Using subject-specific muscle parameters to compare muscle forces and activation between an EMF-driven and OpenSim musculoskeletal model

Date
2010
Journal Title
Journal ISSN
Volume Title
Publisher
University of Delaware
Abstract
Experimental gait analysis was developed to analyze and calculate mechanical aspects such as joint kinetics and kinematics to understand the characteristics that lead to human walking. Unfortunately, gait analysis is limited because it alone cannot determine individual, in vivo muscle function. Redundant systems are created from multiple muscles spanning one joint, thus eliminating the possibility to directly calculate muscle activity. The recent use of musculoskeletal modeling has allowed researchers to extend the functionality of gait analysis, predict individual muscle activity and forces, and gain insight to the underlying neuromuscular function. Many different optimization approaches with varying degrees of complexity exist to estimate individual muscle forces and activation. Previous studies have used electromyograms (EMG) to derive subject-specific muscle parameters that match inverse dynamics, while others use a generic model and redistribute muscle forces across synergistic muscles based on a particular cost function. Because generic musculoskeletal models are limited by their lack of specificity, we hoped to include subject-specific muscle parameters that would lead to improved force predictions. Prior sensitivity studies have determined the effect of altering one parameter within a specified range and calculating the difference in forces. However, to date, a global set of muscle parameters has not been implemented within a forward dynamic simulation. In this thesis, we compared forward simulations of healthy subjects using generically scaled and subject-specific muscle parameters. Specifically, differences in muscle force and activation using optimized values of maximum isometric force, optimal fiber length, and tendon slack length were examined. An EMG-driven model was used as a means to determine subject-specific muscle parameters. OpenSim, an open-source software, was used to provide a computationally efficient forward simulation based on a computed muscle control algorithm. Output forces from subject-specific EMG-driven and OpenSim forward dynamic simulation models were compared to determine differences between the two fundamentally contrasting models. The EMG-driven model matched the experimental inverse dynamic joint moments well with average r2 values of 0.924 and 0.954 for the knee and ankle, respectively, giving confidence that the optimized muscle parameters are indicative of true in vivo values and provide good predictive ability for novel tasks. OpenSim forward dynamic simulations produced lower average activations with optimized subject-specific parameters than generically scaled parameters for all muscles except the BFLH. Average ratios of muscle force from the scaled to optimized OpenSim models were below 1.0 for all muscles except the Sol, indicating larger optimized force values. Because both models apply the same cost function, differences in force are due solely to changes in muscle parameters. Similarly, a comparison of the muscle forces from the EMG-driven to optimized OpenSim models revealed ratios below 1.0 for all muscles except the Sol. Fundamental differences in the model cost functions and their calculation of forces are responsible for these differences as the muscle physiology between the two models is held constant. Although the models showed varying individual muscle forces, it is still encouraging that similar net joint forces were produced at both joints, and in particular, the ankle. r2 values for the net ankle forces ranged between 0.917 and 0.999. Timing and magnitude were also similar for the ankle with differences in peak values only ranging from 24 to 592N. This thesis examined the global use of subject-specific parameters within an EMG-driven and OpenSim forward dynamic model. While an exact cause-and-effect relationship is difficult to determine for each particular change in a muscle parameter, we believe that the use of subject-specific parameters will improve the utility of forward dynamic simulations and can be extended to gain insight into the deficiencies in neuromuscular function of pathologic patients.
Description
Keywords
Citation