Scattering convolutional network based predictive model for cognitive activity of brain using empirical wavelet decomposition. (April 2021)

• Record Type: Journal Article
• Authors: B., Lakshmi Priya; S., Jayalakshmy; Pragatheeswaran, Jayanthi K.; D., Saraswathi; N., Poonguzhali
• Is Part Of: Biomedical signal processing and control. Volume 66(2021)
• Publication Date: 2021-04
• Subjects: Brain decoding -- Electroencephalogram -- Magnetoencephalogram -- Long short term memory -- Bi-directional long short term memory -- Gated recurrent units; 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.2021.102501
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: Addressed the challenging issues of decoding cognitive activity of human brain from EEG and MEG signals. Signals are decomposed using Empirical Wavelet Transform (EWT) and dimensionality reduction is carried out using wavelet scattering network. Classification of wavelet scattering coefficients is performed using LSTM, BiLSTM and GRU and performance comparison is done. GRU based classifier recorded better performance in terms of classification accuracy, reduced time and computational complexities. Abstract: Understanding the cognitive activity of brain is a challenging task in brain computer interface (BCI) applications. This work aims at exploring the capability of empirical wavelet transform in decoding the brain wave pattern acquired in response to a thought process and visual stimuli. Empirical wavelet transform (EWT), when combined with the wavelet scattering coefficients is found to efficiently decode the brain wave using recurrent neural network (RNN) based classifier. Electroencephalogram (EEG) and magnetoencephalogram (MEG) are the two modalities considered in this work. The proposed framework is assessed using three different RNN architectures namely long short term memory (LSTM), bi-directional long short term memory (Bi-LSTM), gated recurrent units (GRU). The experimental results show that wavelet scattering coefficients extracted from the dominant mode of EWT decomposition record better performance of 90.23 % and 84.25 % for EEG and MEG signals using GRU as classifier. Furthermore, the wavelet scattering network which involves no learning process achieves better classification at reduced time and computational complexities.