Women s Imaging Original Research Hakim et al. Digital Breast Tomosynthesis Women s Imaging Original Research Christiane M. Hakim 1 Denise M. Chough 1 Marie A. Ganott 1 Jules H. Sumkin 1 Margarita L. Zuley 1 David Gur 2 Hakim CM, Chough DM, Ganott MA, Sumkin JH, Zuley ML, Gur D 1 Department of Radiology, Magee-Womens Hospital, University of Pittsburgh, Pittsburgh, PA. 2 Department of Radiology, University of Pittsburgh, Radiology Imaging Research, 3362 Fifth Ave., Pittsburgh, PA 15213. Address correspondence to D. Gur (gurd@upmc.edu). WEB This is a Web exclusive article. AJR 2010; 195:172 176 WOMEN S IMAGING Keywords: breast cancer, diagnostic workup, digital breast tomosynthesis, digital mammography DOI:10.2214/AJR.09.3244 Received June 29, 2009; accepted after revision January 15, 2010. C. M. Hakim, D. M. Chough, and M. A. Ganott have participated as readers in independent studies for Hologic, Inc. J. H. Sumkin and M. L. Zuley receive support from Hologic, Inc., to perform nonrebated applied research projects on digital breast tomosynthesis. This work was supported by grant BCTR0600733 (to the University of Pittsburgh) from the Susan G. Komen foundation and also received support from Hologic, Inc. 0361 803X/10/1952 172 American Roentgen Ray Society Digital Breast Tomosynthesis in the Diagnostic Environment: A Subjective Side-by-Side Review OBJECTIVE. The purpose of our study was to subjectively compare additional mammographic views to digital breast tomosynthesis (DBT) in the characterizing of known masses, architectural distortions, or asymmetries. MATERIALS AND METHODS. Four experienced radiologists serially reviewed the imaging studies of 25 women with known masses, including full-field digital mammography (FFDM), additional views, and DBT. After review of the examinations, radiologists rated their relative preference in terms of classifying the finding in question when aided by the additional views versus aided by DBT, their combined diagnostic BI-RADS rating of the finding when both examinations were available, and whether or not they felt comfortable eliminating ultrasound in the specific cases being evaluated as a result of the DBT. RESULTS. FFDM and DBT (combined) were perceived to be better for diagnosis in 50% (50/100) of the ratings (25 cases four readers = 100 ratings) compared with FFDM and additional diagnostic views. Over all readers, 92% of the ratings for verified cancer cases and 50% of the ratings for high-risk cases were rated as BI-RADS 4 or 5. In 12% (12/100) of the ratings, radiologists indicated that the availability of DBT would have eliminated the need for ultrasound as a part of the diagnostic process. CONCLUSION. DBT may be an alternative to obtaining additional mammographic views in most but not all cases of patients with a lesion that is not solely calcifications. In a fraction of cases, the use of DBT may eliminate the need for ultrasound. B reast cancer remains one of the leading causes of death in women over the age of 40 years [1, 2]. The use of full-field digital mammography (FFDM) has been shown to result in significant improvements in radiologists performance levels, in particular when interpreting examinations obtained on younger women with dense breast tissue [3]. However, conventional mammographic images are 2D radiographic views, and, as a result, the presence of overlapping dense fibroglandular breast tissue substantially reduces the conspicuity of some breast lesions. This limits the sensitivity of mammography and constitutes one important reason for missing breast cancers and for recalling a substantial fraction of women without breast cancer for a diagnostic workup [4]. Full-field digital detectors with high resolution and large sensor size [5] offer not only better contrast sensitivity but also great opportunities to develop advanced digital imaging techniques that could improve conspicuity of breast lesions by enabling bet- ter tissue visualization through the provision of 3D nonoverlapped tissue information. As a result, new imaging techniques, such as digital breast tomosynthesis (DBT) [6] and conebeam breast CT [7], are being investigated. A number of investigators believe that DBT has potential in both screening and diagnostic settings [8 14]. In the diagnostic environment, one of the objectives is to improve the efficiency of the workup. However, the more important objective is to improve the selection of patients recommended for biopsy. This would improve the positive predictive value for biopsy and reduce the number of unnecessary biopsies with all associated costs and patient-related effects, such as time spent and anxiety [8]. Anecdotal information on possible improvements in detection, as well as possible reduction in recall rates, is beginning to surface [9 11, 13, 14]. For a variety of reasons, the current primary intended use of this technology, as specifically pursued by many investigators and most commercial companies involved, is in screening. W172 AJR:195, August 2010
