The utility of intracranial EEG feature and channel synergy for evaluating the spatial and temporal behavior of seizure precursors.

摘要

Individuals with epilepsy experience numerous limitations resulting from the unpredictability of their seizures including cognitive and behavioral impairments and risk of injury. There is currently an explosion of interest in predicting epileptic seizures from intracranial electroencephalographic signals that has its roots in experimental and theoretical work first published in the 1970s. Despite over 40 years of investigation into the physiology of epilepsy, it still is not possible to explain how and over what time spontaneous clinical seizures emerge from the relatively normal brain state observed between them. Reliable technologies to warn individuals of a potential seizure and ultimately to trigger intervention to eliminate seizures completely are desperately needed.; The objective of this research was to develop and apply efficient algorithms to multiple channels of intracranial EEG and multiple features extracted from these signals to provide a benchmark for identifying information capable of predicting seizures, thereby demonstrating the utility of feature and spatial synergy for evaluating seizure precursors in patients with mesial temporal lobe epilepsy. Short-term seizure precursors, minutes prior to seizure onset were investigated.; The methodology developed for this study employed preprocessing, three levels of feature extraction, a hybridization of genetic and classifier-based feature selection, classification and validation. These steps were necessary to identify patient-specific features that could identify characteristics common to many different individuals, eventually limiting the search space to a relatively small group of first level or derived features.; An average probability of prediction or block sensitivity of 62.5% was achieved with an average block false positive rate of 0.2775 false positive predictions per hour, corresponding to 90.47% specificity. The average probability of correct classification achieved was 91% with an 8% probability of false alarm. The current available technology used to control epileptic seizures is an open loop system that provides adjunctive therapy by applying stimulation on average, for 30 seconds every five minutes, which translates into less than 41% efficacy. If a system could be developed to predict over 60% of seizures at the level of performance reported in this work, dramatic improvement would be realized.