Towards low-dimensionsal proportional myoelectric control

Area of Science:

• Biomedical Engineering
• Rehabilitation Technology
• Signal Processing

Background:

• Powered myoelectric prostheses aim for enhanced dexterity through simultaneous control of multiple degrees-of-freedom (DOFs).
• Surface electromyogram (sEMG) signal features can reconstruct finger movements, but involve numerous correlated predictors and target variables.
• Synergistic patterns in sEMG data have been previously used to improve kinematics decoding.

Purpose of the Study:

• To develop a novel framework for simultaneous input-output dimensionality reduction for sEMG-based prosthetic control.
• To compare the proposed method with existing dimensionality reduction techniques like PCA and FRR.

Main Methods:

• A framework for simultaneous input-output dimensionality reduction was proposed, utilizing the generalized eigenvalue problem formulation of multiple linear regression (MLR).
• The method was evaluated against principal component analysis (PCA) for input-output dimensionality reduction.
• Prediction accuracy was compared with the full rank regression (FRR) method.

Main Results:

• The proposed MLR-based framework demonstrated superior performance compared to PCA for simultaneous input-output dimensionality reduction.
• The methodology achieved prediction accuracy comparable to the full rank regression (FRR) method.
• Effective dimensionality reduction was achieved using only a few relevant dimensions.

Conclusions:

• The proposed generalized eigenvalue problem formulation offers an effective approach for simultaneous input-output dimensionality reduction in sEMG-based prosthetic control.
• This method enhances the efficiency and accuracy of decoding finger movements, paving the way for more dexterous myoelectric prostheses.
• The framework provides a significant advancement in prosthetic control by reducing data complexity while maintaining high prediction accuracy.