Lower limb estimation from sparse landmarks using an articulated shape model

Abstract

Rapid generation of lower limb musculoskeletal models is essential for clinically applicable patient-specific gait modeling. Estimation of muscle and joint contact forces requires accurate representation of bone geometry and pose, as well as their muscle attachment sites, which define muscle moment arms. Motion-capture is a routine part of gait assessment but contains relatively sparse geometric information. Standard methods for creating customized models from motion-capture data scale a reference model without considering natural shape variations. We present an articulated statistical shape model of the left lower limb with embedded anatomical landmarks and muscle attachment regions. This model is used in an automatic workflow, implemented in an easy-to-use software application, that robustly and accurately estimates realistic lower limb bone geometry, pose, and muscle attachment regions from seven commonly used motion-capture landmarks. Estimated bone models were validated on noise-free marker positions to have a lower (p=0.001) surface-to-surface root-mean-squared error of 4.28 mm, compared to 5.22 mm using standard isotropic scaling. Errors at a variety of anatomical landmarks were also lower (8.6 mm versus 10.8 mm, p=0.001). We improve upon standard lower limb model scaling methods with shape model-constrained realistic bone geometries, regional muscle attachment sites, and higher accuracy.