Musculoskeletal disorders (MSDs) pose a significant healthcare challenge. This study addresses diagnostic limitations by proposing a novel algorithm able to estimate key anatomical features of the bone from skin surface. The clinical application concerns the mechanical axis of the femur from thigh skin. Leveraging 50 angioscanner thigh-femur pairs, we pioneer a robust Statistical Shape Model (SSM) capturing correlations between skin and underlying femur bone. Femur inference from a known thigh, uses a Bayesian approach and, demonstrates promising results. Despite non-homogeneous meshes and no prior femur knowledge, our approach outperforms current markerless methods, suggesting its potential as a potential non-invasive musculoskeletal parameter estimation tool for surgery applications.