Development of Detection Filter Based on Density Gradient Analysis for Mammographic Microcalcifications

Abstract

We have developed a new automated-detection algorithm for clustered microcalcifications on digital mammograms. In our technique, the vectors of density gradient were firstly calculated within the area of breast which was segmented automatically. Second, three "circular-shape" filters were developed to extract the specific features for microcalcifications pattern from the vectors. The sen-sitivity and specificity of our algorithm were 81% and 63% with 0.53 false detection per image for our database of 100 mammograms.