Two-thirds of repetitive strain injuries affect the wrist joint. Although force is believed to be one of the major factors, the forces involved in wrist movements have not been thoroughly characterized in vivo. Computer simulations with a musculoskeletal model of the wrist have been used to estimate wrist muscle forces, but only at maximum voluntary contraction and only involving a single degree of freedom (DOF). In this study we present a method for creating a subject-specific model that can be used to estimate muscle forces and joint torques in both degrees of freedom of the wrist over a range of torques applicable to activities of daily living. Ten young, healthy subjects applied three levels of isometric wrist torque (about 7, 15, and 25% of maximum torque) in combinations of wrist flexion-extension and radial-ulnar deviation while joint torque in both DOF and surface electromyograms (sEMG) in the five major wrist muscles were measured. To find subject-specific parameters, we followed a two-step process. First, a pre-existing, generic musculoskeletal model of the wrist was scaled to individual subjects' height. Second, we compared joint torques predicted from measured sEMG using forward simulations of muscular dynamics to measured torques and minimized this error to optimize for subject-specific model parameter values. The model parameters optimized were the maximum isometric force and tendon slack length of each muscle. Optimization constraints were added to ensure physiologically plausible combinations of parameter values. The optimization produced model parameters that 1) were in a reasonable physiological range for each test subject and 2) significantly improved the accuracy of the model’s torque estimation. Scaling the generic model reduced the root mean squared (RMS) error between predicted and measured joint torques by 2.8±4.6% (mean±SD), whereas optimizing the scaled model further reduced the RMS error by 51.4±18.9% for the torque level at which the model was optimized. Testing the optimized model at other torque levels still significantly reduced the error between predicted and measured torques compared to the scaled model (43.7±28.0% and 25.0±24.0% for lower and higher torque levels, respectively). The mean error between predicted and measured torque was 0.23±0.04, 0.30±0.04, and 1.17±0.26 Nm at the low-, mid-, and high-torque levels, respectively. The method generally reduced the error in flexion-extension (FE) more than radial-ulnar deviation (RUD), likely in part because sEMG and torque were larger in FE than in RUD. Optimizing for subject-specific model parameters significantly improved prediction over both the generic and scaled models, in both degrees of freedom of the wrist, and at all three torque levels. The presented method for creating subject-specific models can be used in future studies to quantify muscle forces and joint torques of natural wrist movements in vivo.



College and Department

Ira A. Fulton College of Engineering and Technology; Mechanical Engineering



Date Submitted


Document Type





wrist, force, torque, estimation, subject-specific, musculoskeletal modeling, OpenSim



Included in

Engineering Commons