Women s Imaging Original Research

Transcription

1 Women s Imaging Original Research Brandt et al. DBT for Screening Recalls Without Calcifications Women s Imaging Original Research FOCUS ON: Kathleen R. Brandt 1 Daniel A. Craig 1 Tanya L. Hoskins 2 Tara L. Henrichsen 1 Emily C. Bendel 1 Stephanie R. Brandt 1 Jay Mandrekar 2 Brandt KR, Craig DA, Hoskins TL, et al. Keywords: 3D mammography, digital breast tomosynthesis, tomosynthesis DOI: /AJR Received March 9, 2012; accepted after revision September 26, Department of Radiology, Mayo Clinic, 200 First St, SW, Rochester, MN Address correspondence to K. R. Brandt (brandt.kathy@mayo.edu). 2 Department of Biostatistics, Mayo Clinic, Rochester, MN. AJR 2013; 200: X/13/ American Roentgen Ray Society Can Digital Breast Tomosynthesis Replace Conventional Diagnostic Mammography Views for Screening Recalls Without Calcifications? A Comparison Study in a Simulated Clinical Setting OBJECTIVE. This study evaluated digital breast tomosynthesis (DBT) as an alternative to conventional diagnostic mammography in the workup of noncalcified findings recalled from screening mammography in a simulated clinical setting that incorporated comparison mammograms and breast ultrasound results. SUBJECTS AND METHODS. One hundred forty-six women, with 158 abnormalities, underwent diagnostic mammography and two-view DBT. Three radiologists viewed the abnormal screening mammograms, comparison mammograms, and DBT images and recorded a DBT BI-RADS category and confidence score for each finding. Readers did not view the diagnostic mammograms. A final DBT BI-RADS category, incorporating ultrasound results in some cases, was determined and compared with the diagnostic mammography BI-RADS category using kappa statistics. Sensitivity and specificity were calculated for DBT and diagnostic mammography. RESULTS. Agreement between DBT and diagnostic mammography BI-RADS categories was excellent for readers 1 and 2 (κ = 0.91 and κ = 0.84) and good for reader 3 (κ = 0.68). For readers 1, 2, and 3, sensitivity and specificity of DBT for breast abnormalities were 100%, 100%, and 88% and 94%, 93%, and 89%, respectively. The clinical workup averaged three diagnostic views per abnormality and ultrasound was requested in 49% of the cases. DBT was adequate mammographic evaluation for 93 99% of the findings and ultrasound was requested in 33 55% of the cases. CONCLUSION. The results of this study suggest that DBT can replace conventional diagnostic mammography views for the evaluation of noncalcified findings recalled from screening mammography and achieve similar sensitivity and specificity. Two-view DBT was considered adequate mammographic evaluation for more than 90% of the findings. There was minimal change in the use of ultrasound with DBT compared with diagnostic mammography. F indings that may represent breast cancer are identified on 5 10% of screening mammograms and these women are recalled for diagnostic mammography. Supplemental mammographic views, designed to improve spatial resolution and reduce superimposition of overlapping tissue, are the mainstay of the diagnostic workup. These views, such as spot compression views, have been shown to improve the specificity of mammography [1, 2]. Many findings on screening mammography prove to be normal overlapping tissue on further evaluation. Digital breast tomosynthesis (DBT) is also designed to reduce the summation effects of tissue overlap and improve lesion conspicuity. Normal structures are often clearly recognized with DBT and suspicious findings become more apparent [3]. Several studies have compared DBT and diagnostic mammography with regard to im- age quality and lesion characterization. An early study by Poplack et al. [4] compared two-view DBT with film-screen diagnostic mammography for the evaluation of 99 findings on screening mammography. A single unblinded reader subjectively rated the image quality of DBT as equivalent (52%) or superior (37%) to diagnostic mammography in 89% of cases. Calcifications accounted for most of the cases in which DBT had inferior image quality. In another subjective comparison study by Hakim et al. [5], four radiologists reviewed full-field digital mammography (FFDM) images, additional diagnostic mammography views, and two-view DBT of 25 women with known masses. Compared with FFDM and additional diagnostic views, FFDM and DBT were perceived as better for diagnosis in 50%, at least equivalent for diagnosis in 31%, and worse for diagnosis in AJR:200, February

