How complex should lower-limb joint models be for subject-specific musculoskeletal modeling applications?

Main topics: Musculoskeletal modelling; Mathematical simulation in human movement science; Orthopaedics Introduction and aim: Subject-specific musculoskeletal modelling can be applied to study musculoskeletal disorders, providing a valuable approach to calculating muscle and joint forces during movement. The choice of the lower-limb joint models can significantly affect joint contact loads during walking [1]; however, the degree of joint model complexity according to the application has not been extensively studied. The aim of this study is twofold: first, to compare muscle and joint contact forces during common daily activities, using four subject-specific models including different joint models; second, to investigate how the uncertainties in the identification of subject-specific kinematic constraints affect the model outputs using a probabilistic modelling approach. Patients/materials and methods: Experimental data were collected on a healthy volunteer subject, including lower-body MRI scans and gait data (marker trajectories, ground reaction forces and EMG recordings) during walking, chair rising, stair ascending and descending. A baseline musculoskeletal model (M1) was created, which included a 7-segment 3D articulated linkage with spherical joints at the hips, hinge joints at both knees and ankles [2], and 84 musculotendon actuators. More complex knee and ankle models were built upon the baseline model as follows: planar hinge and universal joint [3] (M2), modified spherical and universal joint [4] (M3), and subject-specific planar 4-bar-linkage knee and ankle [5,6] (M4). To test the sensitivity to parameter identification, five operators virtually palpated the anatomical landmarks necessary to create the M4 model. The joint parameters were statistically sampled according to inter-operator palpation variability, and a set of perturbed models was then created. Simulations of movements solving a typical inverse dynamics and static optimization problem were performed using OpenSim, to calculate muscle and joint contact forces. The predicted forces were compared between the different motor tasks and the robustness to the joint parameter identification was analyzed. Results: M2 and M4 tended to predict lower knee forces (Table 1), as coupling planar translations to flexion increased knee extensor moment arms. Forces obtained using M4 were more robust to palpation uncertainties at the knee than at the ankle, where, for high dorsiflexion angles (such as in stair descending), M4 predicted unrealistically high contact forces with high variability (Table 1). Hip contact forces were slightly affected by the degree of knee and ankle joint complexity, except for the M3 model, in all simulated motor tasks. Table 1. peak joint contact loads (BW). M4: mean ± SD. Gait Chair rising Stair ascending Stair descending Hip Knee Ankle Hip Knee Ankle Hip Knee Ankle Hip Knee Ankle M1 3.63 3.40 5.44 3.15 2.96 1.21 3.68 5.30 4.88 4.08 5.13 7.04 M2 3.77 3.80 5.82 3.05 2.54 1.46 3.61 4.61 4.99 4.07 4.80 7.57 M3 4.47 5.80 5.58 3.23 2.59 1.50 3.88 5.87 5.05 3.83 4.39 7.18 M4 4.00 ± 0.14 4.45 ± 0.25 4.96 ± 0.26 3.14 ± 0.02 2.26 ± 0.07 1.12 ± 0.07 3.75 ± 0.01 4.14 ± 0.14 4.61 ± 0.32 4.11 ± 0.01 4.91 ± 0.99 9.28 ± 2.77 Discussion and conclusions: The models predicted different joint loads, and behaved differently in different motor tasks. In the absence of direct validation methods of the predicted forces, the choice of implementing complex subject-specific joints should be justified only when the predictions are robust to the uncertainties in parameter identification. Therefore, before implementing subject-specific kinematic constraints, it is advisable to analyze the sensitivity of predictions over the joint range of motion and, if this choice appears weak (as in the stair descending task), opting for less complex, but more robust and reproducible joint models.