A Preliminary Investigation Into Bio-Inspired Data Collection for Transhumeral Targeted Muscle Reinnervation Prosthetic Control

Abstract

For persons with transhumeral amputation, targeted muscle reinnervation (TMR) has unlocked the potential for inno- vative and intuitive control of myoelectric prostheses. There are many open source datasets available for training machine learning (ML) models for transradial and transhumeral prosthetic control. However, to the best of our knowledge, no datasets have been gathered in different limb positions with the intent of training models specifically for persons with tran- shumeral amputation who have undergone TMR surgery. Moreover, such a dataset is challenging to curate as TMR is still a relatively new surgical technique and there are few people with TMR. In this work we present a novel biologically-inspired protocol for collecting data from persons both with and without upper-limb amputations that can be used to train gener- alized ML models for this growing population of users. Our results from a three-participant pilot study suggest that by choosing targeted sensor placements that correspond to specific limb nerve/muscle compartment associations post-TMR surgery, we can potentially capture control-relevant muscle activation patterns from persons without limb difference that closely resemble expectations of anatomical prime movers. We expect this collection protocol to provide further utility in studying the relationship between limb positions and myocontrol signals, and differences between isotonic and isometric muscle contractions during prosthesis use, leading to a new generation of TMR-ready control solutions.