Now that digital breast tomosynthesis (DBT) has gained FDA approval, many breast-imaging providers find themselves excited about the new technology, but facing uncertainty about reimbursement, implementation, and interpretation workflow. There remain a number of questions related to the display of (and approach to interpreting) DBT that need to be addressed if DBT is to be incorporated optimally into routine practice. As they were for digital mammography (DM) in the recent past, the acquisition and dissemination of DBT technology will be marked by early and late adopters. Early adopters will be willing to work with uncertainty and develop the nuances of implementation and interpretation in their own unique environments; later adopters will pattern their practices on the reported experience of others. Clearly, those who determine how to interpret DBT exams best (and most efficiently) will influence its acceptance and clinical implementation. The mechanism of DBT image acquisition has been well described.¹,² Though standards from the industry and the Mammography Quality Standards Act (MQSA) do not yet exist for all aspects of this technology, practitioners are rapidly becoming knowledgeable and gaining experience in the nuances of best displaying and reviewing 3D DBT image datasets. In general, DBT is displayed as 1-mm image slices that are reviewed using cine or manual scroll modes on dedicated soft-copy workstations. The first DBT image slice begins at the detector, for craniocaudal (CC) and mediolateral oblique (MLO) DBT projections. At the compression-plate aspect of the image set, five additional slices are included to ensure that all tissue is displayed. Thus, one can estimate breast-compression thickness from the number of 1-mm BDT slices in the dataset by subtracting 5 mm from the total number of slices. Having location information readily available from DBT image slices makes evaluating masses, calcifications, or other findings related to the skin very straightforward on any routine screening exam, without the necessity for additional views to clarify that a mammographic finding is dermal in origin. Currently, there are a few artifacts related to image reconstruction that radiologists need to understand and adapt to when interpreting DBT studies. Coarse calcifications (macrocalcifications), metallic (BB or scar) markers, postbiopsy clips, and other high-density objects within or applied to the breast create a coil-spring or slinky artifact (Figure 1). While these are initially distracting, many radiologists find that they are quickly able to read through these artifacts. In the future, advanced reconstruction software might reduce or eliminate these artifacts, and many DBT users agree that the benefits of the technology outweigh its current limitations. Another concept to recognize is that, unlike 3D datasets acquired by breast-MRI volume acquisition techniques, the CC-DBT and MLO-DBT datasets cannot be cross-referenced with display software as, for example, a sagittal and axial breast-MRI imaging sequence. DBT CC and MLO projection data are acquired with nonisotropic voxels and with two separate breast positionings, rendering impossible the cross-referencing easily performed with breast MRI 3D datasets. Currently, both a 2D standard DM image and a 3D DBT dataset are acquired with a single positioning and a single exposure for each of the standard screening views. This is commonly referred to as 2D/3D combo imaging and is performed with a total dose that is less than the maximum permitted by MQSA requirements. Many radiologists learning to interpret DBT find that having the standard 2D to correlate with the 3D DBT images is of value in gaining confidence and reducing the learning curve. In addition, studies³,⁴ have shown, based on screening ACR BI-RADS® ratings, that the combined use of DM and DBT was superior to DM alone in terms of significantly reducing recall rates for further diagnostic work-ups. In the future, one of two scenarios might emerge. In the first, after a period of time when patients have undergone more than one DBT study, radiologists might be comfortable viewing only DBT images. In the second, the standard 2D DM images will be replaced by synthesized 2D images, created from the DBT dataset—which would reduce radiation dose. A Dynamic Study In general, the interpretation of a 2D/3D combo screening study begins with a review of the 2D images. After reviewing the four standard 2D DM images and comparing them with any prior studies, the radiologist reviews the individual DBT image sets in conjunction with the corresponding 2D DM projection (using two monitors). This permits careful correlation of 2D image findings with the DBT image slices. There will, undoubtedly, be differing opinions on how best to view a 3D DBT image set, but the overarching goal is to evaluate all images thoroughly and completely. DBT is a dynamic study; as one scrolls through the DBT slices, both images and the reviewer’s eyes are moving. For radiologists who do not spend a majority of their time viewing 3D image datasets, it is important to develop techniques to ensure viewing all aspects of the image data. Initially, it is useful to approach the DBT study as if the 2D DM exam were negative.
