Musculoskeletal Models from Sparse Medical Images – Towards Personalized Load Management in Sports and Medicine

Abstract

Introduction & Purpose Internal musculoskeletal loads (e.g. muscle-forces, joint-moments, and joint-contact-forces) are key indicators of locomotor loading, relevant to sports medicine and rehabilitation. Since many such variables cannot be measured directly, they are estimated using musculoskeletal modelling, driven by 3D-gait-analysis (3DGA) (marker trajectories, electromyography, ground reaction forces). Personalization of musculoskeletal models ranges from simple scaling of a generic model to fully personalized medical-image-based models; more personalization is assumed to improve accuracy but increases time/resources (Kainz & Jonkers, 2023; Modenese et al., 2021; Stansfield et al., 2025). We present a semi-automatic workflow that personalizes models from sparse images typically collected in clinical settings and evaluate muscle-lengths against a scaled-generic model and a full-image-based model during gait. Methods 3DGA and magnetic-resonance-imaging (MRI) data were collected for one individual (female, age: 16years, mass: 52.1kg, height: 172cm) with ethics approval and parental consent. Gait condition: overground-walking, barefoot, self-selected speed. Bones (pelvis, femur, tibia/fibula) were segmented from MRI using 3Dslicer. To imitate sparse clinical images, we retained proximal/distal 15% of femur/tibia/fibula and distal 45% of pelvis. A Statistical Shape Model (SSM) (Carman et al., 2022) was fit to the sparse segmentations using a genetic algorithm in MATLAB (Conn et al., 1991, 1997; Goldberg, 1989) to reconstruct full bones automatically. Three OpenSim models were generated based on the RajagopalLaiUhlrich-model (Rajagopal et al., 2016; Uhlrich et al., 2022): (1) Generic-scaled: using OpenSim’s Scaling Tool (Delp et al., 2007) (automated); (2) Full-personalized: via OpenSim Creator’s Model Warper (Kewley et al., 2026) using full segmentations, as previously described by Koller et al., 2026 (semi-automatic involving manual landmark setting); (3) Sparse-personalized: same warping pipeline applied to SSM-reconstructed bones (Figure 1). Inverse Kinematics and Muscle Analysis yielded joint-angles and muscle-tendon-lengths. Similarity was assessed qualitatively and by root mean squared difference (RMSD) over the gait cycle. Results Waveforms of muscle-tendon-lengths showed similar patterns for all models (Figure 1). Sparse-model was more similar in magnitudes (lower RMSD) to the Full-model than the Generic-model for 85% of muscles. Exceptions were found predominantly in adductors and hip flexors, for which the Generic was closer to the Full than the Sparse-model (Figure 1), likely reflecting pelvis reconstruction challenges when proximal anatomy is missing. Maximum RMSD was 2.5cm for Full-vs.-Sparse-comparison (iliacus), and 5.3cm for Full-vs.-Generic-comparison (tensor fasciae latae). Discussion Our sparse-image-based personalization approximated full-image models for most muscle-tendon-length waveforms using routine-like MRI subsets. Remaining errors highlight the need for stronger pelvis constraints/proximal landmarks. Relative to prior work, we: (i) reconstruct subject-specific bones from sparse MRI via SSM; (ii) integrate this with a semi-automatic OpenSim warping pipeline; and (iii) benchmark against generic and fully image-based models. Upon validation on larger datasets, reconstructing full bone geometry from sparse clinical images enables retrospective modelling in cohorts lacking research-grade scans. As an N=1 pilot, findings are preliminary. Conclusion Sparse-image-based personalization primarily advances retrospective musculoskeletal modeling from routine clinical imaging, broadening applicability across sites and age groups. Further work will validate in a larger, diverse cohort and extend outcomes (joint-kinematics, joint-contact-forces, muscle-forces), with targeted improvements to pelvis/hip reconstruction. References Carman, L., Besier, T. F., & Choisne, J. (2022). Morphological variation in paediatric lower limb bones. Scientific Reports, 12(1), 3251. https://doi.org/10.1038/s41598-022-07267-4 Conn, A. R., Gould, N. I. M., & Toint, P. (1991). A Globally Convergent Augmented Lagrangian Algorithm for Optimization with General Constraints and Simple Bounds. SIAM Journal on Numerical Analysis, 28(2), 545–572. https://doi.org/10.1137/0728030 Conn, A. R., Gould, N., & Toint, Ph. L. (1997). A globally convergent Lagrangian barrier algorithm for optimization with general inequality constraints and simple bounds. Mathematics of Computation, 66(217), 261–289. https://doi.org/10.1090/S0025-5718-97-00777-1 Delp, S. L., Anderson, F. C., Arnold, A. S., Loan, P., Habib, A., John, C. T., Guendelman, E., & Thelen, D. G. (2007). OpenSim: Open-Source Software to Create and Analyze Dynamic Simulations of Movement. IEEE Transactions on Biomedical Engineering, 54(11), 1940–1950. https://doi.org/10.1109/TBME.2007.901024 Goldberg, D. E. (1989). Genetic Algorithms in Search, Optimization & Machine Learning. Addison-Wesley. Kainz, H., & Jonkers, I. (2023). Imaging-based musculoskeletal models alter muscle and joint contact forces but do not improve the agreement with experimentally measured electromyography signals in children with cerebral palsy. Gait & Posture, 100, 91–95. https://doi.org/10.1016/j.gaitpost.2022.11.019 Kewley, A., Van Beesel, J., & Seth, A. (2026). OpenSim Creator: An Interactive User Interface for Building OpenSim Musculoskeletal Models. Journal of Open Source Software, 11(117), 8284. https://doi.org/10.21105/joss.08284 Koller, W., Wallnöfer, E., Rathmair, L., Egner, C., Kitir, K., & Kainz, H. (2026). Foot progression angle modifications alter hip joint contact forces and femoral growth plate mechanics in children. Frontiers in Bioengineering and Biotechnology, 14, 1799347. https://doi.org/10.3389/fbioe.2026.1799347 Modenese, L., Barzan, M., & Carty, C. P. (2021). Dependency of lower limb joint reaction forces on femoral version. Gait & Posture, 88, 318–321. https://doi.org/10.1016/j.gaitpost.2021.06.014 Rajagopal, A., Dembia, C. L., DeMers, M. S., Delp, D. D., Hicks, J. L., & Delp, S. L. (2016). Full-Body Musculoskeletal Model for Muscle-Driven Simulation of Human Gait. IEEE Transactions on Biomedical Engineering, 63(10), 2068–2079. https://doi.org/10.1109/TBME.2016.2586891 Stansfield, E., Koller, W., Gonçalves, B., & Kainz, H. (2025). A Workflow to Create Personalised Musculoskeletal Models Based on Magnetic Resonance Images. Biophysics. https://doi.org/10.1101/2025.07.23.666289 Uhlrich, S. D., Jackson, R. W., Seth, A., Kolesar, J. A., & Delp, S. L. (2022). Muscle coordination retraining inspired by musculoskeletal simulations reduces knee contact force. Scientific Reports, 12(1), 9842. https://doi.org/10.1038/s41598-022-13386-9