Related Experiment Video
Updated: May 1, 2026
Author Spotlight: Using Motor Imagery Brain-Computer Interface to Improve Motor and Cognitive Function in Stroke Patients
Published on: September 1, 2023
Classification of motor imagery performance in acute stroke
Chayanin Tangwiriyasakul1, Victor Mocioiu, Michel J A M van Putten
1Neural Engineering, Institute for Biomedical Technology and Technical Medicine, University of Twente, Enschede, The Netherlands. Clinical Neurophysiology, Institute for Biomedical Technology and Technical Medicine, University of Twente, Enschede, The Netherlands.
Classifying motor imagery using electroencephalography (EEG) is crucial for therapy. Combining common spatial pattern (CSP) filtering with linear discriminant analysis (LDA) and using unaffected hemisphere data improved classification accuracy, particularly with more electrodes.
Area of Science:
- Neuroscience
- Biomedical Engineering
- Rehabilitation Technology
Background:
- Motor imagery (MI) therapy relies on accurate classification of sensorimotor rhythm suppression.
- Optimizing electroencephalography (EEG) classification methods for MI requires high accuracy and minimal channel usage.
- Investigating advanced signal processing techniques like common spatial pattern (CSP) filtering combined with linear discriminant analysis (LDA) is essential for improving MI detection.
Purpose of the Study:
- To evaluate the added benefit of CSP filtering to an LDA classifier for motor imagery detection.
- To assess the impact of different channel configurations on classification performance.
- To determine if EEG data from the unaffected hemisphere can effectively train a classifier for stroke patients.
Main Methods:
- EEG data were recorded from 10 acute stroke patients and 11 healthy subjects using 60 channels.
- A linear discriminant analysis (LDA) classifier was trained using motor execution tasks.
- The performance of LDA combined with common spatial pattern (CSP) filtering was evaluated using varying numbers of electrodes (3, 11, and 45) and data from affected and unaffected hemispheres.
Main Results:
- The addition of CSP to LDA showed no significant improvement with limited channels (AU-ROC ≈ 0.70).
- Expanding to 45 electrodes significantly improved LDA+CSP performance (AU-ROC ≈ 0.90), though no single 'most responsible' electrode was identified.
- Classifiers trained using EEG data from the healthy hemisphere of mild-to-moderate stroke patients achieved acceptable performance (AU-ROC ≈ 0.70).
Conclusions:
- For motor imagery classification, utilizing EEG data solely from the unaffected motor cortex is sufficient for training a classifier.
- While CSP-LDA shows promise, optimal performance in motor imagery detection may require a broader spatial sampling of EEG signals.
- This finding has implications for developing more efficient and accessible neurofeedback systems for motor rehabilitation.
