A modified Neyman-Pearson technique for radiodense tissue estimation in digitized mammograms

Abstract

The percentage of radiodense tissue in the breast has been shown to be a reliable marker for breast cancer risk. In this paper, we present an image processing technique for estimating radiodense tissue in digitized mammograms. First, the mammogram is segmented into tissue and nontissue regions. This segmentation process involves the generation of a segmentation mask that is developed using a radial basis function neural network. Subsequently, the image is processed for estimating the amount of radiodense tissue. The estimation process involves the generation of a modified Neyman-Pearson threshold to segment the radiodense and radiolucent tissue. Typical research results are presented-these have been independently validated by a radiologist.