Segmentation of Proximal Femur in 3D Magnetic Resonance Images for Detection of Cam Type FAI

Abstract

Segmentation of osseous structures from clinical MR images has remained a challenging task for many years now due to inevitable acquisition artifacts and inhomogeneous intensity of bones. To come up with the associated challenges, we devised a novel parametric deformable model framework for segmentation of 3D magnetic resonance (MR) images in computer-aided diagnosis (CAD) of cam type femoro-acetabular impingement (FAI). Our framework has two phases: (i) we introduce radial basis function (RBF) interpolation for semi-automatic piecewise registration of a proximal femur atlas model to the region of interest (ROI) using landmarks displacements and (ii) a parametric deformable model for coarse-to-fine level segmentation based on acting internal and external forces. We tested our segmentation scheme on a 3D synthetic image data as well as clinical datasets of MR hip images with different resolutions and validated the results of real data in comparison with three related parametric deformable model techniques. Accordingly, we found our methodology to be robust against artifacts and intensity inhomogeneity existing in high and low-resolution images and relatively resistant to under- and over- segmentation problems.