2 Brandt et al. 19%. The most common reason for a negative DBT rating in that study was a lesion that was not imaged at DBT on at least one view. In a study by Noroozian et al. [6], four readers evaluated 67 known masses with one- or two-view DBT and mammographic spot views. DBT performed comparably to mammographic spot views for characterizing a mass as benign or malignant. Mass visibility was slightly better with DBT, but this difference in visibility reached statistical significance for only one reader. Tagliafico et al. [7] compared mammographic spot views and two-view DBT of 52 women recalled from screening mammography and found the diagnostic accuracy of DBT to be at least equal to that of mammographic spot views. Most published studies have evaluated two-view DBT; however, Gennaro et al. [8] found one-view DBT was not inferior to twoview digital mammography in a population undergoing diagnostic mammography. In none of these published studies did the DBT assessment include the full complement of data that would be available in a clinical setting, such as comparison mammograms and breast ultrasound results. This study evaluated DBT as an alternative to conventional diagnostic mammography views in the workup of noncalcified asymmetries, areas of distortion, and masses recalled from screening mammography. The performance of DBT, which was compared with that of conventional diagnostic mammography, was assessed in a simulated clinical setting that incorporated comparison mammograms and breast ultrasound results. Subjects and Methods This study was performed under a protocol that included signed informed consent by participants to undergo DBT for research purposes and was approved by the institutional review board. The study coordinator, who is not a radiologist, reviewed the list of patients recalled from screening mammography to identify potential subjects. Any woman older than 40 years with a screening mammography abnormality that did not show calcifications was initially eligible for this study. Women who were pregnant, were lactating, or had implants were not eligible. All eligible subjects were offered enrollment when the DBT unit, the study coordinator, and a DBT-trained technologist were available. One hundred nine subjects prospectively enrolled between March 2009 and September Forty-one subjects prospectively enrolled between March 2010 and April From October 2009 to March 2010 the study was closed to recruitment because of DBT unit downtime. Enrolled subjects underwent conventional diagnostic imaging for the screening-detected abnormality and a two-view DBT examination. The breast radiologist assigned to the diagnostic clinic at the time of the evaluation determined the type of conventional diagnostic images for each subject on the basis of the appearance of the abnormality on screening mammography and his or her preference. The conventional diagnostic evaluation may have included digital spot compression views with or without magnification, rolled views, lateral views, exaggerated views, and ultrasound. The conventional diagnostic workup was completed in one visit and the final clinical BI-RADS category was determined by the interpreting breast radiologist on the basis of his or her analysis of all imaging findings. The DBT images were obtained on a Hologic Selenia Dimensions beta unit. At the time of this study, DBT had not yet been approved by the U.S. Food and Drug Administration (FDA). The DBT examination consisted of 15 low-dose projection images in the craniocaudal (CC) and mediolateral oblique (MLO) planes. The DBT images were sent to a research archive and were not available for review by the clinical radiologist at the time of the conventional diagnostic evaluation. All management decisions were based on conventional workup results only. At least 6 months later, three study breast radiologists independently viewed the DBT images on a research workstation (SecurView 6, Hologic). Readers were instructed to use DBT to evaluate the screening abnormality as if it were the primary diagnostic mammography workup. The screening mammograms, with annotations identifying the abnormality, and comparison mammograms that were present at the time of the screening mammography interpretation had been preloaded on a research workstation by an assistant and were reviewed by study readers. The research workstation was identical to the Hologic clinical workstation but had DBT viewing capability. Study readers did not have access to any images obtained during the conventional diagnostic workup. In addition, readers were blinded to patient identity, the radiologist who interpreted the conventional diagnostic evaluation, the diagnostic workup report, and the clinical results. Each reader had completed the 8 hours of DBT training required to be a certified reader but had little additional DBT experience. Readers 1, 2, and 3 had 16, 17, and 10 years of experience in breast imaging, respectively. The 15 DBT projection images and 1-mm reconstructed images were available to study readers in the CC and