Automated creation and tuning of personalised muscle paths for OpenSim musculoskeletal models of the knee joint

Abstract

Computational modelling is an invaluable tool for investigating features of human locomotion which cannot be measured except for highly invasive techniques. Recent research has focussed on creating personalised musculoskeletal models using medical imaging. Although progress has been made, robust definition of two critical model parameters remains challenging; (i) tibiofemoral (TF) and patellofemoral (PF) joint motions, and (ii) muscle tendon unit (MTU) pathways and kinematics. The aim of this study was to develop a highly automated framework for the definition of personalised musculoskeletal models, consisting of personalised bone geometries, TF and PF joint models, and MTU pathways and kinematics. Informed from medical imaging, personalised rigid body 6 degree-of-freedom TF and PF joint models were created. Using atlas-and optimisation-based methods, personalised MTU pathways and kinematics were created with the aim of preventing MTUs penetrating bones and smooth MTU kinematics that follow literature patterns. This framework was deployed in Musculoskeletal Atlas Project Client for 6 participants, where 5 models with incremental personalisation were created. Three comparisons were made; (i) non-optimised and optimised models with nonpersonalised joint models; (ii) non-optimised and optimised models with personalised joint models; (iii) both aforementioned optimised models. Following optimisation, significant improvements were consistently shown in pattern similarity to literature data. Although not significant, improvements to all metrics were shown with reduced MTUs penetrating bone and increased kinematic smoothness. Comparing optimised models, no significant difference was identified for any evaluation metric, with a trend for more desirable results in the nonpersonalised joint models in most evaluation metrics. Methods Gait data testing and magnetic resonance imagingData were collected at Griffith University as part of an ongoing project (Ethics reference: PES/36/10/HREC). A subset of 6 participants were selected from a larger cohort to span a large range of age, height, and mass (Table 1). Participants had no history of musculoskeletal injury, trauma or lower-limb surgeries. Each participant provided their written and informed consent prior to undergoing comprehensive motion capture (MOCAP) and medical imaging sessions.Three-dimensional (3D) marker positions during a static calibration trial were converted from standard MOCAP format (i.e., .c3d) to standard OpenSim format (i.e., .trc) using MOtoNMS (Mantoan et al. 2015) for use in the MAP-Client.Each participant underwent magnetic resonance imaging (MRI) of their lower-limbs at a local radiology clinic (QScan Southport, QLD, Australia) performed on the preceding or same day as the MOCAP session. Axial T1-weighted 3D fast field echo sequences were acquired bilaterally from above the iliac crest to below the toes, while the participant lay supine in a 3 T

Experimental recommendations for estimating lower extremity loading based on joint and activity

“…We assumed that our musculoskeletal models represented accurate segment masses for each subject. While we didn’t employ advanced scaling techniques (Killen et al, 2021; Valente et al, 2017), our sensitivity analysis tested extremes that far exceeded any plausible differences between the segment masses defined in the musculoskeletal model and subject-specific segment masses. We focused our analyses on instances when the foot is contacting the ground because anytime the foot is off the ground, joint loading errors will be directly proportionate to segment mass errors.…”

Between-Limb Symmetry in ACL and Tibiofemoral Contact Forces in Athletes After ACL Reconstruction and Clearance for Return to Sport

“…While keeping the translational degrees of freedom as a function of the knee flexion angle, we extended the 1 degree of freedom knee with knee varus-valgus and knee internal-external rotation and added the knee ligaments. Ligament origin and insertion points, described in the model of Xu et al, 67 were registered in the Catelli model using host mesh fitting, 27 and ligament properties were the same as described by Xu et al 67 We assumed that graft properties after reconstruction were similar to those of the native ligament and that the ligaments produced passive forces during elongation, given joint kinematics. The maximum isometric force of each muscle was tripled to allow the generation of high forces required to perform the dynamic movements.…”

An EMG-Assisted Muscle-Force Driven Finite Element Analysis Pipeline to Investigate Joint- and Tissue-Level Mechanical Responses in Functional Activities: Towards a Rapid Assessment Toolbox

“…Additionally, we plan to incorporate automated creation and tuning of personalized muscle paths [10], [81] into the workflow developed in the current study. Utilizing image processing, machine learning, and surrogate modeling and techniques [10], our further studies aim to reduce the simulation time even towards real-time EMG-assisted MS-FE analyses (e.g., as done for the Achilles tendon [82]), and to make the whole pipeline automatic (i.e., scaling and morphing of the MS and FE models' geometries) [81].…”