Digital Breast Tomosynthesis In reality, the primary use of DBT may initially be in diagnosis rather than in screening. When an abnormality is not solely a microcalcification cluster, the use of DBT could alter the diagnostic protocol substantially. Because a large fraction of biopsies are performed for benign entities and the ability to assess lesion margins when using DBT is frequently exquisite, it is possible that DBT may be helpful in reducing the number of benign lesions that undergo biopsy. For example, we suspect that the correct benign classification of lymph nodes, fibroadenomas, and cysts may improve substantially with DBT. Therefore, to begin to assess the possible future clinical utility of DBT in the diagnostic evaluation of asymmetries, masses, and architectural distortions, we assembled a group of such cases that had FFDM, DBT, and additional diagnostic views. We subjectively assessed in a side-by-side review whether or not simultaneous viewing of FFDM and DBT is perceived to be worse, comparable, or better for diagnosis than FFDM and the additional diagnostic views that actually had been acquired during the diagnostic workup. Materials and Methods Patients and Study Protocol Acquisition of all DBT examinations used in this study was performed under an institutional review board (IRB)-approved protocol for other research purposes that included a signed informed consent by the participant. Eligible subjects who participated in our IRB-approved protocol for acquiring DBT examinations included, among others, a group of women 18 years or older identified by conventional FFDM screening mammography as having a BI-RADS classification 0 for a suspected lesion requiring a diagnostic workup or patients who during a diagnostic evaluation were found to have an abnormality that warranted biopsy. The women were informed by their technologist of a research study taking place at the facility. If the potential subject was interested in hearing more about the study then she was asked if a research coordinator, or coinvestigator, could speak with her about the study. The research coordinator then explained the DBT device to the potential subject and reviewed with her the consent form. For consenting participants, the DBT examinations (craniocaudal and mediolateral oblique projections) were acquired on an experimental DBT system (Gemini Tomosynthesis Digital Mammography System, Hologic, Inc.) at our institution, typically on the day of their scheduled visit. Details of the acquisition protocol are described elsewhere [14, 15]. In brief, during acquisition, the breast is compressed in a conventional manner, and the x-ray tube moves along a limited arc allowing 15 low-dose images to be acquired rather than the single image acquired during a conventional FFDM examination. DBT acquisitions were performed at a combined radiation dose comparable to a FFDM examination, with an average midbreast dose of approximately 2 mgy per view. After acquisition, the data from the projection images were used to reconstruct between 50 and 90 parallel 1-mm-thick slices (i.e., the 3D DBT data set), depending on the thickness of the compressed breast. Cases were selected if they included a mass, an area of asymmetry, or architectural distortion that was identified and either later biopsied (n = 22) or confirmed at ultrasound as an anechoic cyst (n = 3). Through review of our research database from May 13, 2007, to March 17, 2009, 25 subjects were enrolled in our IRB-approved protocol who had undergone FFDM; additional mammographic views; DBT; and biopsy for a mass, architectural distortion, or asymmetry. Cases were selected based solely on the diagnostic and pathology reports, without direct viewing of the imaging examinations relevant to this study. Of these 25 cases, one case was described as architectural