Database-Aided Suppression of False Positives of Mammographic Calcifications

The widespread use of screening mammography has significantly altered the composition of breast cancer being diagnosed today. Cancers are being diagnosed earlier and the mortality rate from breast cancer is dropping, mostly attributable to the use of screening mammography [1]. At breast care centers, approximately 25% of new breast cancers are now being diagnosed when still in-situ (i.e. confined to the ducts). These are nearly 100% curable, generally with breast conserving surgery. The mammographic finding most commonly associated with in-situ cancer is mammographic calcification. Unfortunately, it is often difficult to differentiate malignant type calcification from those associated with a myriad of benign processes. As a result calcification lesions have a greater false-positive rate than mass lesions. Among biopsies of calcification lesions, over 80 percent are benign. In the U.S., the vast majority of mammograms are interpreted by general radiologists whose experience with mammographic calcifications is limited. As “delay in diagnosis of breast cancer” is the most common source of medical-legal lawsuits against diagnostic radiologists, the tendency is to practice defensively. We describe a computer aid that offers to reduce the false positive rate of calcification lesions and thereby to reduce the incidence of defensive reading of mammograms. The objective of this study is to evaluate the ability of a patented two-dimensional mapping of calcification lesions to help the radiologist-interpreter reduce the incidence of false positives in screening apparent calcifications for cancer. We refer to this mapping as a relational map. We introduced the relational map to the radiology community in June 2000 [2]. This paper reports on a recent test of the ability of the relational map to help reduce the numbers of falsely diagnosed suspicious cancers and unnecessary recalls associated with the analysis of calcification regions of interest (ROIs).