Segmentation of suspicious densities in digital mammograms' computer-aided detection
Changyuan Wang · Journal of Taishan Medical College · 2005
Objective: To develop and evaluate a more effective segmentation algorithm in computer-aided detection on masses of mammograms. Methods: In this work, a two-stage mass segmentation method was used in pixel-level segmentation and region-level segmentation which was sensitive in detection the suspicious densities (mass) in digital mammograms. Results: The method was used to segment a consecutive set of 62 digital mammograms taken from the Visible Human Data (VHD) and mammogram phantom's image. Evaluation of the performance of the method was done in two different ways. In the first experiment, the segmentations of masses were compared with annotations made by the radiologists. In the second experiment, the ROC curves were calculated to examine the segmentation results. Conclusion: The performance of two-stage mass segmentation can reject false-positive regions, and thus the specificity increases while high sensitivity is maintained, and the suspicious areas can be segmented more closely.