Linear and nonlinear regression techniques for simultaneous and proportional myoelectric control

Area of Science:

• Biomedical Engineering
• Rehabilitation Engineering
• Neuroprosthetics

Background:

• Current prosthetic hand control strategies are limited by sequential, single degree-of-freedom (DoF) control.
• Increasing numbers of active joints in prostheses necessitate advanced control methods.
• Existing myoelectric control lacks independent, simultaneous, and proportional joint actuation.

Purpose of the Study:

• To systematically compare linear and nonlinear regression techniques for myoelectric control of two-DoF wrist movements.
• To evaluate control accuracy based on electrode count and training data characteristics.
• To provide guidance for clinical implementation of advanced prosthetic control.

Main Methods:

• Investigated linear regression, mixture of linear experts (ME), multilayer-perceptron, and kernel ridge regression (KRR).
• Utilized electro-myographic (EMG) signals from ten able-bodied subjects and one individual with a limb deficiency.
• Performed offline analysis of control accuracy as a function of system parameters.

Main Results:

• Kernel ridge regression (KRR) demonstrated superior performance among the tested methods.
• Linear models achieved comparable accuracy to KRR with feature space transformations, offering lower computational cost.
• Mixture of linear experts (ME) showed significant potential for advanced prosthetic control applications.

Conclusions:

• Nonparametric methods like KRR provide high accuracy for simultaneous myoelectric control.
• Linear models, particularly ME, are promising for efficient and physiologically inspired prosthetic control.
• Optimizing training data and electrode configuration is crucial for effective clinical application.