EEG-Based Automated Detection of Schizophrenia Using Long Short-Term Memory (LSTM) Network

Abstract

Schizophrenia, a severe long-term mental disorder, can be found in every society and culture all around the globe. This paper focuses on detection of schizophrenia employing long short-term memory (LSTM), a deep learning technique, by extracting features from the EEG signal. The nonlinear features such as Katz fractal dimension (KFD) and approximate entropy (ApEn), along with the time-domain measure of variance values, are calculated from the EEG signal. An LSTM architecture with four hidden LSTM layers having 32 neurons in each layer is used for the detection. A sample of 6790 feature vectors was calculated from EEG signals of 14 schizophrenia patients and 14 healthy controls. The EEG signals from the subjects were recorded under resting state of eyes-closed condition. The model was trained with 6000 feature vectors and tested with the other 790 vectors. The LSTM model could accurately distinguish people suffering from schizophrenia from the controls with an accuracy of 99.0{\%}. The accuracy of the model validates the detection efficiency of the LSTM network over the performance of the machine learning methods of feedforward neural network (FFNN) and support vector machine (SVM).