Separation of movement direction concepts based on independent component analysis algorithm, linear discriminant analysis, deep belief network, artificial and fuzzy neural networks. (September 2020)

• Record Type: Journal Article
• Authors: Sabzevari, Mohammadamin; Imani, Ehsan
• Is Part Of: Biomedical signal processing and control. Volume 62(2020)
• Publication Date: 2020-09
• Subjects: Brain signals -- Independent component analysis -- Linear discriminant analysis -- Artificial neural network -- Fuzzy neural network -- Deep belief network; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2020.101950
• Languages: English
• ISSNs: 1746-8094

Abstract

Abstract: Brain signals have various scientific and practical applications, such as Medical Science, Cognitive Science, Neuroscience, and Brain Computer Interfaces. Brain signal analysis is faced with complex challenges including small sample size, high dimensionality and noisy signals. Because of the non-stationarity of brain signals and the impacts of mental states on brain function, brain signals are associated with an inherent uncertainty. In this study, it is tried to present a plausible method for detecting and distinguishing the directions from EEG signals. Recording single-polarized signals was carried out utilizing a 19-channel cap Micromed device with the use of Cz as reference electrode. The statistical population used involved ten 25–35 year old male volunteers. The designed task consisted of 24 slides of up, down, left and right directions. After preprocessing level, ICA algorithm was employed to extract artifacts, to decrease signal dimension and to determine the target signal. In feature extraction section, AR coefficients extracted to feed the ANN, FNN and LDA. Data for Deep Belief Network provided from Autoregressive power spectral density estimate with order of 20 employed on the data set. Classifiers' results reveal that 2.5 s time window leads to the best separation accuracy. DBN surprisingly leads to the highest level of accuracy in comparison to the other proposed methods. Based on the 10-fold Cross Validation, the performance of the classifiers measured in terms of accuracy, sensitivity and specificity. The obtained average accuracies for LDA, ANN, FNN and DBN respectively are 61.86 ± 1.69 %, 57.18 ± 1.88 %, 66.79 ± 2.14 % and 91.06 ± 0.68.