An integrated detection method of clustered microcalcifications in mammography based on multiscale Hessian Matrix
Ying Wang, Hong Zhao, Hong Li, Xiaoguang Pan, Yan Kang · 2012
As is well known, microcalcification cluster is one of the early and important signs of breast cancer. However, microcalcifications with low contrast can hardly be distinguished by physicians sometimes, and CAD system plays a more and more important role in clinical diagnosis. An integrated detection method for microcalcification clusters in digital mammography was proposed in this paper. It consisted of three parts: the preprocessing and segmentation of region of interest, detection of the candidate regions of microcalcification and reduction of false positives. We used an adaptive threshold in a local window to detect big ringed dot calcification in the image which should be excluded from the candidate region. Then we applied Hessian Matrix in multiscale to detect dot shape. After the individual microcalcification was detected, cluster features were employed to group them into clusters and then false positive regions were removed by feature identification. We used 1,379 sets of clinic data for evaluation the integrated method, in which 547 cases with mirocalcification clusters were included. Results showed that the sensitivity of the method was 92.7%, with the FP 0.19 per image.