False Positive Reduction in Computer Aided Detection of Mammographic Masses Using Canonical Correlation Analysis

X-ray mammography is the most widely used modality for screening breast cancer in the early stages. Computer aided detection (CADe) systems intend to help radiologists in improving the detection rate. However, the drawback of CADe systems is that they result in a high false positive rate (FPR). In this paper, a new feature-fusion-based system is proposed for classifying automatically detected masses in a mammogram as true masses or false positive cases. In this system, unilateral and bilateral information is fused using a multivariate statistical technique called canonical correlation analysis (CCA). The proposed system is validated using a public database called the mammographic image analysis society (MIAS) database. When compared to unilateral, bilateral and conventional-fusion based systems, the overall classification performance of the proposed system is higher by a range of 8%-16%, 12%-16% and 14%-28% in terms of accuracy, area under curve (AUC) and equal error rate (EER), respectively. Further, the reduction in FPR for the proposed system is at least 39%, 35% and 33% at true positive rates (TPRs) of 60%, 65% and 70%, respectively.