Abstract
Machine learning has been densely used in most computer-aided medical diagnosis systems. These systems not only supported the physician's decision but also accelerate the necessitated procedures. Electroencephalography (EEG) is an essential device for measuring the brain's electrical activities. EEG is used to detect a series of brain disorders such as epilepsy, dementia, Parkinson's disease, and Schizophrenia (SZ). In this work, a novel method for detecting SZ using EEG recordings is suggested. Initially, the presented technique breaks down each channel of the input EEG recordings into EEG rhythms. The wavelet transform is employed to achieve this. The 1D local binary pattern (LBP) is then used to code the acquired rhythm signals. Each row of the input picture is formed by concatenating the uniform histograms of the 1D LBP coded beats. The rows of the images are formed from the channels of the input EEG signal, while the columns of the images are constructed from the rhythms. Extreme learning machines (ELM) based autoencoders (AE) are utilized at a data augmentation step. After data augmentation, the SZ and healthy cases are classified using well-known deep transfer learning. Deep transfer learning employs a variety of pre-trained deep Convolutional Neural Network (CNN) models. Various performance assessment indicators are used to evaluate the produced outcomes. An EEG dataset that Lomonosov Moscow State University released is used in experiments, and a 97.7% accuracy score is obtained. The obtained results are also compared with several recently published methods. The comparisons show that the proposed method outperforms the compared methods.