Accurate registration of dynamic contrast-enhanced breast mr images with robust estimation and linear programming

Abstract

Accurate registration of the dynamic contrast-enhanced (DCE) MR breast images is a well-known difficult problem. It is because the breast motion is non-rigid and the intensity variations of tumor between pre-contrast and post-contrast images can cause the unexpected tumor shrinking/expanding effect in the registration process. To obtain accurate registration, we propose two techniques: a novel image similarity measure based on the robust estimation and a new global optimization technique by reformulating the registration problem as solving a linear programming. The novel similarity measure can help to handle the shrinking/expanding problem while the global optimization technique offers more accurate estimation of the breast motion.