MLO planes. DBT images could be viewed as individual images or slabs of images, using either manual scrolling or a cine mode. After reviewing the abnormal screening mammogram, comparison mammograms, and DBT images, study readers recorded a forced DBT BI-RADS category (1 5, not 0) and a confidence score (1 3) for each screening-detected abnormality. Confidence scores, as defined in Table 1, were used to assess if DBT was sufficient for final diagnostic evaluation or if additional diagnostic imaging was desirable, such as conventional diagnostic mammography views or breast ultrasound. Readers were instructed to make this determination as if each case were an active clinical case. After completing the DBT evaluation for the 158 findings, one of the study radiologists retrospectively reviewed each subject s pertinent breast imaging reports and breast biopsy results. For each screening-detected abnormality the following information was recorded: a description of the screening mammography finding, the number of extra diagnostic mammography views obtained during the conventional diagnostic evaluation, ultrasound findings (if performed), the final clinical BI-RADS category, pathologic results from biopsies, and follow-up imaging results. In our practice, the clinical BI-RADS category was a combined mammography and ultrasound category if ultrasound was performed. The same study radiologist reviewed the recorded forced DBT BI-RADS category (1 5) and confidence score (1 3) for each reader and determined a final DBT BI-RADS category for each finding in the following manner. If the reader indicated a confidence score of 1 or 2 and ultrasound had been performed as part of the clinical evaluation, the results of ultrasound for that screening-detected abnormality, whether positive or negative, determined the final DBT BI-RADS category. If ultrasound was not performed clinically in these subjects, the forced DBT BI-RADS category remained the final DBT BI-RADS. The forced DBT BI-RADS also remained the final DBT BI-RADS TABLE 1: Confidence Score Definitions Confidence Score Definition 1 Additional conventional mammographic views and ultrasound needed 2 Ultrasound needed for further evaluation; ultrasound findings (positive or negative) would be accepted as the final result 3 No additional imaging needed for final BI-RADS assessment; if ultrasound is performed, only reason would be to determine whether ultrasound could guide biopsy 292 AJR:200, February 2013

3 DBT for Screening Recalls Without Calcifications when the confidence score was 3, whether or not ultrasound had been performed during the clinical evaluation. The final DBT BI-RADS category was used in the statistical analysis. For the purposes of statistical analysis, BI-RADS scores were collapsed from five categories to three categories based on how the score would be managed clinically: BI-RADS scores of 1 and 2 were grouped together because both require no additional followup. BI-RADS scores of 4 and 5 were combined because both scores would result in a biopsy. BI-RADS scores of 3 were maintained as a separate category because short-interval follow-up would be recommended. To assess agreement of the DBT BI-RADS category with the clinical BI-RADS category (derived from standard workup using diagnostic mammography with or without ultrasound), weighted kappa statistics were used. Interrater agreement about the DBT BI-RADS category was calculated between each pair of raters using weighted kappa statistics calculated with linear weights (Cicchetti- Allison method). The reference standard used for calculation of sensitivity, specificity, and accuracy was the result of the entire clinical workup, including imaging and biopsy if performed, as well as follow-up in those cases not biopsied. Positive lesions were those with biopsy-proven diagnosis of either cancer or a high-risk abnormality (i.e., atypical ductal hyperplasia [ADH], atypical lobular hyperplasia [ALH], lobular carcinoma in situ [LCIS]). Negative lesions were either biopsy-proven benign or had no evidence of breast cancer at follow-up. Subjects without follow-up imaging at our institution were not included in this analysis. Only lesions in the original callback areas were included in these calculations; additional findings noted only with DBT were reported separately. Sensitivity was calculated for both DBT and diagnostic mammography as the percentage of positive lesions that would have been biopsied based on a BI-RADS category of 4 or 5. Specificity was calculated for each modality as the percentage of negative lesions with follow-up that correctly would not have been biopsied (BI-RADS categories 1 3). Accuracy was defined as the percentage correctly classified as positive or negative according to those definitions. We calculated 