A modified row-sparse multiple measurement vector recovery algorithm for reconstructing multichannel EEG signals from compressive measurements. (July 2020)

• Record Type: Journal Article
• Authors: Aghababaei, Mohammad H.; Azemi, Ghasem
• Is Part Of: Biomedical signal processing and control. Volume 60(2020)
• Publication Date: 2020-07
• Subjects: EEG tele-monitoring -- WBAN -- Compressed sensing -- Inter-channel correlation -- Row-sparsity -- DCT basis; 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.2020.101956
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: The problem of efficiently recovering multichannel EEG signals from their compressive measurements is addressed. The algorithm exploits sub-matrices of a given multichannel EEG matrix for which the assumption of row-sparsity is satisfied. The results show significant improvements of up to 4% in the reconstruction accuracy measured by the normalized mean squared error. The main limitation of the proposed recovery method is the need for an uncompressed multichannel EEG epoch to calculate the optimum threshold for the inter-channel correlation. The proposed method can be deployed in wireless body area network based EEG monitoring systems. Abstract: In this paper, a new method for reconstructing multichannel EEG signals from their compressive measurements is proposed which exploits the inter-channel correlation. Such correlation is used in the row-sparse multiple measurement vector (RSMMV) recovery approach to improve its performance in reconstructing multichannel EEGs. In this approach, it is assumed that the multichannel EEG signal ensemble is a row-sparse matrix in some domain such as wavelet, discrete cosine transform (DCT), or Gabor. However, this assumption does not hold in practice for real EEG signals. In order to overcome this limitation, a modified RSMMV algorithm is proposed which exploits sub-matrices of a given EEG matrix for which the assumption of row-sparsity is satisfied. The proposed method is applied to 2 multichannel EEG datasets chosen from the BCI competition III and the OSF databases and its performance is evaluated and compared with that of the RSMMV recovery approach; which has been proven to be superior to well-known single-channel recovery methods. Experimental results show that, in comparison with the RSMMV recovery approach, the proposed method achieves improvements of up to 4% and 2% in the reconstruction accuracy measured by the normalized mean squared error in the BCI and the OSF datasets respectively. The results also show that the proposed algorithm outperforms the blind compressed sensing method and has a comparable performance with the low-rank approach. The proposed method can therefore be deployed in wireless body area network based EEG monitoring systems.