Quality and denoising in real-time fMRI neurofeedback: a methods review

Abstract

<p>Neurofeedback training using real-time functional magnetic resonance imaging (rtfMRI-NF) allows subjects voluntary control of localized and distributed brain activity. It has sparked increased interest as a promising non-invasive treatment option in neuropsychiatric and neurocognitive disorders, although its efficacy and clinical significance are yet to be determined. Maximization of neurofeedback learning effects in accordance with operant conditioning requires the feedback signal to be closely contingent on real brain activity, which necessitates the use of effective real-time fMRI denoising methods to prevent sham feedback. In this work, we present the first extensive review of acquisition, data processing and quality reporting methods available to improve the quality of the rtfMRI neurofeedback signal. Furthermore, we investigated the state of denoising and quality control practices in a set of 128 recently published rtfMRI-NF studies. We found: (i) that less than a third of the studies reported implementing standard real-time fMRI denoising steps; (ii) significant room for improvement with regards to methods reporting; and (iii) the need for methodological studies quantifying and comparing the contribution of denoising steps to the quality of the neurofeedback signal. Advances in the field of rtfMRI-NF research depend on reproducibility of methods and results. To this end, we recommend that future rtfMRI-NF studies: (i) report implementation of a set of standard real-time fMRI denoising steps according to a proposed COBIDAS-style checklist (https://osf.io/kjwhf/); (ii) ensure the quality of the neurofeedback signal by calculating and reporting community-informed quality metrics and applying offline control checks; and (iii) strive to adopt transparent principles in the form of methods and data sharing and the support of open-source rtfMRI-NF software.</p>

Graded fMRI Neurofeedback Training of Motor Imagery in Middle Cerebral Artery Stroke Patients: A Preregistered Proof-of-Concept Study

“…(2) Another factor that compromised the data was head motion. Although head motion poses a major challenge for fMRI research with stroke patients (Siegel et al, 2017 ) and fMRI-neurofeedback more generally (Heunis et al, 2020 ), several previous fMRI studies that investigated motor imagery stroke patients have not reported criteria to control for it (Sharma et al, 2009b ; Sharma and Baron, 2013 ; Liew et al, 2015 ). The present study employed rigorous criteria informed by previous recommendations (Power et al, 2012 , 2014 ; Siegel et al, 2017 ) leading to a substantial discarding of data (an increase by the factor of nearly 10 compared to a similar study conducted in healthy participants) and one patient from the analysis.…”

“…Albeit promising, both approaches are highly susceptible to spurious correlations resulting from excessive head motion. Successful implementation in neurological populations will thus afford using reliable real-time head motion and artifact correction methods (Heunis et al, 2019 , 2020 ).…”

Connectome-wide search for functional connectivity locus associated with pathological rumination as a target for real-time fMRI neurofeedback intervention

“…train the participants to regulate the target connectivity. As a matter of fact, many rtfMRI-nf studies demonstrated successful self-regulation training even without a comprehensive real-time noise reduction process (Heunis et al, 2018), and the correlation between the online-and offlinecalculated signals was not high when the real-time noise reduction process was not comparable to the offline one (Misaki and Bodurka, 2019). When we performed a simulation with only the motion correction in real-time processing -which is a conventional rtfMRI process used in many studies -the correlation between the online and offline FC was less than 0.5 even for the highest online FC neurofeedback signal (10-TR sliding-window).…”

Simultaneous EEG-fMRI during a neurofeedback task, a brain imaging dataset for multimodal data integration

“…Real-time fMRI pre-processing (slice-time and motion correction) was performed directly by the NF control unit using a custom Matlab script based on SPM 8 (FIL, Wellcome Trust Centre for Neuroimaging, UCL, London, UK). More details, according to the COBIDAS-inspired template 21 , are reported in Table 1. XP1. Data collected during the motorloc session were used to identify a ROI for fMRI NF computation.…”

“…(a) Time-Frequency maps showing ERD (red) and ERS (blue) in the contralateral motor electrode and the corresponding ERD time series averaged across all frequencies (mean + standard error across subjects). (b) ERD topographic maps in the alpha[8][9][10][11][12] Hz and beta[13][14][15][16][17][18][19][20][21][22][23][24][25][26][27][28][29][30] Hz frequency bands. (c) ERD cortical maps.…”