In recent years there is a growing interest in the potential application of Brain-Computer interfaces (BCI) for psychiatric and neurological disorders, including stroke (1). BCIs enable the use of brain signals to control an external effector or a virtual device and to provide on-line information about brain activity to the experimental subjects, leading to the modulation of dysfunctional brain regions. BCIs based on fMRI (fMRI-BCI) provide high spatial resolution, and allows the study and modulation of both cortical and subcortical regions (2).

A stroke affecting the primary motor cortex (M1) is commonly followed by a “deactivation” of the contralateral cerebellum cortex, presumably by the loss of excitatory afferent inputs (3). Imaging studies support the role of the cerebellum in the functional recovery from stroke (4,5,6).

Considering this “network dysfunction” after stroke, it would be relevant to reactivate the connectivity among MI and cerebellum for motor rehabilitation (7,8).

Five healthy individuals (23.2 ± 5.4 years, males, right handed) were trained in a real-time fMRI-BCI protocol with contingent visual feedback and motor imagery of the right hand during 3 days, 6 training runs per day (~6 minutes each), alternating baseline (Base) and up-regulation (Upreg) blocks.

The feedback signal was computed including the BOLD signal and the correlation (functional connectivity) between the regions of interest (ROIs), i.e. left M1 and right cerebellar hemisphere, according to:

F = ((BOLD_Upreg–BOLD_Base)_Cerebellum + (BOLD_Upreg–BOLD_Base)_M1) * (1+Correlation)

A first group of subjects (n=4) was trained with a feedback equation that weighted equally the contribution of each ROI (equal-feedback group). Since this group displayed a better volitional regulation of cerebellum than M1 activity (please see results), a subsequent group of subjects (n=2) was trained with a feedback equation that weighted each ROI differentially (30% cerebellum and 70% M1) (weighed-feedback group).

Online functional connectivity was calculated using the correlation of the BOLD signal of both ROIs (cerebellum and M1) in a moving window of 7 volumes TR, across every run.

The fMRI-BCI system was implemented using a 1.5T MR Scanner (Philips Achieva, The Netherlands) running a real-time reconstructor package (DRIN) with functional image acquisition using FFE-EPI sequence with TR/TE=1500/45 ms, voxel size=3.28x3.28x5mm3 and 238 measurements. We used Turbo Brain Voyager software (Brain Innovations, The Netherlands) and custom MATLAB scripts to generate feedback information. Presentation software (NBS, USA) was used to show the feedback (thermometer) in an MR-compatible Visual System (NNL AS, Norway).

Anatomical T1-weighted images were acquired every day. FMRI volumes were preprocessed and analyzed using SPM and additional MATLAB scripts.

To evaluate self-regulation success we use a ratio to calculate the percent change of BOLD signal during up-regulation compared to baseline blocks for each run in a selected ROI of cerebellum and M1 (Figure 1).

As reference, we also calculated the percent change of the BOLD signal during overt movement of the hand. Therefore, three conditions were included for analysis: equal-feedback, weighted- feedback and overt-movement.

All subjects were able to self-regulate the activity of both ROIs, albeit to different degree.

There was an inter-group significant difference of the percent of BOLD signal change (ratio) in both ROIs (cerebellum and M1) (Kruskal-Wallis, p<0.05, data not-normally distributed, Kolmogorov-Smirnov and Shapiro-Wilk test, p<0.05) (Table 1), across the BCI training: the weighted-feedback group showed a significantly higher capability to self-regulate the BOLD signal (Mann-Withney; p<0.05) in both ROIs, compared to the equal-feedback group.

Additionally, there was a significant difference between the cerebellum and M1 BOLD self-regulation (Wilcoxon; p<0.01), with higher cerebellum than M1 BOLD signal modulation in the equal-feedback group (Figure 2 and Figure 3).

Regarding connectivity, there was an inter-group significant difference of the correlation values (Kruskal-Wallis, p<0.05) in the up-regulation block: we found significant differences between the equal-feedback and the weighted-feedback groups, and also between the equal-feedback and the overt-movement condition (Mann-Withney; p<0.05) (Table 2). Moreover, there was a significant difference in the correlation values in up-regulation between the first and third session for the weighted-feedback group, with higher values in the last session (Wilcoxon; p<0.012).

The results indicate that volitional self-regulation of cerebellum-M1 connectivity is feasible with fMRI-BCI.

The data suggests that cerebellum activity is more easily recruited than M1. The weighted feedback methodology provides the possibility to enhance the activation of certain ROI with fMRI-BCI. This strategy allowed achieving self-regulation in both ROIs and similar correlation levels (functional connectivity) between cerebellum and M1 with fMRI-BCI than with overt movement.