Ultrasound Imaging and Machine Learning to Detect Missing Hand Motions for Individuals Receiving Targeted Muscle...

Area of Science:

• Biomedical Engineering
• Rehabilitation Medicine
• Surgical Innovation

Background:

• Targeted muscle reinnervation (TMR) evolved from prosthetic control to pain management post-amputation.
• Modern TMR surgical approaches may compromise conventional electromyography (EMG) signal quality for prosthetic control.
• Factors like muscle depth and crosstalk can hinder EMG-based bionic limb function.

Purpose of the Study:

• To investigate if current TMR surgical techniques still support effective bionic prosthetic control.
• To explore novel methods for decoding residual muscle activity after TMR.
• To assess the feasibility of using sonomyography and machine learning for prosthetic control in TMR patients.

Main Methods:

• Recruited 4 participants with TMR and varying upper limb loss.
• Utilized ultrasound imaging and machine learning (sonomyography) to measure muscle activity.
• Applied a K-nearest neighbors algorithm to classify attempted hand movements based on muscle deformation patterns.

Main Results:

• Attempted missing hand movements generated distinct deformation patterns in reinnervated muscles.
• The sonomyography and machine learning approach accurately predicted 4-10 hand movements per participant.
• Prediction accuracy ranged from 83.3% to 99.4% across participants.

Conclusions:

• Despite shifts in surgical goals, TMR retains significant potential for advanced bionic prosthetic control.
• Sonomyography offers a promising alternative to EMG for decoding motor intent in TMR patients.
• This study validates the utility of TMR for both prophylactic benefits and restoring intuitive prosthetic function.