Automatic identification of schizophrenia using EEG signals based on discrete wavelet transform and RLNDiP technique with ANN

• Published in: Expert systems with applications Vol. 192; p. 116230
• Main Authors: Sairamya, N.J., Subathra, M.S.P., Thomas George, S.
• Format: Journal Article
• Language: English
• Published: New York Elsevier Ltd 15.04.2022; Elsevier BV
• Subjects: Artificial neural network (ANN); Artificial neural networks; Brain; Computation; Diagnosis; Discrete Wavelet Transform; Discrete wavelet transform (DWT); Electroencephalogram (EEG); Electroencephalography; Feature extraction; Histograms; Relaxed local neighbour difference pattern (RLNDiP); Rhythm; Schizophrenia; Schizophrenia (ScZ); Wavelet transforms; Discrete wavelet transform (DWT); Schizophrenia (ScZ); Artificial neural network (ANN); Electroencephalogram (EEG); Relaxed local neighbour difference pattern (RLNDiP)
• DOI: 10.1016/j.eswa.2021.116230
• ISSN: 0957-4174

•Novel RLNDip technique for automatic identification of schizophrenia using EEG signals is proposed.•A fusion approach of DWT with RLNDiP technique is introduced in this work.•Analysis of EEG signals in different brain rhythms is evaluated.•Obtained results conclude that alpha rhythm achieved a better classification performance using ANN. Schizophrenia (ScZ) is a detrimental condition of the brain often associated with depression, anxiety, and socio-psychological problems. In the traditional diagnosis approach, the results are subjective, prone to error, and biased, as they solely depend on the subject’s response and the psychiatrist's experience. Hence, in this work, to overcome the aforesaid problems a computer-aided diagnosis of ScZ from the electroencephalogram (EEG) signals using the novel relaxed local neighbour difference pattern (RLNDiP) technique is proposed. To seize the entire characteristics of disrupted connectivity in ScZ, the combination of RLNDiP features from both time domain (TD) and time–frequency domain (TFD) is proposed. In the TD, the proposed technique is employed to transform the EEG signals into the RLNDiP domain, by computing the RLNDiP code for each sample in the EEG signals. Secondly, the histogram features are computed from the RLNDiP domain. In the TFD, the discrete wavelet transform is used to decompose the signals into five brain rhythms, namely delta, theta, alpha, beta, and gamma. In the next step, each brain rhythm is converted into the RLNDiP domain, and the histogram features are computed. The features extracted from different brain rhythms and the TD features are integrated using various fusion approaches for accurate discrimination of ScZ from normal subjects. The prominent features describing the effective connectivity is selected using the Kruskal-Wallis test (p < 0.05) and the selected features are fed into the artificial neural network (ANN), for automatic diagnosis of ScZ. The proposed approach attained a maximum accuracy of 100%, with the fusion of alpha brain rhythm and the TD features. Compared to the state-of-the-art methods, the proposed approach attained a maximum classification performance.