Self-modulation of motor cortex activity after stroke: a randomized controlled trial

Abstract

Real-time functional MRI neurofeedback allows individuals to self-modulate their ongoing brain activity. This may be a useful tool in clinical disorders that are associated with altered brain activity patterns. Motor impairment after stroke has previously been associated with decreased laterality of motor cortex activity. Here we examined whether chronic stroke survivors were able to use real-time fMRI neurofeedback to increase laterality of motor cortex activity and assessed effects on motor performance and on brain structure and function. We carried out a randomized, double-blind, sham-controlled trial (ClinicalTrials.gov: NCT03775915) in which 24 chronic stroke survivors with mild to moderate upper limb impairment experienced three training days of either Real (n = 12) or Sham (n = 12) neurofeedback. Assessments of brain structure, brain function and measures of upper-limb function were carried out before and 1 week after neurofeedback training. Additionally, measures of upper-limb function were repeated 1 month after neurofeedback training. Primary outcome measures were (i) changes in lateralization of motor cortex activity during movements of the stroke-affected hand throughout neurofeedback training days; and (ii) changes in motor performance of the affected limb on the Jebsen Taylor Test (JTT). Stroke survivors were able to use Real neurofeedback to increase laterality of motor cortex activity within (P = 0.019), but not across, training days. There was no group effect on the primary behavioural outcome measure, which was average JTT performance across all subtasks (P = 0.116). Secondary analysis found improvements in the performance of the gross motor subtasks of the JTT in the Real neurofeedback group compared to Sham (P = 0.010). However, there were no improvements on the Action Research Arm Test or the Upper Extremity Fugl-Meyer score (both P > 0.5). Additionally, decreased white-matter asymmetry of the corticospinal tracts was detected 1 week after neurofeedback training (P = 0.008), indicating that the tracts become more similar with Real neurofeedback. Changes in the affected corticospinal tract were positively correlated with participants neurofeedback performance (P = 0.002). Therefore, here we demonstrate that chronic stroke survivors are able to use functional MRI neurofeedback to self-modulate motor cortex activity in comparison to a Sham control, and that training is associated with improvements in gross hand motor performance and with white matter structural changes.

Keywords: motor cortex; neurofeedback; real-time fMRI; stroke; white matter.

© The Author(s) 2022. Published by Oxford University Press on behalf of the Guarantors of Brain.

Figures

Figure 1

Study design and timeline. (A) Participants attended three NF training sessions separated by 24 and 48 h, a baseline session and two follow-up sessions at 1 week and 1 month post NF training. (B, Left) Regions of interest used for NF training were defined during a functional localizer. Regions of interest were centred on peak activity in M1 during movements of both hands. (B, Right) During NF training, only the stroke-affected hand was moved and participants viewed two bars on the screen. The red bar represented activity in the stroke-affected hemisphere and the blue bar represented activity in the unaffected hemisphere. During movement blocks, participants were instructed to make movements with their stroke-affected hand to increase the size of the red bar, while keeping the blue bar as low as possible.

Figure 2

Laterality of motor cortex activity is increased within, but not across NF training days. (A) Laterality of M1 activity (LI) is displayed for the Real (green/left, n = 12) and Sham (grey/right, n = 12) group over all the training runs and days. Bars represent group means, grey lines show individual participant data and error bars represent SEM. (B, Top) When considering NF learning ‘within' training days by averaging LI for each run (pooled over training days), a significant interaction could be observed between Run and Group (P = 0.037), with the Real group increasing LI over runs, while the Sham group did not change. Estimated marginal means and confidence intervals are shown for each run. (B, Bottom) When assessing NF learning across training days by averaging LI for each day (pooled over runs), there were no significant interaction or main effects. See also Supplementary Table 4. (C) There was no group-level transfer effect; participants in both groups have similar change in LI on the transfer runs (data displayed as in A, see also Supplementary Table 5). (D) However, the relationship between transfer change and NF success differs between the groups (P = 0.002), with a positive relationship found for the Real group, and a negative one for the Sham group. Each dot represents a participant, baseline LI is regressed out of the average transfer change score; therefore, values are centred around zero. For raw values see Supplementary Table 5. There are missing data for one participant in each group due to motion artefacts (see Supplementary material). *P < 0.05.

Figure 3

Participants in Real group improve on gross motor performance (JTT) compared with Sham. (A) Participants in both the Real (green) and the Sham (grey) group got faster at performing the JTT over time when all subtasks were considered (P = 0.017) and no group differences were found. Bars represent mean time to complete all subtasks (log), error bars represent SEM and data from individual participants are shown by the grey dotted lines. (B) When comparing performance on the gross motor subtasks of the JTT, the Real group had faster performance compared to the Sham group (P = 0.01). Estimated marginal means are shown with 95% confidence intervals at each follow-up time point. *P < 0.05. (C) The same data are shown for fine motor subtasks of the JTT where no significant group differences were found (see also Supplementary Table 6). There was no group effect on the ARAT (D) or the UE-FM (E) at 1 week or 1 month after NF training. Data displayed as in (A); see also Supplementary Table 7.

Figure 4

FA Asymmetry of CSTs is reduced after Real NF training. (A) The Real group (n = 10) had lower FA asymmetry in the CSTs (inset) after NF training compared to the Sham group (n = 11). Change in FA asymmetry between baseline and 1-week follow-up is plotted for the Real and the Sham group. The black line represents median change, coloured boxes represent 95% confidence intervals and individual participant data points are shown with open circles. (B) The same information is shown for FA change between baseline and 1-week follow-up in the affected (top) and unaffected (bottom) CSTs. (See also Supplementary Table 8.) (C) Neurofeedback success was positively correlated with FA change in the stroke-affected hemisphere in the Real group, whereas no correlation was found in the Sham group with the two correlations being significantly different from each other (P = 0.004). *P < 0.05.

Figure 5

Increased activity in brain regions associated with NF learning. Changes in brain activity before and after NF or Sham training were assessed during a controlled visuomotor squeeze task. Three significant clusters were found where the Real group had greater change in activity following NF than the Sham group (Real > Sham, voxelwise GLM, P < 0.05, corrected), located in the putamen, LOC and the POC of the unaffected hemisphere. For visualization purposes, the mean percent signal change of the significant clusters is plotted on the right, as well as the data from individual participants (represented by open circles). Error bars represent SEM. See also Supplementary Table 9.