Development of a new case based computer-Aided detection scheme for screening mammography

In this study, we examine a new case based approach to computer-Aided detection (CAD) schemes of screening mammograms based on the computation of many global bilateral asymmetry and mammographic density image based features. The current commercialized CAD schemes have high false positive detection rates, which also have high positive lesion detection correlations with radiologists. Thus, we developed a new global image feature based CAD scheme that can cue the warning sign on the cases with high risk of being positive using an extended feature set of 158 mammographic density and texture based features computed on all four craniocaudal (CC) and mediolateral oblique (MLO) view images. We utilized a modified fast and accurate sequential floating forward selection feature selection algorithm and applied selected features to a 'scoring fusion' artificial neural network (ANN) classification scheme to produce a final case based classification score. We tested our methods using a ten-fold cross-validation scheme on 924 cases (476 cancer and 448 recalled or negative). The area under the receiver operating characteristic curve was AUC = 0.742±0.016. Odds ratios increased from 1 to 15.43 as the CAD-generated case based detection scores increased. The results of the study show that useful information can be derived from the global mammographic density image based features that can be examined further as a new paradigm/approach of CAD for screening mammograms.