Muscle Synergies as the Basis for the Control of a Hand Prothesis
pdf
In this ongoing study, we are exploring the suitability of a novel technique based on the estimation of muscle synergies to control a hand prosthesis. The objective of the study is to predict thirteen distinct hand grasp movements using EMG data.
The method relies on projecting the EMG data using a basis derived via analysis of the muscle synergies.
Five healthy subjects performed a set of thirteen distinct hand grasp movements while sEMG data from seven muscles have been recorded.
The non-negative matrix factorization algorithm was applied to the sEMG envelopes to derive four muscle synergies, which we found to be sufficient to reconstruct the muscle activation patterns with an average R2 - across all sEMG channels -of 0.85 and a minimum R2 of 0.70. These four muscle synergies were used as a projection basis and the output of the algorithm was analyzed using a template matching technique to detect the performance of the hand grasp movements.
The average (across grasps) classifier accuracy varied from 87.4 (subject #1) to 94.9% (subject #3).
Submitted by Paolo Bonato
on
In this ongoing study, we are exploring the suitability of a novel technique based on the estimation of muscle synergies to control a hand prosthesis. The objective of the study is to predict thirteen distinct hand grasp movements using EMG data.
The method relies on projecting the EMG data using a basis derived via analysis of the muscle synergies.
Five healthy subjects performed a set of thirteen distinct hand grasp movements while sEMG data from seven muscles have been recorded.
The non-negative matrix factorization algorithm was applied to the sEMG envelopes to derive four muscle synergies, which we found to be sufficient to reconstruct the muscle activation patterns with an average R2 - across all sEMG channels -of 0.85 and a minimum R2 of 0.70. These four muscle synergies were used as a projection basis and the output of the algorithm was analyzed using a template matching technique to detect the performance of the hand grasp movements.
The average (across grasps) classifier accuracy varied from 87.4 (subject #1) to 94.9% (subject #3).