Clustered Microcalcification Detection in Digital Mammography for Various Breast Densities

Abstract

Background: Early detection of breast cancer is critical to improving breast cancer prognosis and reducing mortality rates. Microcalcification clusters viewable through digital mammography are a major sign of the early stages of breast cancer. However, detecting microcalcification clusters is difficult because they are small and have low contrast compared to normal breast tissue, especially in a dense breast. Methods: In this study, we proposed an automatic detection method for microcalcification clusters in digital mammography and evaluated the algorithm for various breast densities. First, we extracted the breast region, and then removed high-intensity artifacts such as labels or scanning artifacts. Second, we enhanced the contrast to emphasize small microcalcifications in dense breast regions. Because the candidates for microcalcification are observed as small bright blobs, we detected them by combining the Laplacian of Gaussian method and Foveal algorithm. Among the candidates, their false positives were then reduced by using knowledge-based rules. Results: The proposed method for detecting microcalcification clusters based on 232 abnormal mammograms and 216 normal mammograms achieved a sensitivity of 97.43% at 0.27 false positives per image. Conclusion: The proposed algorithm showed a reliable accuracy in mammograms with various densities. Therefore, this method can be useful in detecting microcalcification clusters for early diagnosis of breast cancer.