@inproceedings{reichert_bold_2018,
	address = {San Diego, California, United States},
	title = {{BOLD} signal is more reliable than sensorimotor {EEG} signals in decoding hand movements. {Program} {No}. 225.27},
	abstract = {Sensorimotor rhythms (SMR) are often used to control a brain-computer interface (BCI). However, the degree of SMR modulation varies between participants, such that some people are not able to achieve control and hence are termed BCI illiterate. It has been hypothesized that BCI literacy is associated with an ability to recruit motor areas. Here we investigated the relationship between the detectability of hand movement in fMRI and in EEG recordings from the same participants to assess whether illiteracy is specific to participants or to the recording technique. The fMRI/EEG sessions followed a crossover design with 6 right-handed and 5 left-handed healthy participants. In intervals of 20s duration, participants moved their left hand, right hand or rested. EEG signals were recorded at 27 locations primarily covering motor areas. We classified 2s intervals of ongoing α/β activity (8-15Hz) using a CSP filter and a naïve Bayes classifier. Classification of fMRI data was performed in anatomical ROIs (AAL template) using principal component analysis and a naïve Bayes classifier. While classification accuracy of movement and rest based on EEG data showed an SMR-typical heterogeneous distribution (mean = 81.9\%; 59.1-98.4\% SD = 14.4\%), the BOLD signal of single scans in the contralateral precentral gyrus was classified with accuracies above 89\% (mean = 93.9\%; SD = 2.2\%) in all participants. Further ROIs yielding high accuracies were postcentral gyrus (93.4\%), SMA (90.6\%), cingulum (88.9\%), paracentral lobule (86.1\%), Rolandic operculum (88.7\%); ipsilateral cerebellum (91.9\%) as well as vermis (91.1\%). Accuracy could not be explained by handedness in either EEG or fMRI classification. Importantly, fMRI and EEG classifier accuracy did not correlate significantly ({\textbackslash}textbackslashtextbarr{\textbackslash}textbackslashtextbar{\textbackslash}textbackslashtextless0.4, p{\textbackslash}textbackslashtextgreater0.05). However, in contrast to previous studies of motor imagery (MI), the EEG decoding accuracy in most ROIs correlated negatively with the BOLD signal, with significant correlations in ipsilateral SMA (r=-0.43) and cerebellum (r=-0.49), in contralateral SMA (r=-0.44) and cingulum (r=-0.43) as well as in the vermis (r=-0.48). Our finding of a higher, more homogeneous classification rate using fMRI suggests that the lack of SMR discrimination in EEG-based BCIs is not due to a lower ability of individual users to recruit motor areas as postulated based on MI studies. In participants with low SMR modulation, high γ activity, which is not detectable with surface EEG but is reflected in BOLD recordings, might be more prominent. Our study supports the notion that noninvasive SMR-based BCIs are applicable only to a limited group of users, but access to high γ activity might extend the user group.},
	booktitle = {Neuroscience {Meeting} {Planner}},
	author = {Reichert, Christoph and Dürschmid, Stefan and Hinrichs, Hermann and Heinze, Hans-Jochen and Sweeney-Reed, Catherine M.},
	year = {2018}
}