Nonparametric On-Line Detection of Changes in Signal Spectral Characteristics for Early Prediction of Epilepsy Seizure Onset

The present study introduces the method for solving the problem on early prediction of epilepsy seizure onset based on analysis of multi-channel electroencephalogram (EEG). This problem is considered as the problem of on-line detection of multiple abrupt changes in spectral characteristics of the process under consideration. With EEG characteristics not being changed abruptly, to describe growing changes the use was made of multiple disharmony model. The quantity to characterize the degree of spectral instability is applied as a detector. The nonparametric sequential method of detecting disharmony is realized in computational algorithms effective enough to be used in real time for 32-channel EEG and to open possibilities for creating the system of automatic prediction