Limitations in distinguishing malignant from benign lesions of the breast by systematic review of mammograms.

Our objective was to assess whether or not a systematic review of mammographic descriptors would result in improved classifications of nonpalpable lesions of the breast and reductions of biopsies for benign conditions. The prelocalization mammograms of 200 consecutive patients who had 231 biopsies were reviewed in random order by four radiologists experienced in mammography. Each reader assessed the presence of 32 specific descriptors regarding masses and calcifications. The following descriptors were highly associated with malignancy: for masses, inhomogeneous density; irregular shape; spiculated borders; and for calcifications, many calcifications; microcalcifications; linear or branching patterns, or both, and high spatial density. For masses, the descriptors highly associated with benignity were homogeneous density, oval shape and sharp or lobulated borders. For calcifications, the descriptors highly associated with benignity were solid, round and irregularly shaped. Readers indicated their over-all rating of each lesion, ranging from definitely benign to definitely malignant. Ratings were compared with the findings of pathologic reports (181 benign to 50 malignant), and receiver operating characteristic analyses were performed. The areas under the curves (A[z]) varied from 0.65 to 0.78. If biopsies had not been performed on those patients who were judged to have most likely benign lesions, one could reduce the number of biopsies for benign conditions by 48 to 69 per cent; however, 18 to 46 per cent of the malignancies would be missed. We conclude that diagnoses based on a systematic review of descriptors are not sufficiently accurate to reduce the number of mammary biopsies.