OPTIMIZED RECURRENT CONDITIONAL WASSERSTEIN GENERATIVE ADVERSARIAL NETWORK–FOSTERED SCHIZOPHRENIA DETECTION IN EEG SIGNAL

Schizophrenia is a serious mental disease that can lead to lifelong disability. EEG-dependent schizophrenia diagnosis mostly relies on custom or hand-crafted feature extraction methods. But, the typical manual feature extraction techniques are time-consuming, prone to inaccuracies, and struggle to achieve a balance between accuracy and efficiency. To overcome these complications, this paper proposes an Optimized Recurrent Conditional Wasserstein Generative Adversarial Network–fostered Schizophrenia Detection in EEG Signal (RCWGAN-SD-EEG). Here, the input data are taken from the EEG-Schizophrenia dataset. Then, the collected data are fed to the pre-processing stage. The input signal is pre-processed by a Multi-window Savitzky–Golay Filter (MSGF) approach to remove noise and artifacts from signals. The preprocessed signal is fed into the feature extraction process. The feature extraction is done by High-order Time-reassigned Synchrosqueezing Transform (HTST) to extract EEG features, such as Fp2, F8, T4, T6, O2, Fp1, F7, T3, T5, and O1. The extracted features are given to the Recurrent Conditional Wasserstein Generative Adversarial Network (RCWGAN) for detecting and classifying schizophrenia disease as healthy and schizophrenia. Finally, the Leaf in Wind Optimization Algorithm (LWOA) is employed to improve the RCWGAN weight parameters that classify the type of schizophrenia exactly. The performance metrics, like accuracy, precision, specificity, computation time, and RoC are considered. The RCWGAN-SD-EEG technique is implemented in MATLAB. The proposed approach achieves 26.36%, 20.69%, and 30.29% better accuracy, 19.12%, 28.32%, and 27.84% better precision, 12.04%, 13.45%, and 22.80% better recall when compared to the existing techniques: Exploring deep residual network dependent features for automatic schizophrenia diagnosis from EEG (DRN-ASD-EEG), Schizophrenia classification utilizing ML on resting state EEG signal (DNN-SC-EEGS), a 3-dimensional convolutional neural network for schizophrenia diagnosis utilizing EEG-dependent functional brain network (3DCNN-SD-EEG) respectively.