Random Point Sub-sampling in Rigid Registration of 2D Proximal Femur Surfaces

Abstract

This paper presents a study on evaluation of the point sets obtained via Random Point Sub-sampling (RPS) in rigid registration of 2-dimensional (2D) proximal femur surfaces. In the first stage of the proposed study, the edges of the proximal femurs, which were manually segmented from bilateral hip magnetic resonance (MR) images, were extracted and the surface information of these proximal femurs was obtained as point clouds. In the further stages, the sizes of the existing point clouds were reduced by performing RPS on point clouds describing the proximal femurs, and the right and left proximal femurs were registered rigidly with the Iterative Closest Point (ICP) method. In this way, it was aimed to register the proximal femur surfaces faster and more successfully over a small number of points. In performance evaluations performed on 13 hip MR images of 13 patients diagnosed with Legg-Calve-Perthes (LCP) disease, it was ensured that proximal femur surfaces were successfully registered faster than the normal case by using the points obtained via RPS. The proposed approach can be applied to different problems such as shape quantification and shape modelling/matching of proximal femurs. Furthermore, it is also promising for the real-time clinical applications in the related fields.