Neural shape models enable cohort-scale simulation of knee cartilage mechanics

Knee osteoarthritis affects over 350 million people worldwide, is a leading cause of disability, and carries substantial costs. Cartilage mechanical load is believed to play a critical role in osteoarthritis pathophysiology, yet clinical biomechanics research uses external measures of joint loading that do not capture tissue-level load. We developed a fully-automated pipeline that builds biomechanical knee models from a single MRI using deep learning-based segmentation and neural shape models, and elastic foundation contact to simulate gait. Applied to 756 knees from the Osteoarthritis Initiative, each simulated under the same 10 standardized gaits, the pipeline converged in 99.6% of 7,560 simulations. Medial tibial pressure decreased from healthy to incident to mild osteoarthritis as disease-related shape changes enlarged contact area, an effect reproduced in simulations of synthetic knees. Yet within knees, medial cartilage thinned fastest over six years in regions of highest pressure, with stronger relationships in incident and mild osteoarthritis knees. Automated simulations provided cohort-scale, tissue-level support for mechanical theories of osteoarthritis progression and a framework for prospectively quantifying cartilage load to guide mechanical interventions.