Attribute-Driven Granular Model for EMG-Based Pinch and Fingertip Force Grand Recognition.

Area of Science:

• Biomedical Engineering
• Machine Learning
• Rehabilitation Technology

Background:

• Precise control of prosthetic hands, particularly pinch type and force, is crucial for fine manipulation.
• Decoding both pinch type and force from myoelectric signals presents a significant challenge.

Purpose of the Study:

• To propose and evaluate an attribute-driven granular model (AGrM) for simultaneous classification of pinch types and prediction of fingertip force from myoelectric signals.
• To improve the accuracy and efficiency of prosthetic hand control systems.

Main Methods:

• Developed an attribute-driven granular model (AGrM) utilizing supervised granulation with an additional attribute as a latent variable.
• Collected 16-channel surface electromyographic (EMG) signals and continuous fingertip force data from subjects performing eight types of hand pinches.
• Compared AGrM with clustering-based granulation methods.

Main Results:

• Achieved approximately 97.2% accuracy in pinch-type recognition, an improvement of 1.8%.
• Obtained over 90% accuracy in fingertip force prediction across granular levels greater than six.
• Demonstrated robustness through sensitivity analysis regarding channel combinations and interferences.
• AGrM showed comparable pinch recognition accuracy to other methods but with lower computational cost and higher force prediction accuracy.

Conclusions:

• The AGrM effectively decodes both pinch type and force from myoelectric signals, advancing prosthetic hand control.
• The model offers a robust and computationally efficient solution for multifunctional prosthetic hand manipulation.
• This approach holds promise for developing more intuitive and capable prosthetic devices.