A Markov Random field model for bony tissue classification

Abstract

3D biomedical images constitute an indispensable source of information for the clinical diagnostic. In the case of bone structure images, a system that automatically interprets and presents a 3D shape reconstruction of the bone would be of great aid in areas such as bone remodeling, fracture prediction and prothesis design. In these tasks, external geometry needs to be precisely defined and lesions and pathologies identified. The object recognition task can rarely be carried out without knowledge on the domain. This knowledge may be introduced as a set of constraints over features and relationships between the regions obtained by means of a presegmentation. A formal scheme for the integration of this set of constraints and the solution of the interpretation problem is provided by the Markov Random Field (MRF) model. In this work we present a MRF model for identification of lesions and pathologies in the proximal tibia.