distortion, four cases were described as asymmetry, three cases as multiple cysts, and the remaining 17 cases were identified as masses. Cases depicting only microcalcifications were not selected for this study. The selection of cases was performed by one radiologist who did not participate in the subjective rating study and knew from the relevant reports the actual diagnosis of all cases, in particular, the index lesion of interest. The index lesion of interest location was recorded for each case by this radiologist on a data form so that readers knew which lesion to evaluate. Lesion size was not reported for the four asymmetry cases and for one case with multiple cysts. For the remaining 20 cases, the average size of the reported index lesion was 1.98 ± 0.59 cm. Twenty-two cases had ultimately been recommended for biopsy. Of these, one case of asymmetry had an unsuccessful biopsy attempt and, as a result, MRI was performed. MRI in this case revealed that the asymmetry was due to benign fibroglandular tissue. Of the other 21 biopsied cases, 13 were benign, two were highrisk lesions, and six were cancers. Of the 14 benign cases, two (14%) had been rated in the clinic (during the diagnostic workup) as BI-RADS 3, 11 (79%) had been rated BI-RADS 4, and 1 (7%) had been rated BI-RADS 5. One of the two high-risk lesions was an atypical ductal hyperplasia that had been rated as BI-RADS 4 in the clinic, and the other was an intraductal papilloma that had been rated as BI- RADS 5 in the clinic. The six biopsy-proven cancer cases had been rated as either BI-RADS 4 (four cases) or BI-RADS 5 (one case), with the exception of one case rated as BI-RADS 3. The remaining three examinations had anechoic cysts verified by ultrasound. The subjective breast density ratings distribution (BI-RADS) of these cases were two of 25 (8%), eight of 25 (32%), 14 of 25 (56%), and one of 25 (4%) for tissue density almost entirely fat, scattered fibroglandular density, heterogeneously dense, and extremely dense, respectively. Four board-certified, Mammography Quality and Standards Act qualified radiologists with varying breast imaging experience ranging from 9 to 26 years volunteered for the study. Readers received a study information packet before beginning the study. This packet included three documents: an Instructions to Observers document; a questions, definition, and rating instruction document; and a scoring form. The Instructions to Observers document defined the type of examination used in this study and that ultrasound may have been performed as part of the diagnostic workup in some of the cases but is not provided in this study. We also described the general setup and the process for reviewing and rating the examinations. Readers were told to assume the screening FFDM examination was the woman s baseline examination; hence, no prior FFDM examinations were provided for comparison. The questions, definition, and rating instruction document described the three tasks of the reader study. The objective of the first task was for the readers to provide a subjective assessment of how well the combination of FFDM and DBT examinations compared with the combination of FFDM and additional diagnostic views for the purpose of evaluating the index lesion. A scale of 3 to 3 was provided and used with 3 indicating that FFDM plus DBT was definitely worse for diagnosis compared with FFDM plus additional diagnostic views, 2 indicating that FFDM plus DBT was substantially worse for diagnosis, 1 indicating that FFDM plus DBT was somewhat worse for diagnosis, 0 indicating that FFDM plus DBT was equivalent or comparable for diagnosis compared with FFDM plus additional diagnostic views, 1 indicating that FFDM plus DBT was somewhat better for diagnosis, 2 indicating that FFDM plus DBT was substantially better, and 3 indicating that FFDM plus DBT was definitely better for diagnosis compared with FFDM plus additional diagnostic views. The objective of the second task was for the readers to provide a single BI-RADS score for each index lesion, taking into account all the information including the FFDM, DBT, and additional views. Readers were asked not to give a BI-RADS 0. The objective of the third task was for the readers to provide a binary answer (yes or no) to the question of whether or not the availability of DBT in each case eliminated the need for breast ultrasound. Because all images in this study were anonymized AJR:195, August 2010 W173
