Abstract:
This research addresses challenges in stroke rehabilitation by proposing an alternative to
current methods that are often costly and resource-intensive. With limited accessibility to
rehabilitation interventions due to constraints in skilled therapists and advanced
technology, there is a need for innovative solutions. Past research predominantly focused
on 2-DoF systems, limiting exploration in diverse multiple DoF EMG classification
systems. The initial phase of this study involves a comprehensive investigation into the
potential of user-specific non-functional hand movements. By examining 16 distinct wrist
and finger motions using five classifiers (Wide Neural Networks, Ensemble Bagged Trees,
Cubic SVM, Fine KNN, Logistic Regression Kernel), significant differences were
revealed. One-Way Anova testing (p<0.05) revealed significant classifier differences while
Wilicoxon signed-rank tests (p<0.05) distinguished individual finger and wrist getsures.
For the development of a 3-DOF real-time game, specific hand movements were
strategically selected based on the best obtained accuracies. Noteworthy movements
include "Abduction of thumb and flexion of others," "Flexion of all fingers," "Extension
of index and flexion of others”, “Wrist Radial Deviation (Clockwise)”, Wrist Extension”,
and “Wrist Supination Axis: middle finger.” A MATLAB APP DESIGNER-based GUI
was developed with two tabs. Real-time metrics for the designed game exhibited promising
results: a completion rate of 94.22% (±2.74%), throughput of 0.40 bits s−1 (±0.03), path
efficiency of 87.02%, and minimal overshoot of 12.68%. These findings underscore the
game's potential as an effective and engaging tool for stroke rehabilitation, offering a novel
approach to address the challenges associated with current methods.