95% CIs using the Wilson score for sensitivity, specificity, and accuracy. The sensitivity, specificity, and accuracy of DBT and diagnostic mammography were compared using the exact sign test for paired proportions. Data analysis was performed using statistics software (SAS, version 9.2, SAS Institute); p values < 0.05 were considered statistically significant. Results Conventional Workup and Clinical Results One hundred fifty subjects prospectively enrolled in this study: Four subjects were later excluded because they did not meet all of the eligibility criteria. One of the excluded subjects did not complete the DBT examination because of equipment malfunction, one was found to have calcifications in the area of concern on the screening mammogram, and two had an abnormality detected on either MRI or molecular breast imaging rather than screening mammography. Ten eligible subjects had two separate abnormalities identified at screening mammography, five in the same breast and five in the opposite breast. One subject had three TABLE 2: Performance of Each Reader Using Digital Breast Tomosynthesis (DBT) Compared With Diagnostic Mammography abnormalities, two in the same breast and one in the contralateral breast. These 158 screening-detected abnormalities, in 146 subjects, form the cohort for evaluation. Findings were in the right breast for 76 (48%) cases and in the left breast for 82 (52%) cases. The screening mammography abnormalities included 31 (20%) possible masses, 28 (18%) questionable areas of distortion, and 99 (63%) asymmetries all without associated calcifications. Additional mammographic views were obtained to further evaluate 157 of the 158 abnormalities. Ultrasound was the only diagnostic test obtained for evaluation of one of the 158 findings. The median number of additional mammographic views per abnormality was three (range, 1 6 views). Ultrasound was used clinically to further evaluate and/or determine a biopsy route in 49.4% (78/158) of the abnormalities. The clinical workup resulted in 86.1% (136/158) of the abnormalities with a final clinical BI-RADS (combined mammography and ultrasound if performed) category of 1 or 2 that required no additional workup. BI-RADS 3 was assigned to 3.8% (6/158). These six cases were unchanged to date with total follow-up ranging from 6 to 27 months (average, 16.3 months). The remaining 10.1% (16/158) of abnormalities were classified as BI-RADS 4 or 5 and underwent core needle biopsy. Five were invasive cancer and three were high-risk marker lesions (one each of ADH, ALH, and LCIS). The ADH was upstaged to invasive cancer and the other two were unchanged after surgical excision. Thus, six cancers and two high-risk marker lesions Clinical (Diagnostic Mammography With or Without Ultrasound) DBT With or Without Ultrasound BI-RADS Category 1 or 2 BI-RADS Category 3 BI-RADS Category 4 or 5 Reader 1 BI-RADS category 1 or BI-RADS category BI-RADS category 4 or Reader 2 BI-RADS category 1 or BI-RADS category BI-RADS category 4 or Reader 3 BI-RADS category 1 or BI-RADS category BI-RADS category 4 or Note Data are number of lesions. Concordant results are underlined and discordant are set in boldface. AJR:200, February

4 Brandt et al. were diagnosed from this screening population. The other eight BI-RADS 4 or 5 lesions were benign at needle biopsy or cyst aspiration. Of the 146 subjects in this study, 121 (83%) have undergone subsequent mammography at our institution. The average followup for these subjects was 19.4 months (range, 6 39 months). Two subjects had a significant A C breast abnormality diagnosed during the follow-up period. One subject was diagnosed with breast cancer 25 months after the study mammogram when an abnormality was detected on a molecular breast imaging study. The screening mammography at the time of diagnosis was interpreted as negative. In the other case, new calcifications were found on B D a screening mammography 15 months after the study mammogram. Needle biopsy revealed ADH that was confirmed by surgical excision. Although in the same breast, neither of these abnormalities was in the area recalled from screening mammography as part of this study. Thus, there was no evidence that the conventional diagnostic evaluation had failed to diagnose a cancer in study subjects. There was no follow-up breast imaging at our institution for 25 (17%) subjects. Final Digital Breast Tomosynthesis Results Each of the 158 lesions was evaluated independently by the three study readers using DBT instead of conventional diagnostic mammography. A comparison of the final DBT BI-RADS category for each reader with the clinical BI-RADS category is shown in Table 2. Weighted kappa statistics between DBT BI-RADS and the clinical diagnostic mammography BI-RADS were 0.91, 