Low-dimensional Manifolds of Neural Dynamics of Acute Stroke Patients in Motor Imagery
Poster No:
856
Submission Type:
Abstract Submission
Authors:
Tao Liu1, Jay Wang2, Sadia Shakil3, Raymond Tong3
Institutions:
1Nuffield Department of Clinical Neurosciences, University of Oxford, Oxford, Oxfordshire, 2The Hong Kong Polytechnic University, Department of Biomedical Engineering, Kowloon, Hong Kong, 3Department of Biomedical Engineering, The Chinese University of Hong, Sha Tin, Hong Kong
First Author:
Co-Author(s):
Jay Wang
The Hong Kong Polytechnic University, Department of Biomedical Engineering
Kowloon, Hong Kong
The Hong Kong Polytechnic University, Department of Biomedical Engineering
Kowloon, Hong Kong
Sadia Shakil
Department of Biomedical Engineering, The Chinese University of Hong
Sha Tin, Hong Kong
Department of Biomedical Engineering, The Chinese University of Hong
Sha Tin, Hong Kong
Raymond Tong
Department of Biomedical Engineering, The Chinese University of Hong
Sha Tin, Hong Kong
Department of Biomedical Engineering, The Chinese University of Hong
Sha Tin, Hong Kong
Introduction:
Disruption of neural circuits and population dynamics in specific brain regions is a major contributor to motor dysfunction following stroke (Campos et al., 2023). Neural modes represent the coordinated spatiotemporal activity of neuron groups during movement (Gallego et al., 2017). Studies have demonstrated that movement-related neural population dynamics can be captured as stable, low-dimensional neural manifolds within the neural mode space, consistent across time and subjects (Gallego et al., 2020; Safaie et al., 2023). However, research on these manifolds in stroke patients remains limited, largely due to the reliance on invasive multi-electrode recordings. To address this gap, we present a novel approach to derive low-dimensional neural manifolds of neural population dynamics from non-invasive EEG data of acute stroke patients performing motor imagery tasks and evaluate their cross-subject stability.
Methods:
32-channel EEG was recorded from 30 acute stroke subjects (23 males, 7 females, 14 left hemiplegia, 16 right hemiplegia) during motor imagery grasping task.
To remove artifacts in EEG signals, we performed preprocessing following process mentioned in (Chen et al., 2021). The pre-processed EEG data were subsequently used for source localization with eLORETA (Pascual-Marqui, 2007) based on public head model template, projecting EEG from electrode to voxel space. Eight regions of interest recommended by (Khan et al., 2021) were selected based on the Automated Anatomical Labeling atlas (Tzourio-Mazoyer et al., 2002), within which voxel activities simulated neural population activity.
To capture neural population dynamics during motor imagery, voxel time series were divided into non-overlapping 30ms windows, with each window averaged to represent real-time voxel activity. The resulting trials × voxels × windows matrix (N × V × T) was used for analysis. Principal component analysis was applied to the concatenated input matrix along trials axis (V × N*T), transforming each trial's V × T matrix. Each principal component represents different neural modes. We used canonical correlation analysis to align the neural manifolds for easier comparison.
Neural manifolds from different subjects were paired and all combinations were traversed. Canonical correlation analysis was applied to each pair, and the average canonical correlation across all pairs was used to measure cross-subject similarity. A support vector classifier was trained to identify neural manifolds during motor imagery task from resting state. The classifier was trained on the data of one subject in each pair while tested on the other. Mean accuracy across all pairs assessed the classification performance. Figure 1 shows the overall process of obtaining low-dimensional neural manifolds based on whole-brain EEG recordings.
To remove artifacts in EEG signals, we performed preprocessing following process mentioned in (Chen et al., 2021). The pre-processed EEG data were subsequently used for source localization with eLORETA (Pascual-Marqui, 2007) based on public head model template, projecting EEG from electrode to voxel space. Eight regions of interest recommended by (Khan et al., 2021) were selected based on the Automated Anatomical Labeling atlas (Tzourio-Mazoyer et al., 2002), within which voxel activities simulated neural population activity.
