Lower Limb Landmark Prediction

Abstract

Computational musculoskeletal models estimate the internal forces that produce human movement by representing the bones, muscles, and joints of the body. The accuracy of these simulations depends on how well the model reflects the anatomy of the individual. Current practice personalises models by linearly scaling a generic template. Scaling cannot accurately reproduce internal anatomical landmarks such as hip joint centres, resulting in deviations of 20-40mm from imaging-based ground truth.
This study develops a method to predict anatomical landmark positions of the lower limbs directly from external skin surface geometry. The proposed approach achieves a mean prediction error of 25.2mm across 18 lower-limb landmarks. These are of the same order as the hip joint centre deviations reported for linear scaling.
The results show that external surface geometry contains sufficient information to estimate these internal anatomical locations, and that the regression model captures this relationship across participants. Prediction accuracy is limited by the consistency of the input skin representation.
These findings show that anatomical landmarks relevant for musculoskeletal modelling can be estimated from skin geometry, providing a non-invasive approach to obtaining subject-specific anatomical information. This establishes a framework for exploring surface-based methods for musculoskeletal model personalisation.