A bilateral analysis scheme for false positive reduction in mammogram mass detection. (1st February 2015)

• Record Type: Journal Article
• Authors: Li, Yanfeng; Chen, Houjin; Yang, Yongyi; Cheng, Lin; Cao, Lin
• Is Part Of: Computers in biology and medicine. Volume 57(2015)
• Page Start: 84
• Page End: 95
• Publication Date: 2015-02-01
• Subjects: Bilateral analysis -- Mammogram -- Mass detection -- Similarity measurement -- Region matching; Medicine -- Data processing -- Periodicals; Biology -- Data processing -- Periodicals; 610.285
• DOI: 10.1016/j.compbiomed.2014.12.007
• Languages: English
• ISSNs: 0010-4825

Abstract

Abstract: In this paper, a bilateral image analysis scheme is developed for the purpose of reducing false positives (FPs) in the detection of masses in dense mammograms. It consists of two steps: a region matching step for determining the correspondence between a pair of mammograms, and a bilateral similarity analysis step for discarding FPs in the detection. For the first step, a matching cost is defined to quantify the credibility of the corresponding region in a pair of bilateral mammograms. For the second step, a similarity measurement is introduced to discriminate between mass and normal for a pair of bilateral regions based on both global and local image appearances. The proposed scheme is tested on a set of 332 mammograms. The results show that the proposed scheme could obtain better performance when compared with several existing bilateral analysis schemes. With detection sensitivity at 85%, the proposed bilateral scheme could reduce the FP rate of a unilateral scheme from 3.64 to 2.39 per image, a 34% reduction. Highlights: The proposed region matching method using shape context is robust to segmentation error. Matching cost is defined to quantify the matching credibility, which is discarded by existing bilateral analysis schemes. A hierarchical similarity is designed for FP reduction in bilateral mass detection. Promising FP reduction is obtained using bilateral analysis in mass detection.