COMPUTER AIDED DETECTION OF RADIO OPAQUE LESIONS IN DIGITIZED MAMMOGRAMS THROUGH DISTRIBUTION FREE KNOWLEDGE BASED CLASSIFICATION TECHNIQUES

摘要

Segmentation of Lesions is a vital step in computerized mass detection scheme for digitized mammograms. In this paper, two robust algorithms have been developed for the segmentation of Radio Opaque Lesions known to be K-Means Bootstrap Subgroup (KMBS) and Expectation Maximization Bootstrap Subgroup (EMBS). The proposed algorithms are capable of segmenting the regions of varying intensity distribution in a mammogram. A number of image regions being 8 and 5, it yields True Positive (TP) rate of 94.5% and 92.3% for EMBS with false positive per Image of 0.26 and 0.33 respectively. When KMBS method is applied for the same data set, it results in TP rate of 93.4% and 91.3% with false positive per image of 0.33 and 0.37. The regions of Radio Opaque Lesions are segmented and the assessments of the segmentation results by radiologist are compared. The efficiency of algorithm is measured using Free Receiver Operating Characteristics (FROC) curve and the results are highlighted.