Continuous grip force estimation from surface electromyography using generalized regression neural network.

Area of Science:

• Biomedical Engineering
• Rehabilitation Technology
• Neuroprosthetics

Background:

• Accurate grip force estimation is crucial for developing advanced prosthetic limbs.
• Current prosthetic control systems require improved methods for intuitive and precise hand function.
• Surface electromyography (sEMG) offers a non-invasive approach for capturing muscle activity related to grip.

Purpose of the Study:

• To develop and validate a method for estimating continuous grip force from sEMG signals.
• To assess the efficacy of the method across different forearm postures (supination, neutral, pronation).
• To compare the performance of Generalized Regression Neural Network (GRNN) and Multilinear Regression Model (MLR) algorithms for grip force estimation.

Main Methods:

• sEMG signals were recorded from six forearm muscles in ten able-bodied subjects and one transradial amputee.
• Grip force was measured synchronously with sEMG data.
• Two algorithms (GRNN and MLR) were evaluated using various EMG features and grip force profiles (triangle, trapezoid, fast triangle).

Main Results:

• The optimal regressor combining time-domain features and GRNN achieved high accuracy (R2=96.33%) in able-bodied subjects.
• The method demonstrated significant accuracy (R2=86.86%) for the unilateral amputee, indicating successful grip force estimation.
• Accurate estimation of grip force across multiple postures was achieved using mirrored bilateral training.

Conclusions:

• The proposed sEMG-based method shows strong potential for precise force control in prosthetic hands.
• This approach facilitates more natural and functional control for individuals with unilateral limb differences.
• Mirrored bilateral training enhances the adaptability and accuracy of prosthetic control for amputees.