assessed the
accuracy of eight simple classifiers that determine, based on the statistical
frequency of EEG microstate occurrences in the EEG recording, whether a
subject is healthy or has probably dementia. The most accurate classifier
was trained using the support vector machine method based on microstates
acquired using the SCIMS method, with an AUC of 0.76.
|