Abstracts

Depending on the vigilance state different brain areas are active. Normally, an exact localization is possible only from short (few ms) averaged segments by means of computational demanding methods such as dipole fitting. In contrast to this, we used principle component analysis (PCA) and independent component analysis (ICA) to approximate the global source distribution, i.e., the origins of the postsynaptic potentials, from stationary segments (20 seconds) of the EEG. We analyzed intracranial EEG recordings of six patients with mesial temporal lobe epilepsy. The obtained map of the source landscape is then assigned to a distinct vigilance state. We investigated whether a constant and recurrent distribution of the sources in the hippocampus can be found for the same vigilance state. Furthermore, we tried to identify differences between the focal and non-focal side. We used PCA and ICA to transform a multi-channel recording, e.g., EEG, into principle and most independent components, respectively. The matrix obtained from this transform contains the information about the source distribution. Subsequently we derived a similarity measure which shows the change of this transform matrix over time and therefore reveals the temporal activations of the source. We calculated these similarity values between all segments (from different times) resulting in an image which exhibits typically a chess pattern. Edges of the squares in these images correspond to a change in the vigilance state. We applied the algorithms on intracranial EEG of six patients with unilateral mesial temporal lobe epilepsy which were recorded continuously during night. A sleep stage classification was carried out by an expert EEG reader using a simultaneously recorded surface EEG. In the obtained images we found:
a.) pronounced blocks (many segments) with constant source distributions
b.) beginning and ending of these blocks correspond to a change of the sleep stage
c.) same sleep stages show the same source distributions during the whole night
As for the comparison of PCA and ICA we obtained similar results for both techniques. However, because ICA is a more sophisticated approach the data must fulfil also more assumptions what is not always provided. So PCA is normally preferred. Our method provides a very fast and robust way to identify changes of the vigilance state from the analysis of multi-channel EEG. For all patients we found stationary and exactly recurrent source distributions which were strongly correlated with the sleep stages obtained from surface EEG. This means that always the same sources are active in the same sleep stages. Furthermore, we found predominately stronger correlation of our results with the sleep stages classification for the left hemisphere. Aspects of further investigations should concern the identification of vigilance stages during the day and the possible differences between focal and non-focal hemisphere.