A prior model for bone microarchitecture reconstruction with a very limited number of projections

The goal of this work was to find a new parametric model suited to the representation of three-dimensional (3D) bone microarchitecture that can be used as a prior model in tomographic reconstruction or image segmentation. For this purpose, Markov random fields (MRFs) are well adapted. Bone microarchitecture is typically organized as a pseudoperiodic arrangement of parallel plates with a given thickness (∼ 100 μm), and spacing (∼ 700 μm). As standard Ising MRFs could not represent the complexity of bone, we investigated energy functions based on the design of a neighborhood system including local and distant interactions. The parameters of this MRF model were related to the standard morphometric parameters of bone microarchitecture.