A computerized instrument is proposed for automated spike and seizure detection. The objective is to use novel methods to increase the effectiveness of monitoring, and to increase the quality and reduce the cost of epilepsy diagnostics. The proposed spike detector will utilize a combination of methods. These will include parametric description of the EEG and its Laplacian derivatives, an artificial neural network (ANN) producing likelihood estimates for each event of being a spike, rules to select candidate events for the ANN, further rules to select events based on the ANN output, and spike localization. A graphical user interface will allow the investigator to review results quickly and to retrain the ANN according to specific needs. In Phase I, software for the proposed detector will be developed along with tools to measure the effect of detection parameters on the sensitivity and selectivity of the system. A database of spikes and non-epileptic EEG events will be established to train and test the detector, to demonstrate its performance, and to optimize detection parameters. Phase II studies will employ a larger data base and will include seizure as well as interictal and non- epileptiform EEG.
