Automated Muscle Analysis with Wearable Ultrasound and Motion Sensors

Sonomyography (SMG) is the quantitative analysis of muscle function using ultrasound imaging, emerging as an essential tool that provides valuable clinical insights into muscle activities and anatomical structures. Given its recent development, it is crucial to establish a standardized SMG platform for both acquisition and processing. In this study, we present a wearable and automated system designed for monitoring and analyzing muscle activity, which integrates an ultrasound transducer with motion sensors, including electromyography, mechanomyography, goniometers, and pressure sensors, all housed within a compact, wearable device. Central to this system is the SMG-master software, which facilitates real-time SMG acquisition, calculation, and visualization. The software employs deep learning strategies and post-processing algorithms for muscle structure segmentation, tracking, and quantification. Notably, our automated tracking algorithm excels in muscle boundary detection, even replacing manual initialization with efficient automated segmentation, achieving a Dice coefficient of 0.842. The Attention U-net and Unet++ models have demonstrated strong performance in fascicle segmentation, with a mean surface distance (MSD) of 0.367 mm, and orientation estimation, yielding a root mean square error (RMSE) of 3.065°, successfully reconstructing partially visible fascicles. In addition to validating the automated features of the SMG system, we observed real-time motion patterns during bicep curls and treadmill walking through multiple biomedical and SMG signals, confirming the system’s feasibility. This innovative approach offers a standardized platform for studying muscle and holds significant potential for muscle functional analysis and biofeedback training in sports science, rehabilitation, and medical fields. Ultimately, it aims to enhance the reliability of SMG as a diagnostic tool.