Hakim et al. TABLE 1: Distribution of Subjective Ratings for Diagnosis Using Full-Field Digital Mammography (FFDM) and Digital Breast Tomosynthesis (DBT) Compared With FFDM and Additional Diagnostic Views for Diagnosis Over All Readers Benign High Risk a Verified Cancer Overall Rating and given a case number in the DICOM header, the scoring form provided a table with the corresponding anonymized series of case numbers in the order in which the cases were to be reviewed on the workstation, with clearly marked spaces for manually entering the ratings for all three tasks. This study was performed using a modified SecurView mammography workstation (Hologic Inc.). This research workstation, which is PC based, includes two 5-megapixel LCD displays with a mammography workflow keypad. The system includes tools for magnification, zoom, and contrast adjustment as well as a drag-anddrop image display. The image display software allows the viewing of one, two, or four images per display for each monitor, so up to eight images could be displayed simultaneously. During the review, the craniocaudal and mediolateral oblique views of the breast with the index findings were displayed simultaneously on the left and right monitors, respectively. Then, in a crossmode approach, for the first 12 cases the readers were first shown the screening FFDM examination then the additional diagnostic views followed by the DBT examination. In the remaining 13 cases, the readers were first shown the FFDM examination then the DBT examination followed by the additional diagnostic views. Before providing the single overall BI-RADS score that took into account all modes (objective 2), the readers were allowed to review any of the modes. In 21 of the 25 cases at least one reader asked to do this before rating the case. The study was performed in one session with all the readers present at the workstation. Each reader reviewed the cases individually without any discussion of the case and rating was done completely independently. One case was presented at a time to all readers, and they were Frequency % Frequency % Frequency % Frequency % 95% CI for Percentage Definitely worse 0 0 0 0 Substantially worse 1 1.4 0 2 8.3 3 3.0 0.00 6.40 Somewhat worse 12 17.6 0 4 16.7 16 16.0 8.69 23.31 Equal 26 38.2 0 5 20.8 31 31.0 21.78 40.22 Somewhat better 17 25.0 2 25.0 6 25.0 25 25.0 16.36 33.64 Substantially better 9 13.2 4 50.0 3 12.5 16 16.0 8.69 23.31 Definitely better 3 4.4 2 25.0 4 16.7 9 9.0 3.29 14.71 Total 68 8 24 100 Note Dash indicates not applicable. a High-risk cases include atypical ductal hyperplasia and intraductal papilloma. given approximately 3.5 minutes to review each case. As a result, the session lasted approximately 1.5 hours. Data Analysis First, we computed the frequency and proportion of the subjective ratings and then of the BI-RADS ratings over all readers and cases for benign lesions, high-risk lesions, and cancer cases. For the purpose of this analysis, BI-RADS ratings 1 and 2 were combined. Finally, we computed the frequency of yes responses representing subjective assessments of the cases for which the need for ultrasound could be eliminated on the basis of the availability of DBT. Results Over all cases (25) and readers (four), FFDM and DBT (combined) were perceived to be better for diagnosis in 50% (50/100) of the subjective ratings (25 cases four readers = 100 ratings), at least equivalent for diagnosis in 31% (31/100) of the subjective ratings, and worse for diagnosis in 19% (19/100) of the subjective ratings compared with FFDM and additional diagnostic views (Table 1). The FFDM and DBT (combined) were perceived to be better for classification compared with the FFDM and additional diagnostic views in 54% (13/24) of the subjective ratings for the cancer cases, in 100% (8/8) of the ratings for high-risk cases, and in 43% (29/68) of the ratings for benign cases. Figure 1 shows a cancer case that was rated by all radiologists as definitely better for diagnosis with DBT compared with FFDM and additional diagnostic views. The 19 negative subjective ratings occurred in 13 cases in which at least one radiologist gave a negative subjective rating. Twelve of these were given to cases with heterogeneously dense tissue and seven negative ratings were given to cases that had scattered fibroglandular densities. The distribution of positive (including equivalent