Learning from off-target data: characterizing motor imagery neurofeedback in cerebellar ataxia
Poster No:
1554
Submission Type:
Abstract Submission
Authors:
Jeff Soldate1, Jonathan Lisinski2, Joshua Berenbaum3, Amy Mistri3, Ashley Kucharski3, Luca Lutzel3, Cherie Marvel3, Stephen LaConte4
Institutions:
1Virginia Tech, Ronaoke, VA, 2Virginia Tech, Roanoke, VA, 3Johns Hopkins University, Baltimore, MD, 4Virginia Tech, Blacksburg, VA
First Author:
Co-Author(s):
Introduction:
Motor imagery (MI) training is a promising rehabilitation method for individuals with movement disorders. MI activates the motor system "offline" and thus provides a strategy for regaining motor skill through practice that minimizes both frustration and fatigue. But by its nature, the quality of MI training is difficult to assess for both patients and clinicians. To enhance the quality of MI sessions, we developed a real time functional magnetic resonance imaging neurofeedback (rtfMRI-nf) protocol that tracks brain activity associated with MI using supervised learning and provides an objective marker of MI performance. This protocol faces a technical challenge: supervised learning requires labeled images to build a model, but MI labels (e.g. imagine tapping vs. rest) are covert and are thus challenging to verify. Based on previous studies, we hypothesized that overt motor tasks share overlapping neural circuitry with motor imagery and thus motor runs could serve as "bootstrap" training sets to train models that could track MI in real-time. This study examines the use of such motor activity models in prediction of MI and their success as a neurofeedback signal in cerebellar ataxia (CA).
Methods:
We recruited 16 CA participants and 8 age matched controls. Participants performed paced overt motor tapping (tap) as well as motor imagery neurofeedback in scanner (3T, 2000/30ms TR/TE). Two initial runs of tapping were collected and used to train a tap vs. rest classifier. Participants then performed two runs of MI where neruofeedback was provided based on the above model. Participants performed two additional runs of tapping after MI. (Figure 1A)
Tapping: Participants were asked to tap their right index finger at 1Hz or 4Hz in 28-32s blocks, 3 blocks of each per run. Tapping blocks were interspersed with rest blocks of similar size.
Imagery: Using classifiers trained to separate imagined tapping from rest, we asked participants to drive classifier output in both directions. Feedback was delivered based on classifier predictions: prediction of tapping would cause a central fixation to flash while prediction of rest would cause the fixation to remain solid. (Figure 1B)
Analysis: Using 3dsvm and AFNI's realtime functionality, we trained linear support vector classifiers after the first two tapping runs to provide feedback for the subsequent two MI runs. Offline, we were able to compare additional classifier analyses, including MI training to predict tapping (tap-MI) and MI training to predict MI (MI-MI).
Tapping: Participants were asked to tap their right index finger at 1Hz or 4Hz in 28-32s blocks, 3 blocks of each per run. Tapping blocks were interspersed with rest blocks of similar size.
Imagery: Using classifiers trained to separate imagined tapping from rest, we asked participants to drive classifier output in both directions. Feedback was delivered based on classifier predictions: prediction of tapping would cause a central fixation to flash while prediction of rest would cause the fixation to remain solid. (Figure 1B)
Analysis: Using 3dsvm and AFNI's realtime functionality, we trained linear support vector classifiers after the first two tapping runs to provide feedback for the subsequent two MI runs. Offline, we were able to compare additional classifier analyses, including MI training to predict tapping (tap-MI) and MI training to predict MI (MI-MI).
·Figure 1
Results:
On average, classifiers trained on the first two tapping runs were able to predict MI at above chance accuracy. Classification accuracy for participants with CA was significantly higher than controls (Figure 2A). Despite this, tapping models were more reproducible (brain pattern) for control participants (Figure 2B). Classifiers performance for both groups was similar when looking at cross-task pairs for the entire session. MI-MI accuracy was significantly higher than tap-MI as was correlation between resulting models (Figure 2C). When looking at tapping only, the dominant effect driving differences in model correlation was proximity in time.
·Figure 2
Conclusions:
rtfMRI provides a means of quantifying and displaying MI performance through supervised learning. Our results confirm that classification models built from overt tapping can provide neurofeedback on MI performance. However, MI-MI models were both more accurate classifiers and reproducible. Due to our experimental design, we cannot separate the effect of training on MI guided by neurofeedback and training on MI in general when predicting MI. Future work is needed to examine both the effect of neurofeedback on MI and the need for "bootstrap" training sets. Overall, off-target training is viable for enabling neurofeedback in MI and we believe neurofeedback itself might improve classifier performance for MI.
Disorders of the Nervous System:
Neurodegenerative/ Late Life (eg. Parkinson’s, Alzheimer’s) 2
Modeling and Analysis Methods:
Methods Development 1
Motor Behavior:
Brain Machine Interface
Novel Imaging Acquisition Methods:
BOLD fMRI
Keywords:
Cerebellar Syndromes
Data analysis
Machine Learning
Other - neurofeedback
1|2Indicates the priority used for review
Abstract Information
By submitting your proposal, you grant permission for the Organization for Human Brain Mapping (OHBM) to distribute your work in any format, including video, audio print and electronic text through OHBM OnDemand, social media channels, the OHBM website, or other electronic publications and media.
The Open Science Special Interest Group (OSSIG) is introducing a reproducibility challenge for OHBM 2025. This new initiative aims to enhance the reproducibility of scientific results and foster collaborations between labs. Teams will consist of a “source” party and a “reproducing” party, and will be evaluated on the success of their replication, the openness of the source work, and additional deliverables. Click here for more information. Propose your OHBM abstract(s) as source work for future OHBM meetings by selecting one of the following options:
Please indicate below if your study was a "resting state" or "task-activation” study.
Healthy subjects only or patients (note that patient studies may also involve healthy subjects):
Was this research conducted in the United States?
Are you Internal Review Board (IRB) certified? Please note: Failure to have IRB, if applicable will lead to automatic rejection of abstract.
Were any human subjects research approved by the relevant Institutional Review Board or ethics panel? NOTE: Any human subjects studies without IRB approval will be automatically rejected.
Were any animal research approved by the relevant IACUC or other animal research panel? NOTE: Any animal studies without IACUC approval will be automatically rejected.
Please indicate which methods were used in your research:
For human MRI, what field strength scanner do you use?
Which processing packages did you use for your study?
Provide references using APA citation style.
Decety, J. (1996). The neurophysiological basis of motor imagery, Behav. Brain Res, 77(1), 45-52.
Papageorgiou, T.D., Lisinski, J.M., McHenry, M.A., White, J.P., LaConte, S.M. (2013). Brain-computer interfaces increase whole-brain signal to noise. PNAS, 110(33), 13630-13635.
Yoxon, E., Welsh, T.N. (2020). Motor system activation during motor imagery is positively related to the magnitude of cortical plastic changes following motor imagery training, Behav. Brain Res, 390, 112685.