Figure 1.Cropped craniocaudal (left) and mediolateral oblique digital breast tomosynthesis image slices demonstrate the typical coiled-spring or slinky reconstruction artifact created by breast calcifications. Choose specific subsections (retroareolar, medial, lateral, superior, or inferior) of the image dataset on which to focus for each CC or MLO DBT, and manually scroll forward and backward through the image stack until you feel confident that you have looked at all areas of the exam. Then, refocus on any specific findings noted on the 2D DM and search for a similar finding on the DBT. It is especially useful to have the 2D exam for detecting and rendering initial evaluations of microcalcifications. After viewing each DBT projection alongside the corresponding 2D DM image, it will be necessary (in the future) to compare last year’s DBT with the current DBT. Here, it will be quite helpful to have display software that intelligently considers compression-thickness differences, that links the prior and current DBT image datasets, and that enables the radiologist to scroll fluidly through both datasets simultaneously to review and analyze the images. When the 2D DM and 3D DBT images are reviewed side by side, the DBT often resolves 2D DM findings that would result in recalling the patient if DBT were not immediately available. A common example is a summation artifact or pseudomass. Scrolling through the DBT image stack and focusing on the location where the finding was noted on the 2D exam will show normal tissue or a series of normal anatomic structures (Cooper’s ligaments, vessels, and parenchyma) that sum to create the spurious 2D finding. If a 2D finding is a real mass, it generally appears more conspicuous on the DBT because the underlying and overlying structures are eliminated on DBT image slices. Value Proposition and Pitfalls The value of resolving or reducing summation artifacts using DBT is a reduction in recall rate, with improved screening specificity. The reduction of summation of tissue on DBT means, however, that there is the potential to see benign masses not visible previously in the breast parenchyma on prior DM exams. This phenomenon is similar to the early DM experience, when microcalcifications not previously depicted on screen-film mammography were seen on DM, resulting in a transient increase in the recall rate for magnification views. In evaluating DBT-only (probably benign) masses, there is a need to balance the ordering of additional mammographic views or ultrasound with creating more follow-up DBT cases and/or increasing recall or work-up rates. Detecting microcalcifications on DBT can be challenging because individual calcifications in a cluster might be located on different DBT slices, creating a perceptual challenge. The radiologist might not appreciate the clustering of the calcifications because, as he or she scrolls through the DBT images, one calcification might appear and then disappear, while others become apparent in the next slice, but are not present on the subsequent slice. The calcifications might be quite conspicuous, but the radiologist doesn’t appreciate the clustered distribution. Having the 2D mammogram performed with the 2D/3D combo imaging set enhances detection. In addition, several 1-mm DBT image slices can be grouped into a slab of any thickness to promote better appreciation of the size and extent of a calcification cluster. Although increasing slice thickness will increase the ability to perceive the 3D configuration of a cluster of calcifications, the spatial resolution of each individual calcification is compromised by slabbing. While DBT often negates the need for additional mammography views to evaluate summation artifacts, masses, and architectural distortions, it generally does not negate the need for micro-focus magnification views to evaluate the morphology of microcalcifications. In the future, DBT will be augmented by computer-aided detection to identify high-frequency image information present on DBT image slices indicating microcalcifications; it will aid in the perception, classification, and characterization of calcifications on DBT. Computer-aided detection software for DBT is currently being evaluated and tested in Europe. DBT enhances the detection and conspicuity of breast masses and architectural distortions (Figures 2 and 3). In many cases, DBT images alone are sufficient to make a BI-RADS assessment of these findings, without additional spot-compression mammography. In reviewing masses and distortions on DBT, radiologists should be aware of certain precautions to be taken to prevent undercalls and overcalls. It is important to evaluate mass borders based on information gleaned from all DBT slices encompassing the mass. Because overlying parenchyma is eliminated in DBT image slices, a mass with a border that is not well visualized on the 2D DM might be well visualized on DBT; however, one should avoid assessing mass margins on the basis of a few DBT slices. It is important to assess mass shape and borders based on all DBT slices containing the mass—to avoid designating the mass as circumscribed, while overlooking spiculations or lobulations. If there are lobulated borders, consider a BI-RADS 4a designation (so that a circumscribed-but-lobulated malignant mass is not undercalled). Similarly, architectural distortions often appear more prominent and easily detected on DBT. When one is first reading DBT images, however, normal structures (such as Cooper’s ligaments) might simulate a distortion or spiculation on a single DBT slice. Therefore, it is advisable not to dwell on a 1-mm slice that appears to have a distortion. Rather, evaluate the finding in conjunction with several slices above and below it before determining that additional work-up is indicated.
