Investigation of the dependence of joint contact forces on musculotendon parameters using a codified workflow for image-based modelling

Abstract

The generation of subject-specific musculoskeletal models of the lower limb has become a feasible task thanks to improvements in medical imaging technology and musculoskeletal modelling software. Nevertheless, clinical use of these models in paediatric applications is still limited for what concerns the estimation of muscle and joint contact forces. Aiming to improve the current state of the art, a methodology to generate highly personalized subject-specific musculoskeletal models of the lower limb based on magnetic resonance imaging (MRI) scans was codified as a step-by-step procedure and applied to data from eight juvenile individuals. The generated musculoskeletal models were used to simulate 107 gait trials using stereophotogrammetric and force platform data as input. To ensure completeness of the modelling procedure, muscles' architecture needs to be estimated. Four methods to estimate muscles' maximum isometric force and two methods to estimate musculotendon parameters (optimal fiber length and tendon slack length) were assessed and compared, in order to quantify their influence on the models' output. Reported results represent the first comprehensive subject-specific model-based characterization of juvenile gait biomechanics, including profiles of joint kinematics and kinetics, muscle forces and joint contact forces. Our findings suggest that, when musculotendon parameters were linearly scaled from a reference model and the muscle force-length-velocity relationship was accounted for in the simulations, realistic knee contact forces could be estimated and these forces were not sensitive the method used to compute muscle maximum isometric force.

Automatic Generation of Personalised Skeletal Models of the Lower Limb from Three-Dimensional Bone Geometries

“…In the validation part of the study the JCSs of all models were defined as in Modenese et al (2018) for consistency, but we considered other algorithms (Table 1) to enable different modelling approaches.…”

“…The articulated skeletal models were generated leveraging the MATLAB (The MathWorks, Natick, MA, USA) application programming interface (API) of OpenSim 4.1 (Seth et al, 2018): a rigid body with appropriate inertial properties (Winter, 2009) was created for each leg segment and a SpatialTransform for each joint based on the JCSs identified at step c). The individual JCSs, consistent with Modenese et al (2018), are described in Table 1. For convenience in the validation step, all rigid bodies shared a local coordinate system that was coincident with that of the medical images, as they would have in a model generated using NMSBuilder.…”

“…The models produced by the automatic tool were validated against musculoskeletal models of the lower limb generated for other purposes using NMSBuilder (manual models) and previously employed in published research (Montefiori et al, 2019b) or contributions at international conferences (Modenese et al, 2020;Modenese et al, 2019). These subject-specific models were built following the codified approach of Modenese et al (2018) and using bone geometries available in public datasets plus an in vivo MRI dataset collected with the approval of the Imperial College Research Ethics Committee. A complete description of the datasets, characterized by quality of the bone geometry meshes ranging from very good (LHDL-CT) to low (JIA-MRI), is provided in Table 3 and Figure 2. Articulated skeletal models for all datasets were generated using the STAPLE toolbox (automatic models) and their JCSs compared against those of the manual models, reporting differences in their origin location and axes orientation.…”

“…Previous studies presented methods to generate subject-specific models of the entire lower limb (Marra et al, 2015;Modenese et al, 2018) or individual joints Brito da Luz et al, 2017;Montefiori et al, 2019a;Nardini et al, 2020) and dedicated modelling tools like NMSBuilder (Valente et al, 2017a) or specialized features in the AnyBody software (Damsgaard et al, 2006) are available to implement those workflows. Nevertheless, patient-specific musculoskeletal models are currently employed in small-sized clinical applications (Falisse et al, 2020;Montefiori et al, 2019b;Pitto et al, 2019;Taddei et al, 2012;Valente et al, 2017b), mostly because the generation of each model is a time-demanding operation requiring manual intervention by specialized operators.…”

“…Nevertheless, patient-specific musculoskeletal models are currently employed in small-sized clinical applications (Falisse et al, 2020;Montefiori et al, 2019b;Pitto et al, 2019;Taddei et al, 2012;Valente et al, 2017b), mostly because the generation of each model is a time-demanding operation requiring manual intervention by specialized operators. For example, a codified approach proposed in Modenese et al (2018) reported around 10 hours to build a complete bilateral musculoskeletal model of the lower limbs from segmented bone geometries, while Scheys et al (2006) reported on average 65 minutes to define the lower limb musculature using an atlas-based semi-automated approach. We believe that validated and fully automatic workflows are of paramount importance to enable large-scale use of these computational models.…”