Modélisation statistique multi-organes : application à l’inférence osseuse à partir de données cutanées

This thesis lies within the field of musculoskeletal morphometric modeling and aims to estimate quantitative bone characteristics from skin morphology, without direct reliance on internal medical imaging. While the reconstruction of bone geometry represents an essential intermediate step, the main objective is the extraction of relevant anatomical and biomechanical landmarks — such as axes, planes, and characteristic points of the femur — enabling a geometric and functional analysis directly from the external surface of the thigh. This work relies on the construction of a Statistical Shape Model (SSM) that jointly describes the geometric variability of the thigh and femur. The model was developed from a database of lower-limb CT scans collected at the CHU Morvan Hospital in Brest. Two complementary inference strategies were explored: a stochastic approach based on Markov Chain Monte Carlo (MCMC) methods, and a deterministic approach using geometric regression. These methods allow predicting the internal femoral geometry from the external surface and automatically extracting clinically relevant morphometric parameters. Experimental evaluations demonstrate the feasibility and accuracy of this non-invasive approach, paving the way for the personalization of biomechanical models and for clinical analysis without heavy imaging. The results represent a step toward predictive and accessible digital medicine, capable of exploiting superficial data (3D surface scans, depth cameras, photogrammetry) to estimate internal anatomical characteristics.