Fig 1.
Subject screening, enrollment, and study flow diagram.
Fig 2.
Real-time functional magnetic resonance imaging (fMRI) acquisition and neurofeedback system: Visual stimulation displayed by custom Python software running on a dedicated presentation computer (8-GB RAM, 2.70 GHz, Core i7 processor) running Windows 7; reconstructed image was sent via DRIN protocol (Phillips Medical Systems) to Turbo Brain Voyager software (TBV, version 3.2) running on a dedicated computer (8GB RAM, 3.6GHz, Xeon E5-1620) to identify active voxels within a defined region of interest (ROI) in real-time.
The feedback estimation was performed in TBV and saved as a bitmap (.PNG file) to disk in a shared network directory. These files were picked up by custom Python software for final display to the subject. The feedback stimulus consisted of a brain picture with a purple icon superimposed. The icon changed in size and brightness to reflect the level of activation in the ROI. This visual stimulus was the source of neurofeedback. (With permission, [39]).
Fig 3.
Real-time functional near infrared spectroscopy (rt-fNIRS) neural feedback system.
The command to move was a visual cue including a cartoon of the brain, the words “Wrist extension—GO”, and a cartoon of the wrist moving from a neutral to an extended position in 1 s. Brain activation during the attempt to extend the wrist was captured by NIRx NIRSport. Output from the NIRSport was synchronized with NIRx NIRStar Version 15, which ran on a dedicated acquisition computer (24-GB RAM, 2.60 GHz, Core i7 processor) running Windows 8. The data were pre-processed and thresholded by custom Matlab code. Matlab output informed Python code how to update the visual participant feedback, shown as a purple color in the brain region. Matlab output also triggered functional electrical stimulation (FES) of wrist extensors. (With permission, [39]).
Table 1.
Subject characteristics.
Table 2.
Group data for motor measures.
Fig 4.
Brain map outcome measures acquired at pre-treatment and follow-up.
Brain maps are shown for right and left hemisphere regions: HK, BA4—HK, BA3, and BA6) reflecting the summary results in Table 3 above for pre-treatment and follow-up. (FWEc, p value = 0.05 with small volume correction, used for thresholding each ROI). Key: R = right hemisphere; LH = left hemisphere; RH = right hemisphere. Individual activation values are in the S1 File, Section I. Outcome Measures, Section 1.2. fMRI Outcome Measures, Individual Subject Data; Table 2a-2d in S1 File. 4.1. Subject 1. S1 in S1 File showed lessening of activation from pre-treatment to follow-up in all ROIs. 4.2. Subject 2. S2 in S1 File showed lessening of activation in all ROIs from pre-treatment to follow-up, except for right sensory region which remained consistent. 4.3. Subject 3. S3 in S1 File showed increases in all left lesioned hemisphere ROIs, except sensory which decreased minimally; there was an increase in right Primary Motor-Hand Knob and all other ROIs remained consistent or with only minimal change. 4.4. Subject 4. S4 in S1 File. For S4 in S1 File, from pre-treatment to post-treatment in the right lesioned hemisphere, there was a 26% decrease for the sensory region, with zero or near zero for the remaining right ROIs. For the left hemisphere, there was a decrease in all ROIs, except for a minimal increase in the Hand Knob region.
Table 3.
Patterns of change from pre-treatment to 3 month follow-up*, according to percent volume of activation; wrist extension motor task.
Fig 5.
fNIRS: Range of values for ‘active–rest’ hemoglobin concentration during wrist extension task three conditions, all during wrist extension task.
Panel A. Healthy controls fNIRS range of values during wrist extension. Panel B. Stroke survivors during rt-fNIRS training for wrist extension. Panel C. Stroke survivors during wrist extension without neural feedback (pre-, post-treatment, follow-up. Key: Each vertical bar shows the range of HbO for a given individual, each of whom are identified on the horizontal axis. For y-axis, oxyhemoglobin concentration values were calculated as the difference between ‘active-rest’ condition. This difference variable is a change in oxyhemoglobin concentration from rest to the active movement state, derived from the fNIRS signal. The above rectangles for each participant represent the range of values for that given participant. *Data for healthy controls and S1 were acquired using Hitachi fNIRS system and data for S2, S3, and S4 were acquired using the NIRx fNIRS system.
Fig 6.
Brain maps across three neural feedback training sessions (rows a, b, and c from sessions 1, 2, and 3, respectively) using real time fMRI (rt-fMRI) for each of four cases (6.1, 6.2, 6.3., and 6.4, respectively).
Brain maps show functional magnetic resonance imaging (fMRI) data acquired during each of the three real-time fMRI neural feedback wrist coordination training sessions. A session-specific, t-statistic map was overlaid on the session-specific T1 anatomical image. We used a significance threshold of p < 0.05 with small volume family-wise error correction for each ROI. 6.1. for S1. Brain maps for the left lesioned hemisphere reflect small changes (≤ 10%) or consistent pattern comparing first to last of the three sessions, rows a and c. The right hemisphere showed a marked lessening of activation (19%–62% decreases; details contained in Table 3a, S1 File). 6.2. for S2. Brain maps for the left lesioned hemisphere reflect a consistent pattern of activation comparing first to last of the sessions, rows a and c, for Hand Knob and premotor, and a lessening of activation in Primary Motor-Hand Knob and sensory. The right hemisphere showed a similar pattern across the three sessions, with the exception of a 10% increase in Primary Motor-Hand Knob (details in Table 3b, S1 File). 6.3. for S3. Brain maps reflect a marked lessening of activation comparing the first and last of the three sessions for the left lesioned hemisphere (29% to 59%) and the right hemisphere (11%–56%; details in Table 3c, S1 File). 6.4. for S4. Brain maps reflect lessening of activation comparing the first and last of the three sessions, rows a and c for right lesioned hemisphere (≤ 10%) except for Hand Knob remaining constant. The left hemisphere showed lessening of activation (7%-41%), except for Hand Knob which remained constant; details in Table 3d, S1 File).
Table 4.
Performance during rt-fMRI neural feedback training patterns of change in % volume of activation for specific ROIs across the first and last rt-fMRI neural training sessions.
Fig 7.
Brain activation mean success rate during rt-fNIRS neural feedback training, for each of four subjects.
For each stroke survivor, mean (rectangle height) success rate for the 10 rt-fNIRS neural training sessions is shown with standard deviation (gray standard deviation lines.