Translator Disclaimer
Paper
9 May 2002 Improving the automated classification of clustered calcifications on mammograms through the improved detection of individual calcifications
Author Affiliations +
Abstract
We are developing a semi-automated classification scheme in which the approximate size and location of the cluster (but not the individual calcifications), and a rough estimate of the number of calcifications in the cluster are used to segment individual calcifications. A difference of Gaussians is used to pre-process the ROI centered on the cluster. Next a global and a local grey-level thresholds are applied. The threshold values are determined iteratively based on the approximate number of calcifications in the cluster and the actual number segmented. The center position of each segmented calcification is then determined. These locations are passed to the computer classifier, which determines the likelihood of malignancy for the cluster. Using this approach, 74% of individual calcifications can be detected per cluster compared to 55% when using our cluster detection scheme that does not use the a priori information about cluster. There was no measurable decrease in the classification scheme's performance when using the segmented calcifications from our new approach compared to if all the location of the calcifications were determined manually (area under the ROC curve of 0.85 versus 0.91 respectively, p = 0.2).
© (2002) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.
Robert M. Nishikawa, Maria Fernanda Salfity, Yulei Jiang, and John Papaioannou "Improving the automated classification of clustered calcifications on mammograms through the improved detection of individual calcifications", Proc. SPIE 4684, Medical Imaging 2002: Image Processing, (9 May 2002); https://doi.org/10.1117/12.467097
PROCEEDINGS
7 PAGES


SHARE
Advertisement
Advertisement
Back to Top