Figure 2.Cropped 2D digital mammography image (left) and 3D digital breast tomosynthesis image slice of a 9-mm spiculated breast carcinoma; on the left, the mass is obscured by overlying breast tissue, but it is depicted well on the right. Experts recommend scrolling at a uniform rate through at least 10 DBT slices (to evaluate a potential area of architectural distortion fully) before designating the finding as actionable and making it a call-back case. With practice, one can avoid overcalling distortions. In your initial experience with DBT, it is useful to pay close attention to the borderline findings that you recall for additional work-up so that you can knowledgeably adjust your interpretation threshold over time. Another pitfall to be aware of is that malignant masses sometimes demonstrate areas of lucency of fat within them on DBT. This should not dissuade one from calling for additional work-up if the mass is lobulated or irregular. Only encapsulated fat-containing masses, indicative of lipoma or hamartoma, should be considered benign (BI-RADS 2).
Figure 3.Cropped 2D digital mammography image (left) and 3D digital breast tomosynthesis image slice of a 7-mm spiculated breast carcinoma; on the left, the mass is obscured by overlying breast tissue, but it is depicted well on the right. As with breast MRI, there will be a subgroup of suspicious masses or architectural distortions identified on DBT that cannot be found on diagnostic mammography, ultrasound, or even MRI. For these lesions, a DBT-guided needle localization can be performed in a manner similar to that used for conventional DM localization. After acquiring a 2D/3D combo image (and without releasing compression), scroll through the 1-mm image stack to locate the lesion and then insert the localization needle. Next, obtain an orthogonal 2D/3D combo image, which will have a needle artifact on each slice. In the slice best depicting the lesion, the relationship of the needle tip to the lesion will be seen, and the needle depth can be adjusted accordingly. A single DBT slice from each projection can be printed for the operating surgeon. Although there are different ways to implement and use DBT in a breast-imaging practice, DBT is likely to be most valuable for screening mammography, where its ability to depict small cancers will enhance screening sensitivity, while the potential to reduce recalls will improve specificity. The addition of DBT images to a screening or diagnostic evaluation does increase interpretation time and, like any new technology, there is a learning curve—but it is a relatively rapid one. Because DBT and DM images are acquired similar ways, radiologists are immediately familiar with structures and parenchymal patterns. The primary difference is that the structures and patterns are visualized with greater clarity on DBT because superimposed or obscuring structures are eliminated. Thus, with training and experience, radiologists quickly gain comfort and efficiency with this new breast-cancer–detection technology. Laurie L. Fajardo, MD, MBA, is chair of the department of radiology at the University of Iowa Hospitals and Clinics in Iowa City. Limin Yang, MD, PhD, is clinical assistant professor at the University of Iowa Carver College of Medicine. Jeong Mi Park, MD, is clinical professor at the college.