A Knowledge-Driven Segmentation Method for Ribs in Bone Scintigraphy Using MRF Model

Bone scintigraphy is a useful tool in diagnosing bone diseases and accurate segmentation of ribs in bones can data-sets is an essential step for accurate diagnosis. It´s a challenging task to demonstrate accurate results due to poor image quality and much extra priority should be introduced for performance improvement. To complete this task, we propose a new Markov random Field (MRF) based framework for automatic segmentation in this paper. In particularly, a series of constraints that force the coarse result of Segmentation to have smooth boundary and connected region are defined. In MRF model, segmentation problem is formed as a discrete labeling problem, so how to unify various constraints into the first-order and secondorderclique potentials of energy function is another challenge which is also studied in the paper. Our main contribution can be summarized as, 1 A framework (GSCSP-B) which is the first detailed method using MRF for segmenting ribs in Bone-Scan image is proposed.2 As to the implementation issue, a method for generating seeds in graph-cut automatically is developed for better user experience.3 Propose a series of constraints, especially the geodesic distance and boundary smoothness constraints and the method for incorporating them into MRF model is presented as well.