Estimation of wrist angle from sonomyography using support vector machine and artificial neural network models.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Sonomyography (SMG) utilizes real-time muscle thickness changes from ultrasound to represent muscle contractions.
• Accurate prediction of dynamic joint angles is crucial for various applications, including prosthetics and rehabilitation.

Purpose of the Study:

• To predict dynamic wrist angles using sonomyography (SMG) signals.
• To evaluate the performance of Least Squares Support Vector Machine (LS-SVM) and Artificial Neural Networks (ANN) for this prediction task.

Main Methods:

• Synchronized wrist angle and SMG data were collected from extensor carpi radialis muscles during wrist extension/flexion at different rates.
• LS-SVM, back-propagation (BP) ANN, and radial basis function (RBF) ANN models were developed and trained.
• Models trained at 22.5 cycles/min were used to predict wrist angles at other rates.

Main Results:

• All three methods accurately predicted wrist angles across different rates (RMSD < 0.2, CC > 0.98).
• LS-SVM demonstrated superior performance (RMSD < 0.15, CC > 0.99) compared to ANN models.
• Models trained at 22.5 cycles/min generalized well to predict data from other rates.

Conclusions:

• Dynamic wrist angles can be precisely estimated from extensor carpi radialis muscle thickness changes using SMG.
• LS-SVM and ANN models are effective for SMG-based wrist angle prediction.
• The developed models show potential for real-time applications in human-computer interaction and assistive technologies.