Statistical models for the detection of abnormalities in digital mammography

This paper introduces statistical methods for the detection of abnormalities in X-ray mammograms. Two types of abnormalities which may be encountered are microcalcification clusters and masses, both of which are possible indicators of breast cancer. Microcalcifications are small deposits of calcium in the breast, which are associated with a high incidence of breast cancer, whilst masses may indicate cancerous growth. While masses are typically of a size which will enable detection under current breast screening procedures, the small size of microcalcifications indicates that a computer assisted analysis is appropriate to enable detection at an early stage. The analysis presented here is evaluated using two sets of data. The first of these is a CIRS phantom image containing clusters of microcalcifications in a range of sizes, and the second is a set of digitized film mammograms depicting both masses and microcalcifications. The authors are indebted to Dr Matthew Freedman of the Georgetown University Medical Centre, Washington D.C. For both of these sets of data. Two approaches have been adopted for the analysis of this data. The first approach is based upon a method that was developed by the authors for the application of detecting objects in sidescan sonar images. Here, the authors add a pre-processing algorithm which suppresses the background variability whilst emphasising the abnormalities. A second method, developed specifically for this application, is based upon a Gibbs random field which is designed to model pixel interactions within the image