Mammography is recognized as an important means to reduce breast cancer mortality. However, its accuracy is limited, both in sensitivity (some cancers are missed) and specificity (many non-cancer cases are referred for invasive procedures). This proposal aims at improving the effectiveness of breast cancer screening by the discovery of measurements that can be taken from digital mammograms, and the design of classifiers that result in an automatic computer-generated description of suspicious areas. In particular, classifiers for the standardized lexicon for mass shape, mass margin and mass density, as well as breast composition will be designed. These automaticaily generated descriptions are aimed at increasing the specificity of mammography by providing the radiologist with a probability of rnalignancy for the lesion. The descriptions should also be helpful in conjunction with computer programs that detect suspicious areas, by rejecting those detected areas that do not likely represent cancer (false positive reduction). In addition, automatic description of a marnmographic iesion will reduce reader variability and may heip in training radiologists to use the standardized lexicon. This project will first develop a segmentation program, that finds the border of a mammographic mass. The approach will be a multi-stage knowledge guided system. Measurements to be taken from digital mammograms make use of this border, and include some that have been reported in the literature as well as newly proposed measurements. As the classification problem for mammographic masses is very difficult, hybrid classifiers will be constructed that take advantage of specific abilities of multiple classifiers. A rule based system will be developed to arrive at an aggregate decision for each of the BIRADS descriptions. Finally, these descriptions will be used in a classifier to determine the likelihood of malignancy. The outcome of this project would be the tools to develop an aid to non-expert mammographers to define the characteristics of mammographic masses and help in decisions to biopsy a mammographic lesion, but can also be applied to help enhance sensitivity through computer assisted diagnosis, and in educational tools.
