System theoretic analysis of electroencephalogram data for the early identification of epileptic seizures

This paper describes a new method of seizure detection that identifies seizures before their onset with sufficient time to warn the individual or execute actions to abort the seizure formation. Clinically classified segments of intracranial EEG data obtained from epileptic patients were separated into training and testing sets. After the powers in various frequency bands were calculated as the state-discriminating features for each EEG segment, a system theoretic model - the fuzzy measure theoretic quantum approximation of an abstract system (FMQAS) - was employed to develop a classifier from only the training set data. For this problem, FMQAS relied on defining a bipartite network. An iterative procedure combined with a fuzzy intersection attributed weights, or system measures, to the arcs connecting nodes in the two sets. The system measures were then used to identify a subset of features associated with each EEG state. Lastly, an FMQAS-based classifier was developed to individually analyze the features from short time segments of EEG. When applied to EEG segments in the testing set, the FMQAS-based classifier shows evidence of seizure activity in advance of visual EEG manifestation and also functions well as a seizure detector. Further, the classifier does not indicate signs of seizure activity during interictal periods greater than ten minutes before a seizure onset.