Background <p>Subject-specific musculoskeletal modeling for estimating muscle force or joint torque is challenging as the muscle model parameters are difficult to determine. This study deals with the methodological questions regarding the application of an electromyography (EMG)-driven model combined genetic algorithm with OpenSim application programming interface to estimate muscle force or knee torque, for the quadriceps muscle group performing maximal voluntary isometric tasks.</p> Methods <p>Knee torque is calculated from the Hill muscle model using filtered EMG data as input and compared with the torque measured by a dynamometer. The initial values of muscle parameters include optimal fiber length, tendon slack length, pennation angle at optimal fiber length, and Maximum isometric force from an OpenSim model. Eight participants performed knee isometric tasks and kept contraction for five seconds while recording surface EMG of rectus femoris, vastus lateralis and vastus medialis synchronously at angles 30°, 45°, 60°, 75° and 90°. A genetic-simulated annealing algorithm is used to tune parameters to reduce root mean square error between the predicted and measured torques.</p> Results <p>The proposed approach produced better results with an overall mean RMS error of 3.7 Nm and R² of 0.97. The two curves between the simulated and measured torques were very similar when the four parameters were adjusted simultaneously.</p> Conclusion <p>These results reveal that subject-specific and well-calibrated musculoskeletal model can better predict muscle force or knee joint torque. The proposed method can demonstrate the feasibility to generate personalized MTU parameters of the knee with high estimation accuracy and low error.</p>