Machine Learning for Motor Imagery Wrist Dorsiflexion Prediction in Brain-Computer Interface Assisted Stroke...

Area of Science:

• Neuroscience
• Rehabilitation Medicine
• Biomedical Engineering

Background:

• Stroke significantly impacts survivors' physical, cognitive, and emotional well-being, necessitating effective rehabilitation.
• Rehabilitation therapies aim to restore function and promote independent living.
• Automated training protocols, including Brain-Computer Interface (BCI) systems, enhance rehabilitation efficiency and reduce reliance on professional trainers.

Purpose of the Study:

• To evaluate the efficacy of various machine learning (ML) algorithms for motor imagery (MI) wrist dorsiflexion prediction.
• To apply these ML algorithms within a BCI-assisted stroke rehabilitation framework.
• To identify the optimal ML classifier for stroke rehabilitation using electroencephalogram (EEG) signals.

Main Methods:

• Utilized electroencephalogram (EEG) signals from eleven stroke survivors with paresis.
• Employed the doubling sub-band selection filter bank common spatial pattern (DSBS-FBCSP) as a feature extractor.
• Tested several ML algorithms: Decision Tree (DT), Naive Bayesian (NB), Linear Discriminant Analysis (LDA), Support Vector Machine (SVM), Ensemble Learning Classifier (ELC), and Artificial Neural Network (ANN).

Main Results:

• The study assessed the performance of multiple ML algorithms for MI wrist dorsiflexion prediction.
• The Artificial Neural Network (ANN) classifier demonstrated superior performance compared to other evaluated algorithms.
• The DSBS-FBCSP feature extractor was effective in processing EEG signals for BCI applications.

Conclusions:

• Machine learning algorithms, particularly ANNs, show significant promise for BCI-assisted stroke rehabilitation.
• Accurate MI prediction using EEG signals can drive personalized rehabilitation protocols.
• This approach can enhance recovery and improve functional outcomes for stroke survivors.