Abstract
A new multi-dimensional statistical-based texture segmentation algorithm (STA) is presented in this work. The presented algorithm is based on two and three-dimensional statistical texture analysis. The goal of the algorithm is to accurately detect masses in the breast tissue and to classify regions of the breast into one of three pre-defined types: a mass, glandular tissue or background. The STA is composed of two modules, the 2D-STA and the 3D-STA, which interact with each other for features extraction and classification. This work also introduces the idea of viewing texture as volumetric data. Three-dimensional analysis of textural regions is performed over interpolated volumetric Regions of Interest (ROI's) or temporal planar surfaces. Finally, a method to incorporate knowledge to our segmentation algorithm is presented to create a robust segmentation process. The knowledge used is in the form of rules in a knowledge-base and it ranges from details of the anatomy in question to effects of contrast agents on breast tissue. Results illustrate the capability of the algorithm to accurately segment and classify regions of the breast in two and three dimensions into one of three defined basic textural classifications, i.e., masses, background and glandular tissue.