Unified data structures in a software environment for medical image segmentation

Medical image segmentation has many applications, including tumor localization, radiation therapy planning, and three-dimensional modeling, but its current use falls far short of its potential. To address this shortcoming, we are developing a unified software environment that facilitates the development and deployment of new and existing medical image segmentation algorithms, including classification-based shape-based, region-based, edge-based, and hybrid algorithms. The foundation of this environment is a set of unified data structures, combining region-based and edge-based representations of segmented images: the region-based part of the representation stores the classification label assigned to each pixel, while the edge-based part stores the boundaries between the different classification regions in a topologically consistent manner. Having such a unified data structure allows us not only to develop pixel-based, region-based, edge-based or shape-based segmentation algorithms, but also to integrate them efficiently. In addition to the data structures, and built on their foundation, our software environment provides other basic tools needed for developing and testing image segmentation algorithms. These include image display tools, manual segmentation tools, segmentation editing tools, basic file input and output tools, and validation tools (to compare computer-generated segmentation to hand segmentation). All these tools are provided in a unique modular, extensible, and customizable form, using object-oriented software design and standard distributed objects (Object Linking and Embedding [OLE] and ActiveX from Microsoft). This architecture also allows our software to be seamlessly integrated with other software environments which use OLE and ActiveX (such as the statistical analysis package S-PLUS and the 3-D visualization package AVS), and it allows this software to be embedded on a home page on the World Wide Web.