Neuropathology Detector in EEG using Higher Order Statistics and Deep Learning

This article presents a neuropathology detector, based on the patient's electroencephalogram (EEG). Detection is based on HOSA ("High Order Statistical Analysis") image classification of higher order statistics derived from time series corresponding to EEG of human patients. The classifier is a DL model ("Deep Learning") with the pretrained CNN ("Convolutional Neural Network") architecture: Inception. The CNN training and test set are HOSA images of non-linear and non-Gaussian segments, of signals corresponding to the selected channel of the EEG of patients with neuropathologies (specifically, epilepsy) or healthy. The performance of the classifier is very satisfactory, presenting an accuracy of approximately 94% in the detection of epilepsy.