HCC: Small: Understanding Impaired Muscle Activity to Improve Human-Technology Interfaces for Pediatric Prostheses

Children born with upper limb deficiencies face several unique challenges when operating prosthetic limbs, for example their affected muscles will never have moved a complete hand. So, while many prosthetic limbs are operated by measuring activity in these muscles, the degree to which children can purposefully control the affected muscles or how best to measure this muscle activity for effective prosthetic operation is still not fully understood, which is one of the reasons advanced robotic prosthetic limbs are not widely available for children even though many "hand-like" systems are available for adults.

This research will investigate how well children born with upper limb deficiencies can control their affected muscles, and will then use that information to develop AI algorithms to recognize the movements a child wishes to achieve with their missing hand. The long-term goal is to better understand the capabilities of these children so as to enable creation of more helpful prosthetic limbs that are tailored to relevant factors such as age, gender, and learning.

Project outcomes will include datasets, algorithms, and a deeper understanding of the capabilities of children born with upper limb deficiencies, which will ultimately help medical professionals decide on prosthetic treatment options and will also lead to control techniques for other robotic devices for children, such as exoskeletons. Additional broad impact will derive from the fact that this project will support annual involvement in a multi-day summer camp program designed to help children with upper limb deficiencies learn about their capabilities.

This project will capture muscle activity in children's affected limbs by measuring muscle movements below the skin's surface and the electrical activity of these same muscles. Two human-technology interfaces will be employed: sonomyography, which uses a small ultrasound sensor, image processing, and machine learning to infer the user's intended missing-hand movements from the affected muscle deformations; and electromyography (sEMG) pattern recognition, which uses machine learning to infer the user's intended missing-hand movements from multiple sensors measuring the electrical activity of the affected muscles.

The capacity of children ages 5-17yrs to control their affected muscles will first be characterized using ultrasound imaging and sEMG measures. Post-hoc analyses of this data will then be performed to fine-tune machine learning algorithms that extract classifiable missing hand movement data from the ultrasound imaging and sEMG signals. Finally, the real-time performance of sonomyography and sEMG pattern recognition will be characterized as well as participant learning effects, as subjects perform videogame activities with these systems across multiple testing sessions.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.