Proportional estimation of finger movements from high-density surface electromyography

Area of Science:

• Biomedical Engineering
• Neurorehabilitation
• Signal Processing

Background:

• Restoring hand function after nervous system injury is crucial for rehabilitation.
• Surface electromyography (sEMG) enables user-driven control of rehabilitation robots.
• Active engagement using voluntary activation triggers robotic assistance.

Purpose of the Study:

• To investigate selective estimation of individual finger movements from high-density surface EMG (HD-sEMG).
• To minimize interference between individual finger movements during robotic control.
• To evaluate regression algorithms for real-time and offline finger movement control.

Main Methods:

• Compared Linear Discriminant Analysis (LDA), Common Spatial Patterns Proportional Estimator (CSP-PE), and Thresholding (THR) algorithms.
• Utilized high-density surface EMG (HD-sEMG) signals from nine healthy subjects per test.
• Assessed performance using normalized mean square error (nMSE), classification accuracy (CA), mean false activation amplitude (MAFA), and Pearson correlation coefficient (PCORR).

Main Results:

• CSP-PE outperformed LDA with fewer electrodes (≤24) in offline tests, showing higher precision and less crosstalk.
• LDA and CSP-PE demonstrated similar performance in online tests, with CSP-PE offering more stable results.
• Thresholding (THR) provided comparable accuracy in some cases but lacked consistency across subjects and fingers.

Conclusions:

• CSP-PE is the preferred method for individual finger control with limited electrodes (<24).
• Both CSP-PE and LDA are suitable for higher-resolution recordings.
• While THR is simpler, pattern recognition methods are superior for reliable finger control in rehabilitation robotics.