Overview of the evidence on digital breast tomosynthesis in breast cancer detection

Digital breast tomosynthesis (DBT, or 3D-mammography), a three-dimensional derivative of digital mammography (DM), reduces the effect of tissue superimposition and may improve mammographic interpretation. In this review, we examined the evidence on the accuracy of DBT in clinical studies. Published studies of DBT were relatively small studies, mostly test-set observer (reader) studies or clinical series that included symptomatic and screen-recalled cases, and were generally enriched with cancers. With these limitations in mind, the evidence showed some consistent findings, summarized as follows: two-view DBT has at least equal or better accuracy than standard two-view DM, whereas one-view DBT does not have better accuracy than standard DM; the addition of DBT to standard mammography (for mammographic interpretation or for assessment or triage of screen-recalled abnormalities) increases accuracy; improved accuracy from using DBT (relative to, or added to, DM) may be due to increased cancer detection or due to reduced false positive recalls, or both; and subjective interpretation of cancer conspicuity consistently found that cancers were equally or more conspicuous on DBT relative to DM. Preliminary data from population screening trials suggest that the integration of DBT with conventional DM (screen-reading using combined 2D + 3D mammography) may substantially improve breast cancer detection, although final results are not yet available, and many logistical issues need further evaluation to determine the potential implications and cost of combined 2D + 3D mammographic screening. At present, there is insufficient evidence to justify a change from standard DM to DBT however the available data strongly support investment in new large-scale population screening trials. These trials need to avoid the 'double' acquisitions required for 2D + 3D mammograms, and should therefore focus on evaluating integrated 2Dsynthetic + 3D mammography (where 2D-images are reconstructed from the DBT acquisition), and should consider using a randomized design.