Margarita Louise Zuley MD, Abstract Co-Author: Research Grant, Hologic, Inc

Julie Jean Koo MD, Presenter: Nothing to Disclose

Donna M. Plecha MD, Abstract Co-Author: Advisory Board, Hologic, Inc Research Grant, SuperSonic Imagine

Stephen L. Rose MD, Abstract Co-Author: Consultant, Hologic, Inc

Joseph Charles Benjamin MD, Abstract Co-Author: Nothing to Disclose

David Gur PhD, Abstract Co-Author: Nothing to Disclose

Andriy I. Bandos PhD, Abstract Co-Author: Nothing to Disclose

Jules Henry Sumkin DO, Abstract Co-Author: Scientific Advisory Board, Hologic, Inc

Amy Elizabeth Kelly MD, Abstract Co-Author: Nothing to Disclose

Marie Adele Ganott MD, Abstract Co-Author: Nothing to Disclose

To determine the imaging and breast tissue characteristics associated with false negative tomosynthesis studies.

IRB approval was obtained. 339 tomosynthesis examinations with verified cancer performed at 3 institutions either at the time of diagnosis or up to 12 months prior to the cancer diagnosis were retrospectively reviewed. Tissue density, tomosynthesis views obtained and pathology cell type were collected. One of 6 experienced breast imaging radiologists recorded lesion location on each view, shortest distance to skin and reason for non-visualization (obscured, not included on the view, looks like normal tissue, motion/blur). Lesion location and cancer cell type were known to the reader to assure the correct lesion was evaluated.

Cell type distribution was 24% IDC (82/339), 16% DCIS (53/339), 48% (164/339) mixed IDC/DCIS, 10% (33/339) ILC, and <1% each (3/339) mucinous and (1/339) invasive papillary. Of the 339 cancers detected within 12 months of tomosynthesis acquisition, 54 (16%) were not visible on both views. A substantially larger number of cancers were visible on the CC view (78%; 265/339) as compared to the MLO view (47%;165/339) (p<0.001). The majority of the non-visible cancers were recorded by the readers as not visible because the cancer looked like normal tissue. There was no significant difference in visibility as a function of breast density (p=0.13) but fractionally, visibility was better in lower breast densities. There was no significant difference in visibility by cancer type (p>0.6). On the CC view, better visibility was associated with being located centrally (p<0.04) while there was no significant correlation between fraction of cancers rated not visible with respect to location or distance to skin on the MLO view.

Cancers of all cell types are missed on tomosynthesis on one or both views at all locations, primarily because they look like normal tissue. The CC view seems to depict substantially more cancers than the MLO view.

Tomosynthesis has been shown to improve performance but still a substantial fraction of cancers of all types and at all locations may be missed primarily in non-fatty breasts because they look like normal tissue.

Zuley, M, Koo, J, Plecha, D, Rose, S, Benjamin, J, Gur, D, Bandos, A, Sumkin, J, Kelly, A, Ganott, M, Analysis of Cancers Missed on Digital Breast Tomosynthesis.  Radiological Society of North America 2014 Scientific Assembly and Annual Meeting, - ,Chicago IL. http://archive.rsna.org/2014/14006594.html