Abstract
We have improved our automated detection algorithm for clustered microcalcifications on digital mammograms by adding two techniques. In our schemes, vectors of density gradient are firstly calculated within the area of breast which is segmented automatically. Second, the "triple-ring filter"extracts the specific features for the pattern of microcalcification from the vectors. Third, the shape analysis is performed. As one of new parts, a region growing technique, in which the information of vector intensity is employed to calculate the variable thresholding values, is used to determine the correct region of the microcalcification. The contrast analysis part is also added in order to reduce the false-positive candidates. The sensitivity of our revised algorithm is 87.3% with 0.55false detection of cluster per image in our database of 163 mammograms.
