Efficient Classification of Schizophrenia EEG Signals Using Deep Learning Methods

Schizophrenia, a serious mental disorder, manifests as hallucinations, delusions, and cognitive deprivations. Timely detection and intervention helps the person overcome the distress in leading a healthy life and could be handheld to recovery soon. This work is an attempt to evaluate the deep learning methods of VGG-16 and AlexNet over long short-term memory (LSTM) network in detecting schizophrenia efficiently as deep learning methods are able to identify the subtle patterns hidden in the data. VGG-16, a convolutional neural network (CNN) architecture, is widely used in many learning applications due to the ease with which it can be implemented, while AlexNet, also being a CNN, could learn the representation of features from the data. The results of VGG-16 and AlexNet are compared with that of the classification efficiency of LSTM in detecting schizophrenia. The resting state EEG signals obtained from a publicly available database were sampled at a frequency of 250?Hz. Detection of schizophrenia using LSTM was carried out by extracting nonlinear features such as Katz Fractal Dimension (KFD), approximate entropy (ApEn) along with the statistical time domain measure of variance, from the acquired EEG signals. The LSTM model gave an accuracy of 99% in classifying schizophrenia from healthy controls, while 99.81% and 99.61% were obtained for VGG-16 and AlexNet, respectively. The CNN models outperform the classification efficiency of the feature input LSTM model in detecting schizophrenia, which could be of assistance to the clinicians for an expeditious diagnosis.