Classification Performance and Feature Space Characteristics in Individuals With Upper Limb Loss Using Sonomyography.

Area of Science:

• Biomedical Engineering
• Rehabilitation Technology
• Prosthetics and Orthotics

Background:

• Sonomyography (ultrasound-based muscle deformation sensing) is an emerging technology for upper limb prosthesis control.
• Understanding the learning curve for sonomyography is crucial for its clinical adoption.

Purpose of the Study:

• To characterize user performance and learning over time with sonomyography for upper limb prosthesis control.
• To assess the impact of initial exposure, biofeedback training, and repeated testing on sonomyography proficiency.

Main Methods:

• Collected ultrasound images of hand gestures from individuals with transradial limb loss.
• Evaluated performance via offline classification accuracy and feature space metrics across three experimental scenarios.
• Scenarios included initial exposure, training with biofeedback, and testing on different days.

Main Results:

• High classification accuracy (96.2 ± 5.9%) was achieved during the initial exposure to sonomyography.
• Performance did not significantly improve with biofeedback training or across different testing days.
• While classification accuracy remained stable, some sonomyography signal pattern metrics in feature space did change.

Conclusions:

• Users achieve high proficiency with sonomyography immediately upon initial exposure.
• Subsequent training or testing does not lead to significant performance improvements in classification accuracy.
• This suggests sonomyography may offer rapid usability for upper limb prosthesis control.