or 0) and negative preference ratings by breast density is shown in Table 2. We conducted a postrating review of the 13 cases for which at least one radiologist gave a negative subjective rating of the FFDM plus DBT viewing mode. This postrating review indicated that in three of the 13 cases (accounting for five of the 19 negative ratings) the reason for these negative ratings was the posterior location of the index lesion and that it was not imaged at DBT on at least one view. Over all readers, 69% (47/68) of the ratings for benign cases were rated as BI-RADS 1, 2, or 3, and 92% (22/24) of the ratings for verified cancer cases and 50% (4/8) of the ratings for high risk cases were rated as BI- RADS 4 or 5 (Table 3). In 32% of the cases (8/25 or 12% [12/100]) of the ratings) at least one of the radiologists indicated that the availability of DBT would have eliminated the need for ultrasound as a part of the diagnostic process. These 12 ratings were all associated with benign cases. Discussion There are little published data on the possible utility of DBT in diagnostic breast imaging. This very preliminary subjective study was designed to assess the need for specific future objective studies in this area and to evaluate the type of diagnostic questions that may be addressed with DBT in clinical practice. We found that, competing with a directed examination designed for a spe- W174 AJR:195, August 2010
Digital Breast Tomosynthesis cific purpose and optimized to enable viewing a specific suspected region within the breast (e.g., spot compression) is not an easy task, despite anecdotal and personal perception that DBT should be better for diagnostic purposes. Hence, because DBT is primarily a whole-breast examination that is conventionally positioned to cover the whole breast rather than a specific location, we had several cases in our data set with negative subjective ratings. Although DBT rated comparable or better in most instances, we had several examinations in which the directed examination (e.g., spot compression) was clearly viewed as better for addressing the specific question of interest. We note that in 26% (5/19) of these negative ratings, optimal positioning for imaging the index lesion was identified as the reason that the DBT examination was inferior. Interestingly, when asked about the cases rated for possible elimination of the need for ultrasound, readers commented that, in these cases, mass margins were visible much more clearly on the optimal slices in the DBT series so that they would be very comfortable diagnosing the mass in question as benign without the need for ultrasound. In addition to enabling preliminary observations, preliminary (pilot) studies, such as the one presented here, are extremely useful in refining hypotheses and improving the study design of larger pivotal studies. Most important, these preliminary studies allow a better estimate of the sample size requirement for the larger (pivotal) study. Among the limitations of this preliminary study is the selection of a small nonrandomized set of cases with a distribution of cases that had been actually rated as 3, 4, or 5 in the clinic; hence, this set may not be representative of the entire spectrum of diagnostic cases. Also, the readers may have been biased in that they unintentionally tried to rate the FFDM plus DBT mode as at least equal to if not better than the FFDM plus additional views mode. This is noted by the fact that no definitely worse ratings were given to any of the cases under the FFDM plus DBT viewing mode. On careful review of the six can- TABLE 3: Distribution of BI-RADS Ratings for Diagnosis Over All Readers Rating Benign High Risk a Verified Cancer Overall Frequency % Frequency % Frequency % Frequency % 95% CI for Percentage Assessment incomplete 0 0 0 0 Normal 6 8.8 0 0 6 6.0 1.26 10.74 Benign 20 29.4 0 1 4.2 21 21.0 12.88 29.12 Probably benign 21 30.9 4 50.0 1 4.2 26 26.0 17.25 34.75 Suspicious 17 25.0 4 50.0 10 41.7 31 31.0 21.78 40.22 Highly suspicious 4 5.9 0 12 50.0 16 16.0 8.69 23.31 Total 68 8 24 100 Note Dash indicates not applicable. a High-risk cases include atypical ductal hyperplasia and intraductal papilloma. TABLE 2: Distribution of Positive (Including Equivalent or 0) and Negative Preference Ratings by Breast Density Breast Density cer cases, there were two ratings (one case by two different radiologists) that could be considered false-negative (one rated as benign and one as probably benign). This case contained a mass that had grown from the prior year and the recommendation in the clinic to Fig. 1 57-year-old woman who was recalled for suspected spiculated density in posterior third medial aspect of right breast. A C, During diagnostic workup, right spot craniocaudal medially, right mediolateral oblique spot superiorly, and full 90 views were obtained. Slice images from digital breast tomosynthesis (DBT) image set left mediolateral oblique (A), full-field digital mammography left mediolateral oblique (B), and additional spot-compression left mediolateral oblique view (C) are shown. All readers in this study preferred DBT image for diagnosis of this malignant lesion (arrow). Preference Ratings Positive Negative Total Almost entirely fat 8 0 8 Scattered fibroglandular density 25 7 32 Heterogeneously dense 44 12 56 Extremely dense 4 0 4 Total 81 19 100 AJR:195, August 2010 W175
Hakim et al. biopsy was based on that change over time. Because we did not provide prior images in this study, we believe that this particular cancer would not have been actually missed under a fully simulated clinical setting. In conclusion, our preliminary study suggests that DBT is likely to become an important tool available to radiologists in diagnostic cases in which a possible mass, architectural distortion, or asymmetry is identified, and it is possible that in a fraction of women it will ultimately be selected as the preferred examination for this purpose. Determining the specific type and location of abnormalities for which DBT is likely to be better than additional views is clearly beyond the scope of this preliminary work. References 1. Mettlin C. Global breast cancer mortality statistics. CA Cancer J Clin 1999; 49:138 144 2. Smith RA. Breast cancer screening among women younger than age 50: a current assessment of the issues. CA Cancer J Clin 2000; 50:312 336 3. Pisano ED, Gatsonis C, Hendrick E, et al.; Digital Mammographic Imaging Screening Trial (DMIST) Investigators Group. Diagnostic performance of digital versus film mammography for breast-cancer screening. N Engl J Med 2005; 353:1773 1783 4. Bird RE, Wallace TW, Yankaskas BC. Analysis of cancers missed at screening mammography. Radiology 1992; 184:613 617 5. Lewin JM, D Orsi CJ, Hendrick RE, et al. Clinical comparison of full-field digital mammography and screen-film mammography for detection of breast cancer. AJR 2002; 179:671 677 6. Niklason LT, Christian BT, Niklason LE, et al. Digital tomosynthesis in breast imaging. Radiology 1997; 205:399 406 7. Boone JM, Nelson TR, Lindfors KK, Seibert JA. Dedicated breast CT: radiation dose and image quality evaluation. Radiology 2001; 221:657 667 8. Park JM, Franken EA, Garg M, et al. Breast tomosynthesis: present considerations and future applications. RadioGraphics 2007; [suppl 1]: S231 S240 9. Moore RH, Kopans DB, Rafferty EA, et al. Initial callback rates for conventional and digital breast tomosynthesis mammography comparison in the screening environment. (abstr) Proceedings of the Radiological Society of North America annual meeting. Chicago, IL: Radiological Society of North America, 2007; 381 10. Poplack SP, Tosteson TD, Kogel CA, Nagy HM. Digital breast tomosynthesis: initial experience in 98 women with abnormal digital screening mammography. AJR 2007; 189:616 623 11. Smith A, Niklason L, Jing Z. Performance of breast tomosynthesis as an adjunct imaging modality to digital mammography. (abstr). Proceedings of the European Congress of Radiology annual meeting. Vienna, Austria: European Congress of Radiology, 2008; 18:150 12. Smith A, Rafferty E, Niklason L. Breast tomosynthesis reduces radiologist performance variability compared to digital mammography. (abstr). Proceedings of the European Congress of Radiology annual meeting. Vienna, Austria: European Congress of Radiology, 2009; 19:151 13. Andersson I, Ikeda DM, Zackrisson S, et al. Breast tomosynthesis and digital mammography: a comparison of breast cancer visibility and BI- RADS classification in a population of cancers with subtle mammographic findings. Eur Radiol 2008; 18:2817 2825 14. Gur D, Abrams GS, Chough DM, et al. Digital breast tomosynthesis: an observer performance study. AJR 2009; 193:586 591 15. Good WF, Abrams GS, Catullo VJ, et al. Digital breast tomosynthesis: a pilot observer study. AJR 2008; 190:865 869 W176 AJR:195, August 2010