Reference layer adaptive filtering (RLAF) for EEG artifact reduction in simultaneous EEG-fMRI

Abstract

In conclusion, the results suggests that RLAF and MBRLAF are potentially very effective in improving EEG quality of simultaneous EEG-fMRI. Highlights We present a new and reusable reference layer cap prototype for simultaneous EEG-fMRI We introduce new algorithms for reducing EEG artifacts due to simultaneous fMRI The algorithms combine a reference layer and adaptive filtering Several evaluation criteria suggest superior effectivity in terms of artifact reduction We demonstrate that physiological EEG components are preserved.

Exploring the relative efficacy of motion artefact correction techniques for EEG data acquired during simultaneous fMRI

“…Such overfitting would be particularly problematic in the case of simultaneous EEG-fMRI where single trial features of the EEG response, such as ERP amplitude (e.g., Debener et al, 2005;Eichele et al, 2005;Mayhew, Porcaro, et al, 2010) Previous studies, in which motion metrics were fitted to EEG scalp data, have shown that neuronal signals are recoverable Jorge et al, 2015;LeVan et al, 2013;Masterton et al, 2007;Maziero et al, 2016;Steyrl et al, 2017). Such overfitting would be particularly problematic in the case of simultaneous EEG-fMRI where single trial features of the EEG response, such as ERP amplitude (e.g., Debener et al, 2005;Eichele et al, 2005;Mayhew, Porcaro, et al, 2010) Previous studies, in which motion metrics were fitted to EEG scalp data, have shown that neuronal signals are recoverable Jorge et al, 2015;LeVan et al, 2013;Masterton et al, 2007;Maziero et al, 2016;Steyrl et al, 2017).…”

“…While considerable effort has been applied to removing the GA and PA via reduction of the strength of the artefacts produced during acquisition (e.g., Bonmassar et al, 2002;Chowdhury, Mullinger, & Bowtell, 2015;Chowdhury, Mullinger, Glover, & Bowtell, 2014;Jorge, Grouiller, Gruetter, van der Zwaag, & Figueiredo, 2015;LeVan et al, 2013;Luo, Huang, & Glover, 2014;Maziero et al, 2016;Mullinger, Yan, & Bowtell, 2011;Solana et al, 2014;Steyrl, Krausz, Koschutnig, Edlinger, & Muller-Putz, 2017) and application of post-processing methods (e.g., Abreu et al, 2016;Acharjee, Phlypo, Wu, Calhoun, & Adali, 2015;Allen, Josephs, & Turner, 2000;Allen, Polizzi, Krakow, Fish, & Lemieux, 1998;Bonmassar et al, 2002;Brookes, Mullinger, Stevenson, Morris, & Bowtell, 2008;De Munck, van Houdt, Goncalves, van Wegen, & Ossenblok, 2013;Iannotti, Pittau, Michel, Vulliemoz, & Grouiller, 2015;Krishnaswamy et al, 2016;Luo, Huang, & Glover, 2014;Niazy, Beckmann, Iannetti, Brady, & Smith, 2005;Xia, Ruan, & Cohen, 2014), until recently, little attention had been given to removing the MA. This is because it was thought that the identification of gross MAs, via data inspection, followed by removal of confounded data segments, produced EEG data of high enough quality to use in EEG-fMRI data analysis pipelines (Allen et al, 1998).…”

“…The signals measured from the electrodes on the reference layer are subtracted from the signals measured at the scalp electrodes to eliminate the artefacts . Using this system, they showed that least-mean squares adaptive filtering of the reference layer signals to the scalp layer produced superior performance to the simple subtraction used in the original RLAS implementation (Steyrl et al, 2017). Using this system, they showed that least-mean squares adaptive filtering of the reference layer signals to the scalp layer produced superior performance to the simple subtraction used in the original RLAS implementation (Steyrl et al, 2017).…”

“…The signals measured from the electrodes on the reference layer are subtracted from the signals measured at the scalp electrodes to eliminate the artefacts . This method has been extended by Steyrl et al (2017), who produced a double-layer cap in which the electrodes used to monitor motion are connected via a series of conductive tubes, rather than a continuous layer. Using this system, they showed that least-mean squares adaptive filtering of the reference layer signals to the scalp layer produced superior performance to the simple subtraction used in the original RLAS implementation (Steyrl et al, 2017).…”

“…Furthermore, the methods described above employ different algorithms for fitting the motion metrics to the EEG data. While it has been shown that underlying neuronal signals are present after MA correction using all methods, it is unclear whether over-fitting of the data is occurring, especially in the cases where adaptive filtering is employed (Jorge et al, 2015;Masterton et al, 2007;Steyrl et al, 2017). Such overfitting may attenuate the neuronal signals of interest.…”