Validity of a musculoskeletal model using two different geometries for estimating hip contact forces during normal walking

Prediction of in vivo hip contact forces during common activities of daily living using a segment-based musculoskeletal model

Background: Quantifying in vivo hip muscle and contact forces during activities of daily living (ADL) provides valuable information for diagnosis and treatment of hip-related disorders. The objective of this study was to utilize Freebody, a segment-based musculoskeletal model, for the prediction of hip contact forces using a novel objective function during seven common ADLs and validate its performance against the publicly available HIP98 dataset. Methods: Marker data, ground reaction forces, and hip contact forces during slow, normal, and fast walking, stair ascent and descent, and standing up and sitting down were extracted for 3 subjects from the HIP98 dataset. A musculoskeletal anatomical dataset was scaled to match the dimensions of each subject, and muscle and hip contact forces were estimated by minimizing a novel objective function, which was the summation of the muscle stresses squared and body weight-normalised hip contact force. The accuracy of predictions were quantified using several metrics, and muscle forces were qualitatively compared to experimental EMGs in the literature. Results: FreeBody predicted the hip contact forces during the ADLs with encouraging accuracy: The root mean squared error of predictions were 44.0 ± 8.5, 47.4 ± 6.5, and 59.8 ± 7.1% BW during slow, normal, and fast walking, 44.2 ± 16.8% and 53.3 ± 12.2% BW for stair ascent and descent, and 31.8 ± 8.2% and 17.1 ± 5.0% BW for standing up and sitting down, respectively. The error in prediction of peak hip contact forces were 14-18%, 24-28%, 17-35% for slow, normal, and fast walking, 7-25% and 15-32% in stair ascent and descent, and around 10% for standing up and sitting down. The coefficient of determination was larger than 0.90 in all activities except in standing up (0.86 ± 0.08). Conclusion: This study has implemented a novel objective function in a segment-based musculoskeletal model, FreeBody, for the prediction of hip contact forces during a large range of ADLs. The model outputs compare favourably for all ADLs and are the best in standing up and sitting down, while muscle activation patterns are consistent with experimental EMGs from literature. This new objective function addresses one of the major limitations associated with musculoskeletal models in the literature, namely the high non-physiological predicted hip joint contact forces.

Knee loading in OA subjects is correlated to flexion and adduction moments and to contact point locations

This study evaluated the association of contact point locations with the knee medial and lateral contact force (F_med, F_lat) alterations in OA and healthy subjects. A musculoskeletal model of the lower limb with subject-specific tibiofemoral contact point trajectories was used to estimate the F_med and F_lat in ten healthy and twelve OA subjects during treadmill gait. Regression analyses were performed to evaluate the correlation of the contact point locations, knee adduction moment (KAM), knee flexion moment (KFM), frontal plane alignment, and gait speed with the F_med and F_lat. Medial contact point locations in the medial–lateral direction showed a poor correlation with the F_med in OA (R² = 0.13, p = 0.01) and healthy (R² = 0.24, p = 0.001) subjects. Anterior–posterior location of the contact points also showed a poor correlation with the F_med of OA subjects (R² = 0.32, p < 0.001). Across all subjects, KAM and KFM remained the best predictors of the F_med and F_lat, respectively (R² between 0.62 and 0.69). Results suggest different mechanisms of contact force distribution in OA joints. The variations in the location of the contact points participate partially to explains the F_med variations in OA subjects together with the KFM and KAM.

Automated Generation of Three-Dimensional Complex Muscle Geometries for Use in Personalised Musculoskeletal Models

The geometrical representation of muscles in computational models of the musculoskeletal system typically consists of a series of line segments. These muscle anatomies are based on measurements from a limited number of cadaveric studies that recently have been used as atlases for creating subject-specific models from medical images, so potentially restricting the options for personalisation and assessment of muscle geometrical models. To overcome this methodological limitation, we propose a novel, completely automated technique that, from a surface geometry of a skeletal muscle and its attachment areas, can generate an arbitrary number of lines of action (fibres) composed by a user-defined number of straight-line segments. These fibres can be included in standard musculoskeletal models and used in biomechanical simulations. This methodology was applied to the surfaces of four muscles surrounding the hip joint (iliacus, psoas, gluteus maximus and gluteus medius), segmented on magnetic resonance imaging scans from a cadaveric dataset, for which highly discretised muscle representations were created and used to simulate functional tasks. The fibres’ moment arms were validated against measurements and models of the same muscles from the literature with promising outcomes. The proposed approach is expected to improve the anatomical representation of skeletal muscles in personalised biomechanical models and finite element applications.

Knee medial and lateral contact forces in a musculoskeletal model with subject-specific contact point trajectories

Contact point (CP) trajectory is a crucial parameter in estimating medial/lateral tibio-femoral contact forces from the musculoskeletal (MSK) models. The objective of the present study was to develop a method to incorporate the subject-specific CP trajectories into the MSK model. Ten healthy subjects performed 45 s treadmill gait trials. The subject-specific CP trajectories were constructed on the tibia and femur as a function of extension-flexion using low-dose bi-plane X-ray images during a quasi-static squat. At each extension-flexion position, the tibia and femur CPs were superimposed in the three directions on the medial side, and in the anterior-posterior and proximal-distal directions on the lateral side to form the five kinematic constraints of the knee joint. The Lagrange multipliers associated to these constraints directly yielded the medial/lateral contact forces. The results from the personalized CP trajectory model were compared against the linear CP trajectory and sphere-on-plane CP trajectory models which were adapted from the commonly used MSK models. Changing the CP trajectory had a remarkable impact on the knee kinematics and changed the medial and lateral contact forces by 1.03 BW and 0.65 BW respectively, in certain subjects. The direction and magnitude of the medial/lateral contact force were highly variable among the subjects and the medial-lateral shift of the CPs alone could not determine the increase/decrease pattern of the contact forces. The suggested kinematic constraints are adaptable to the CP trajectories derived from a variety of joint models and those experimentally measured from the 3D imaging techniques.

Influence of the Level of Muscular Redundancy on the Validity of a Musculoskeletal Model

While recent literature has clearly demonstrated that an extensive personalization of the musculoskeletal models was necessary to reach high accuracy, several components of the generic models may be further investigated before defining subject-specific parameters. Among others, the choice in muscular geometry and thus the level of muscular redundancy in the model may have a noticeable influence on the predicted musculotendon and joint contact forces. In this context, the aim of this study was to investigate if the level of muscular redundancy can contribute or not to reduce inaccuracies in tibiofemoral contact forces predictions. For that, the dataset disseminated through the Sixth Grand Challenge Competition to Predict In Vivo Knee Loads was applied to a versatile 3D lower limb musculoskeletal model in which two muscular geometries (i.e., two different levels of muscular redundancy) were implemented. This dataset provides tibiofemoral implant measurements for both medial and lateral compartments and thus allows evaluation of the validity of the model predictions. The results suggest that an increase of the level of muscular redundancy corresponds to a better accuracy of total tibiofemoral contact force whatever the gait pattern investigated. However, the medial and lateral contact forces ratio and accuracy were not necessarily improved when increasing the level of muscular redundancy and may thus be attributed to other parameters such as the location of contact points. To conclude, the muscular geometry, among other components of the generic model, has a noticeable impact on joint contact forces predictions and may thus be correctly chosen even before trying to personalize the model.