Detecting abnormalities on mammograms by bilateral comparison

An automated system for detecting breast abnormalities should significantly reduce the time needed to examine a mammogram. A system should be able to detect most kinds of abnormalities in order for it to have a positive contribution to a clinical situation. One way of examining mammograms that is used frequently by radiologists is the comparison of left and right breast images. Recent attempts to automate the comparison method have produced very promising results. Two new attempts are presented here. Initially, segmentation of the digitised images involves separating breast tissue from their background. Alignment of the 2 mammograms is then carried out using a single reference-the point of maximum curvature on the breast curve. Finally normalisation is used to minimise differences in illumination between X-ray images before comparison. The first method, single image comparison, involves finding corresponding areas whose intensities differ more than a preset threshold. The results are presented in the form of 2 binary images which are median filtered to eliminate artifacts and to smooth rough borders. The second method, multiple image comparison (MIG), involves generating 8 pairs of images for each original pair of left and right images. MIC uses a combination of processes including adaptive histogram modification, normalisation, grey level thresholding, binary image cleaning and region segmentation. The 8 pairs of images are then bilaterally compared and the resulting images recombined into 1 pair of images.