A Universal Segmentation Platform for Computer-Aided Detection

One of the fundamental image processing challenges that a mammographic computed-aided detection system faces is to reliably segment a wide variety of objects from complicated backgrounds regardless of the input source. Rather than develop a completely new approach for each segmentation task and each source of data, we have found that it is possible to use a single active- contour segmentation engine for each problem. The segmentation engine does most of the work for a given application and can easily be interfaced with prob- lem-specific algorithms that define the engine’s input and adapt its operation to the problem at hand. In this paper, we describe the segmentation engine and discuss how we apply it to the segmentation of densities, pectoral muscles, and computed tomography images of lungs.