An object-oriented expert system for the identification of foci of epileptiform activity.

摘要

This dissertation describes the reasoning strategies and key issues involved in the implementation of a system for the reliable automated detection of foci of epileptiform activity in EEG recordings. Our scheme uses a signal processing front-end to detect candidate sharp transients (epileptiform events) and other events in the raw EEG, which are used subsequently by an expert system. In this research, we have designed and developed a system that employs extensive spatial and temporal context information for the detection of foci of epileptiform activity. Our system has been implemented using Objective-C under NEXTSTEP, and can operate both in a research environment and a clinical setting. This system has been extensively evaluated on recordings with and without epileptiform activity. Emphasis has been placed on the elimination of false detections and achieving detection rates that closely match a human expert. In data from thirteen subjects with epileptiform activity and five controls, of about 6.3 hr total duration, we have achieved detection accuracies ranging between 85% and 100% for the detection of epileptiform events with false detections ranging between 5% and 25%. All of the twenty foci found by the expert were correctly detected and localized by the system. Two additional false positive foci were reported by the system due to benign 14- and 6-Hz activity and generalized spike-wave activity which the system was not trained to discriminate.