Auto-encoder Based Automated Epilepsy Diagnosis

Abstract

This work proposes the design of an automated system based on Auto-encoders to detect epilepsy using Electroencephalogram (EEG) signal. Harmonic Wavelet Packet Transform (HWPT) and Fractal Analysis are employed to extract the initial feature vector. HWPT is used to capture spectral information using non-overlapping octave bands, thereby avoiding recursive calculations. Fractal dimension (FD) based on Katz technique is performed on windowed segments of the signal to capture spatial information of the EEG signal. Features comprising HWPT and FD are fed to an Auto-encoder, which is used to compress the high dimensional feature vector to a more discriminative and lower dimensional feature vector, for efficient classification. Finally, a Softmax classifier is used for the binary classification problem of discriminating epilepsy signals from normal cases.