Automated Analysis of Long - Term EEG Signals

The automated classification of electroencephalographic (EEG) signals is a very difficult task, especially when these signals become large and noisy. A comprehensive methodology for automated analysis of multichannel EEG signals recorded during long-term monitoring (LTM) has been proposed. The methodology is based on multichannel adaptive segmentation, subsequent comprehensive feature extraction and automated classification of the segments by supervised learning or by cluster analysis. The greatest effort was focused on designing, implementing and making a successive comparison of an appropriate combination of preprocessing, data representation and visualization methods in order to overcome the disadvantages of existing approaches and to enhance the differentiation of individual neurological states. The structure of the methodology was optimized to EEG signal processing in the field of sleep, newborns, epilepsy and comatose studies, but it can also be applied to general EEG data (both clinical and non-clinical). The proposed methodology was verified on real clinical data acquired in neurological clinics. Some of the illustrative results are shown. In order to achieve increased precision and robustness, additional information was obtained from other physiological signals, such as electrocardiograms (ECG), electromyograms (EMG), electrooculograms (ECG), and respiratory signals (PNG).