Performance evaluation of an EEG-guided robotic glove with machine learning models for hand rehabilitation in injured athletes

Abstract

Background

Assistive rehabilitation technologies play a crucial role in improving motor recovery for individuals with hand injuries particularly athletes where traditional methods often lack real-time neural feedback and adaptability.

Objective

This study aims to evaluate the effectiveness of an EEG-guided robotic glove integrated with machine learning techniques for enhancing hand rehabilitation and motor engagement. A total of 250 subjects including 50 healthy individuals and 200 injured athletes were analysed under conditions with and without the robotic glove. Electroencephalography (EEG) that captures and preprocesses brain signals utilizing MATLAB subsequently performing effective feature extraction and frequency analysis of alpha and beta bands. The research also seeks to uncover notable trends in EEG signals by conducting power spectral density analysis, utilize machine learning techniques for precise classification of injured athletes and healthy individuals and assess system performance using metrics like accuracy and ROC analysis. The Receiver Operating Characteristic (ROC) curve is a crucial assessment tool employed to evaluate the effectiveness of machine learning algorithms in distinguishing EEG signals from athletes with injuries and healthy participants. Moreover the aim encompasses enhancing electrode choice and guaranteeing dependable and uniform outcomes for possible application in rehabilitation tracking.

Methods

Multiple machine learning models including Linear Discriminant Analysis (LDA), Extended Logistic Regression (ELR), Tuned Neural Network (TNN), K-Nearest Neighbor-Based (KNB) and Fine Gaussian Support Vector Machine (SVM) were employed to classify motor engagement levels into low, medium and high categories.

Results

The results showed significant variations in EEG signals (p < 0.05) with injured participants using the robotic glove demonstrating increased beta activity and reduced alpha activity indicating enhanced motor engagement. Among the models LDA achieved the highest classification accuracy of 99%, followed by ELR and TNN at 98.5% while Fine Gaussian SVM showed the lowest accuracy at 73.5%.

Conclusion

Overall, the findings confirm that the EEG-guided robotic glove significantly improves motor function and cognitive engagement and the integration of machine learning provides an effective and reliable approach for advanced rehabilitation systems.