PCA-ICA for automatic identification of critical events in continuous coma-EEG monitoring

Continuous monitoring of the electroencephalogram (EEG) provides information about the condition of the brain of a patient in coma and about the effects of therapy. For these reasons it is exploited to monitor cerebral events in coma patients. Declaring a patient ‘‘dead’’ is a very tricky procedure and it requires a long-time monitoring during which the doctor visually inspects the EEG tracings looking for any waves (critical events) that might account for restart of cerebral activity. Because of this, coma-EEG are very long and sometimes the huge size of data to be visually inspected makes the detection of critical events troublesome. A procedure for automatic critical events identification might support the doctor in the continuous monitoring of coma patients. This is why the performance of a technique, recently proposed by the authors, capable of automatically quantifying how much an EEG epoch is critical, is analysed in this paper. The technique is based on the extraction of descriptive components from overlapping EEG data segments (epochs) and on the extraction of some features from the components: entropy and kurtosis. This analysis was applied to a 3 h continuous coma-EEG and allowed for the detection of critical epochs that were worth being carefully inspected by the expert in order to ascertain whether they accounted for brain activity restart or not. Moreover, the step of descriptive components extraction was performed in three different ways: by Principal Component Analysis (PCA), by Independent Component Analysis (ICA) and the by the joint use of PCA–ICA. Finally the performance of the automatic detection were compared.