Trainable Early Warning System for Epileptic Seizures

DESCRIPTION (provided by applicant): Individuals with epilepsy, their families and the professionals that treat them have long felt the need for reliable warning of an impending epileptic seizure. In Phase I, we developed a unique, multi-parameter, individualized, trainable seizure predetection system which extracts spectral, wavelet and time domain features and uses recurrent neural networks to analyze many hours of multichannel EEG and predetect seizure onsets with a high level of accuracy and reliability. The underlying analytical software for this system was tested on large database of scalp EEG recordings from five epilepsy centers and contained over 373 hours of recordings from 17 patients containing 50 seizures. This is a large database compared to those reported by most investigators in this field. Phase I results were very accurate and reliable (sensitivity: 100%, false positive rate 0.02/hr and detection times that occurred an average of 9.1 seconds, before seizure onset). This is entirely acceptable for our primary target application, of warning of a seizure onset to enable timely diagnostic injection to locate the brain region where the seizure starts using SPECT imaging. In Phase II we will refine the detection process, validate performance in a large data set, implement the software in C++ modules, and test the commercial prototype in a clinical setting. These steps will result in a company supported Phase III product development, integration into existing EEG recording systems, beta testing and commercialization. While Phase 2 will concentrate on the SPECT application using scalp electrodes, the scalp EEG techniques developed will be applicable without change for general use in an epilepsy monitoring facility to alert staff of impending seizures and allow them to attend to patient safety in a timely manner. Predetection will also facilitate neuropsychological or other testing in the epilepsy monitoring environment. With additional work, the techniques developed can be applied to ambulatory devices for seizure alert, or for episodic vagal nerve stimulation to block seizures.