To capture neural population dynamics during motor imagery, voxel time series were divided into non-overlapping 30ms windows, with each window averaged to represent real-time voxel activity. The resulting trials × voxels × windows matrix (N × V × T) was used for analysis. Principal component analysis was applied to the concatenated input matrix along trials axis (V × N*T), transforming each trial's V × T matrix. Each principal component represents different neural modes. We used canonical correlation analysis to align the neural manifolds for easier comparison.
Neural manifolds from different subjects were paired and all combinations were traversed. Canonical correlation analysis was applied to each pair, and the average canonical correlation across all pairs was used to measure cross-subject similarity. A support vector classifier was trained to identify neural manifolds during motor imagery task from resting state. The classifier was trained on the data of one subject in each pair while tested on the other. Mean accuracy across all pairs assessed the classification performance. Figure 1 shows the overall process of obtaining low-dimensional neural manifolds based on whole-brain EEG recordings.
Results:
Figure 2 illustrates our obtained neural manifolds and their cross-subject stability in alpha band. Results show that 3-dimensional trajectories of the obtained neural manifolds share stable structure after alignment (Figure 2(A)), with high cross-subject canonical correlation proving their similarity (Figure 2(B)). Figure 2(C) manifests that classification performance in all regions of interest exceed 65%, higher than random level, further demonstrating the cross-subject similarity and tasks-relevance of our obtained neural manifolds.
Conclusions:
In this study, we identified low-dimensional neural manifolds of neural population dynamics in acute stroke patients performing motor imagery tasks based on EEG. A public acute stroke dataset (Liu et al., 2024) was utilized for our analysis. Our results suggest that stable and task-relevant low-dimensional neural manifolds depicting population dynamics exist across acute stroke patients. Our research offers a novel perspective for understanding the mechanisms of stroke and introduces a potential biomarker for stroke diagnosis and rehabilitation.
Higher Cognitive Functions:
Imagery
Learning and Memory:
Neural Plasticity and Recovery of Function 1
Skill Learning
Modeling and Analysis Methods:
Methods Development 2
Keywords:
Electroencephaolography (EEG)
Motor
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?
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:
Provide references using APA citation style.
2. Chen, C., Yuan, K., Wang, X., Khan, A., Chu, W. C. W., & Tong, R. K. Y. (2021). Neural Correlates of Motor Recovery after Robot‐Assisted Training in Chronic Stroke: A Multimodal Neuroimaging Study. Neural Plasticity, 2021(1), 8866613.
3. Gallego, J. A., Perich, M. G., Chowdhury, R. H., Solla, S. A., & Miller, L. E. (2020). Long-term stability of cortical population dynamics underlying consistent behavior. Nature neuroscience, 23(2), 260-270.
4. Gallego, J. A., Perich, M. G., Miller, L. E., & Solla, S. A. (2017). Neural manifolds for the control of movement. Neuron, 94(5), 978-984.
5. Khan, A., Chen, C., Yuan, K., Wang, X., Mehra, P., Liu, Y., & Tong, K. Y. (2021). Changes in electroencephalography complexity and functional magnetic resonance imaging connectivity following robotic hand training in chronic stroke. Topics in Stroke Rehabilitation, 28(4), 276-288.
6. Liu, H., Wei, P., Wang, H., Lv, X., Duan, W., Li, M., ... & Hao, J. (2024). An EEG motor imagery dataset for brain computer interface in acute stroke patients. Scientific Data, 11(1), 131.
7. Pascual-Marqui, R. D. (2007). Discrete, 3D distributed, linear imaging methods of electric neuronal activity. Part 1: exact, zero error localization. arXiv preprint arXiv:0710.3341.
8. Safaie, M., Chang, J. C., Park, J., Miller, L. E., Dudman, J. T., Perich, M. G., & Gallego, J. A. (2023). Preserved neural dynamics across animals performing similar behaviour. Nature, 623(7988), 765-771.
9. Tzourio-Mazoyer, N., Landeau, B., Papathanassiou, D., Crivello, F., Etard, O., Delcroix, N., ... & Joliot, M. (2002). Automated anatomical labeling of activations in SPM using a macroscopic anatomical parcellation of the MNI MRI single-subject brain. Neuroimage, 15(1), 273-289.