A Workflow to Create Personalised Musculoskeletal Models Based on Magnetic Resonance Images

Abstract

Musculoskeletal simulations typically rely on generic models that may not accurately represent individual anatomy. While personalisation based on medical images can improve model accuracy, current approaches often require time-consuming workflows to create these models. We present a semi-automatic workflow for creating personalised musculoskeletal models based on magnetic resonance imaging (MRI) that does not require bone segmentation. Our workflow uses 3D Slicer and Python scripts employing Thin-Plate Spline transformation to map 106 homologous landmarks from generic models onto participants anatomy. Generic-scaled and MRI-based models were created for eight healthy participants, and simulations were performed using the participants 3D motion capture data. MRI-based models were compared with generic-scaled models through principal component analysis, and joint kinematics and joint contact forces were analysed between both modelling approaches. Clear geometric differences existed between model types, with MRI-based models showing wider pelvises and different femur/tibia proportions. Unlike generic models, MRI-based male and female models displayed systematic differences. Despite anatomical discrepancies, joint kinematics were similar between models of the same individual, except for pelvis tilt. Muscle moment arms were generally aligned with published data from cadaver studies. MRI-based models consistently produced higher joint contact forces with greater inter-individual variation, particularly at knee joints, compared to generic-scaled models. The proposed workflow simplifies MRI-based model creation while revealing significant sensitivity of joint contact forces to individual morphology, highlighting the importance of personalisation for biomechanical analyses. Author SummaryQuestions about healthy or pathological movement patterns in humans--critical for injury prevention, rehabilitation, and sports performance--are often explored with the help of musculoskeletal modelling. This approach uses a priori defined generic models of the human musculoskeletal system to study joint moments, muscle activation patterns and joint contact forces. Typically, a generic model is scaled to match the participants dimensions linearly, which does not allow for an accurate representation of their bone and muscle morphology. We developed a semi-automatic workflow to creating magnetic resonance imaging-based personalisation that does not require bone segmentation but closely matches individual geometry with the help of a non-linear fitting function. By comparing magnetic resonance imaging-based and generic-scaled models in eight individuals, we show systematic bias inherent to one of the most popular musculoskeletal models and demonstrate the importance of model personalisation for healthy adults. Our personalisation pipeline is openly available and easy to set up, which will facilitate musculoskeletal modelling studies based on highly personalised models in clinical and research settings, potentially improving treatment planning and biomechanical assessments in the future.