Sparse Sonomyography-based Estimation of Isometric Force: A Comparison of Methods and Features.

Area of Science:

• Biomechanics
• Medical Imaging
• Rehabilitation Engineering

Background:

• Noninvasive estimation of joint and muscle forces is crucial for clinical and research applications.
• Surface electromyography (sEMG) is limited by poor signal-to-noise ratio and spatiotemporal specificity for accurate force estimation.
• Traditional ultrasound methods for force prediction are computationally complex.

Purpose of the Study:

• To develop an offline sonomyography-based approach for estimating continuous isometric muscle force using sparse ultrasound data.
• To reduce feature dimensionality and computational complexity compared to existing ultrasound methods.
• To evaluate regression models and assess the feasibility of wearable sonomyography systems.

Main Methods:

• Developed an offline sonomyography approach using A-mode scanlines from B-mode ultrasound images.
• Isolated anatomically relevant features to reduce the feature space.
• Evaluated four regression methodologies (including Gaussian process regression) for force prediction.
• Simulated wearable system feasibility by varying transducer placement and number.

Main Results:

• Gaussian process regression models demonstrated superior performance in predicting continuous force levels.
• Accurate force prediction was achieved using only four equispaced transducers in offline settings.
• The proposed method significantly reduced computational complexity.

Conclusions:

• Sonomyography offers a robust and computationally efficient alternative to sEMG for muscle force estimation.
• Gaussian process regression with sparse ultrasound data is effective for continuous isometric force prediction.
• Findings support the development of practical, wearable sonomyography-based force prediction systems.