Computer-aided detection of clustered microcalcifications

Publisher Website
Google Scholar
Abstract
A computerized technique is being developed to automatically detect clustered microcalcifications on digital mammograms. The method consists of three steps. First the signal-to-noise ratio of microcalcifications is enhanced by filtering the image to reduce the normal background structure of the mammogram. Second, signals (potential microcalcifications) are identified by means of global grey-level thresholding, morphological erosion, and a local adaptive grey-level thresholding. Third, the number of falsely detected signals is reduced by examining the power spectrum of individual signals, determining the spatial distribution of the signals, and examining the relationship between size, shape, and background pixel value of microcalcifications. Using this approach, the computer scheme was tested using 78 mammograms, half containing subtle clusters of microcalcifications and half containing no clusters. The scheme was capable of detecting 87% of true clusters with, on average, two false clusters detected per image.<>