Help Privacy Policy Disclaimer
  Advanced SearchBrowse




Journal Article

REG-ICA: A hybrid methodology combining Blind Source Separation and regression techniques for the rejection of ocular artifacts

There are no MPG-Authors in the publication available
External Resource
No external resources are shared
Fulltext (public)
There are no public fulltexts stored in PuRe
Supplementary Material (public)
There is no public supplementary material available

Klados, M., Papadelis, C., Braun, C., & Bamidis, P. D. (2011). REG-ICA: A hybrid methodology combining Blind Source Separation and regression techniques for the rejection of ocular artifacts. Biomedical Signal Processing and Control, 6(3), 291-300. doi:10.1016/j.bspc.2011.02.001.

Cite as: http://hdl.handle.net/11858/00-001M-0000-0024-C5D7-5
There are so far two main methodological approaches for rejecting ocular artifacts from electroencephalographic (EEG) and magnetoencephalographic (MEG) signals: regression- and Blind Source Separation (BSS)-based techniques, both having merits, as well as, some serious limitations. In this piece of work, a hybrid methodology that combines the main advantages of these two methods is proposed. We hypothesize that the artifactual independent components (ICs) extracted by a BSS method include more ocular and less cerebral activity than the contaminated EEG signals. We thus propose to apply a regression algorithm to the ICs rather than directly to the recorded signals. The analysis was carried out with synthetic mixtures of real EEG and electroocculographic (EOG) recordings. A BSS method, the extended INFOMAX version of Independent Component Analysis (ICA), was initially used to decompose the artificially contaminated EEG signals into spatiotemporal ICs. Then, a regression scheme, based on a stable version of the Recursive Least Squares algorithm, sRLS, was applied to the artifactual components in order to remove only the ocular artifacts, maintaining the underlying neural signals intact. The processed ICs were then projected back, reconstructing the artifact-free EEG signals. The performance of the proposed technique was compared with two automatic techniques; a regression technique based on Least Mean Square (LMS) algorithm and a BSS-based artifact rejection technique called wavelet-ICA (W-ICA) on the artificially contaminated data. For comparison, two metrics were used to assess the different methods’ performance: the first quantified how successful each technique was in removing the ocular artifacts from the EEG recordings, and the second one quantified how much each technique distorted the ongoing brain activity in both time and frequency domains. Confirming our main hypothesis, results have shown that the artifactual ICs contained more ocular and less cerebral activity (p < 0.04) (artifact to signal ratio (ASR) = 1.83 ± 3.65) in contrast to the contaminated electrode signals (ASR = 0.69 ± 3.40). Our results reveal that the proposed methodology, namely REG-ICA, removes the ocular artifacts more successfully than W-ICA (p < 0.01) or LMS (p < 0.01). It also distorts less the brain activity in the time domain when compared to W-ICA and LMS. In the frequency domain, it distorts the brain activity less than the W-ICA in all frequency bands, and less than the LMS for the delta, beta, and gamma bands. Our results suggest that the proposed methodology is evidently an attractive alternative to other already proposed artifact rejection methodologies.