Session: 05-09-02: Computational Modeling in Biomedical Applications - II

Paper Number: 99522

99522 - Estimating Upper Extremity Muscle Activations Based on Muscle Synergy Analysis and Emg-Driven Modeling 

Obtaining muscle forces as well as joint moment is a good indicator to quantify improvement of rehabilitating stroke patients. It is impossible to measure muscle forces directly and computational techniques run into the muscle redundancy problem. One way to avoid this redundancy is to obtain muscle activation directly from the muscles. Muscle electrical activity obtained through electromyography (EMG) is critical to predict joint moment with EMG-driven musculoskeletal modeling [1]. But it gets complicated when EMG data of a muscle gets missing that might have contributed substantially to the joint moment. This may happen due to lack of sufficient number of EMG channels or inability to access deeper muscles through surface EMG sensors. This research aims to predict those missing muscles’ activations through a combination of muscle synergy analysis, EMG-driven modeling and musculoskeletal modeling. Muscle synergy analysis reduces the dimensionality of muscle control by decomposing many muscle activations to a limited number of muscle synergies [2]. The method explores muscles of the upper extremity for four specific tasks in various degrees of freedom (DOFs): (1) elbow flexion-extension (1 DoF), (2) triceps kickback (1 DOF), (3) shoulder abduction-adduction (1 DOF) and (4) a reaching task (5 DOF). A healthy 30-year-old male subject (mass 103 kg, stature 188.1 cm) was recruited for the experimental data collection. For each task, surface EMG data was collected from 6 channels of the upper extremity representing 13 muscle-tendon units using surface electrodes. The motion capture data was collected for the tasks simultaneously which was later used to obtain joint moments from OpenSim inverse dynamics. The muscle parameters of the subject were obtained through the OpenSim upper extremity model optimization [3]. Among the 13 muscles, one muscle at a time (biceps, triceps or deltoid muscle) was marked missing. The remaining 12 muscles’ activation were then decomposed to time varying muscle synergy excitation and time invariant synergy weights using non-negative matrix factorization formulation. The missing muscle’s synergy weights were then predicted by minimizing the errors between joint moment obtained by the EMG-driven model and OpenSim inverse dynamics model as non-linear optimization formulation [4]. For the first task, elbow flexion-extension, the approach successfully predicted the activation of the biceps muscle but failed to predict the triceps muscle’s activation. The opposite happened for the second task, triceps kickback, where the method could predict the triceps muscle activation but not the biceps muscle. Triceps for Task 1 and biceps for Task 2 showed minimal activity due to the nature of the task. Therefore, a fourth task was performed where both biceps and triceps showed significant activity when the subject tried to reach a target point in front, involving both elbow and shoulder joint with 5 DOFs. The approach successfully predicted the target muscles’ activation. This method may contribute to the upper extremity rehabilitation research where it is critical to obtain unmeasured muscle activations.

References

1.   Meyer, A. et al., PloS one 12.7 (2017): e0179698.

2.   Safavynia, S. et al., Topics in spinal cord injury rehabilitation 17.1 (2011): 16-24.

3.   Modenese, L. et al., Journal of biomechanics 49.2 (2016): 141-148.

4. Ao, D. et al., Frontiers in computational neuroscience 14 (2020): 588943.

Presenting Author: James Yang Texas Tech University

Presenting Author Biography: Dr. Yang is a Professor at the Department of Mechanical Engineering, Texas Tech University.

Authors:

Shadman Tahmid Texas Tech University
James Yang Texas Tech University