COMPUTER-AIDED DETECTION OF CLUSTERED MICROCALCIFICATIONS - AN IMPROVED METHOD FOR GROUPING DETECTED SIGNALS

A computerized scheme for the automated detection of clustered microcalcifications from digital mammograms is being developed. This scheme is one part of an overall package for computer-aided diagnosis (CAD), the purpose of which is to assist radiologists in detecting and diagnosing breast cancer. One important step in the computer detection scheme is to group or cluster microcalcifications, since clustered microcalcifications are more clinically significant than are isolated microcalcifications. Previously a ''growing'' technique in which signals (possible microcalcifications) were clustered by grouping those that were within some predefined distance from the center of the growing cluster was used. In this paper, a new technique for grouping signals, which consists of two steps, is introduced. First, signals that may be several pixels in area are reduced to single pixels by means of a recursive transformation. Second, the number of signals (nonzero pixels) within a small region, typically 3.2 X 3.2 mm, are counted. Only if three or more signals are present within such a region are they preserved in the output image. In this way, isolated signals are eliminated. Furthermore, this method can eliminate falsely detected clusters, which were identified by a previous detection scheme, based on the spatial distribution of signals within the cluster. The differences in performance of the CAD scheme for detecting clustered microcalcifications using the old and new clustering techniques was measured using 78 mammograms, containing 41 clusters. The new clustering technique improved the detection scheme by reducing the false-positive detection rate from 4.2 to 2.5 per image, while maintaining a sensitivity of approximately 85%.