Simultaneous Multi-object Segmentation Using Local Robust Statistics and Contour Interaction

Abstract

In this work, we present an active contour scheme to simultaneously extract multiple targets from MR and CT medical imagery. A number of previous active contour methods are capable of only extracting one object at a time. Therefore, when multiple objects are required, the segmentation process must be performed sequentially. Not only may this be tedious work, but moreover the relationship between the given objects is not addressed in a uniform framework, making the method prone to leakage and overlap among the individual segmentation results. On the other hand, many of the algorithms providing the capability to perform simultaneous multiple object segmentation, tacitly or explicitly assume that the union of the multiple regions equals the whole image domain. However, this is often invalid for many medical imaging tasks. In the present work, we give a straightforward methodology to alleviate these drawbacks as follows. First, local robust statistics are used to describe the object features, which are learned adaptively from user provided seeds. Second, several active contours evolve simultaneously with their interactions being governed by simple principles derived from mechanics. This not only guarantees mutual exclusiveness among the contours, but also no longer relies upon the assumption that the multiple objects fill the whole image domain. In doing so, the contours interact and converge to equilibrium at the desired positions of the given objects. The method naturally handles the issues of leakage and overlapping. Both qualitative and quantitative results are shown to highlight the algorithm's capability of extracting several targets as well as robustly preventing the leakage.