0.84, and 0.68 for readers 1, 2, and 3, respectively. Readers 1, 2, and 3 showed concordance with the clinical diagnostic mammography BI-RADS category in 153 (97%), 150 (95%), and 142 (90%) of the lesions, respectively. Interrater agreement for the final DBT BI- RADS (three categories [1 or 2, 3, 4 or 5]) was a weighted kappa of 0.87 between readers 1 and 2, 0.67 between readers 1 and 3, and 0.66 between readers 2 and 3. A weighted kappa value of < 0.40 was considered poor agreement; 0.40 κ 0.75, good agreement; and > 0.75, excellent agreement. Examples of DBT cases are shown in Figures 1 3. The sensitivity, specificity, and accuracy estimates for each DBT reader and for clinical diagnostic mammography are reported in Table 3. Findings without follow-up imaging (n = 27) that were not biopsied were not included in this analysis. There was no significant difference between DBT and diagnostic mammography with respect to sensitivity for any of the readers. Readers also did not differ significantly from diagnostic mammography on specificity (p = 1.0, p = 1.0, and p = 0.13) or accuracy (p = 1.0, p = 1.0, p = 0.07). CIs for sensitivity were wide because of the Fig year-old woman with invasive ductal carcinoma (arrows) detected on screening mammography and digital breast tomosynthesis (DBT). A and B, Mediolateral oblique (MLO) (A) and craniocaudal (CC) (B) screening mammographic views. C and D, MLO (C) and CC (D) DBT images. All three DBT readers would have biopsied this lesion. 294 AJR:200, February 2013

5 DBT for Screening Recalls Without Calcifications A C E B D F small number of positive cases. The outcome for DBT BI-RADS category 4 or 5 lesions and additional findings with DBT in the same breast are recorded in Table 4 for each reader. Readers assigned BI-RADS 4 or 5 to 12 different findings that were assessed clinically as BI-RADS category 1 or 2 or category 3 and that, thus, were not biopsied. In five of these 12 cases, DBT readers indicated that ultrasound, if performed, would have affected the final DBT BI-RADS category. The other seven lesions would have been biopsied by study readers without ultrasound. Study readers also reported seven additional findings with DBT in the same breast. Five of these findings had no evidence of malignancy on follow-up imaging ranging from 21 to 36 months, one was a cyst, and one had no follow-up. There was no evidence that DBT correctly identified a significant lesion that was not seen on conventional imaging in the same breast as the screening abnormality. The confidence score recorded for each abnormality was used to assess the completeness of the DBT evaluation. Results indicated that two-view DBT was considered adequate mammographic evaluation for 99% (156/158) of the findings for readers 1 and 2 and for 93% (147/158) for reader 3. Readers 1, 2, and 3 would have ordered breast ultrasound for further evaluation in 44% (69/158), 28% (45/158), and 48% of cases (76/158), respectively. Breast ultrasound would have been ordered by readers in an additional 5% (8/158), 4% (7/158), and 7% (11/158) only to determine whether ultrasound could be used to guide a biopsy (confidence score of 3). Discussion DBT has been approved for clinical use in the screening and diagnostic settings, but its role in breast imaging is still evolving. Published studies support screening DBT to reduce unnecessary recalls due to overlapping normal tissue [3, 4, 9]. However, there are issues to be resolved before screening DBT is likely to gain widespread acceptance. These issues include identifying the optimal imaging protocol (one- or two-view), determin- Fig year-old woman with invasive ductal carcinoma (arrows) detected on screening mammography and digital breast tomosynthesis (DBT) and conventional diagnostic images (for comparison). A and B, Mediolateral oblique (MLO) (A) and craniocaudal (CC) (B) screening mammographic views. C and D, MLO (C) and CC (D) DBT images. E and F, MLO (E) and CC (F) spot compression views. All three DBT readers would have biopsied this lesion. AJR:200, February

6 Brandt et al. A C Fig year-old woman recalled from screening mammography for possible mass (arrows) that was found to be normal dense parenchyma by conventional diagnostic mammography and ultrasound. A and B, Mediolateral oblique (MLO) (A) and craniocaudal (CC) (B) screening mammographic views. C and D, MLO (C) and CC (D) DBT images. Screening mammogram obtained 1 year later (not shown) was negative for malignancy. None of three readers would have biopsied this area and two of three would not have performed ultrasound. B D ing how to triage patients for DBT if it cannot be offered to all screening patients, and accounting for the higher cost of DBT units and digital storage [9 11]. In addition, early studies suggested that radiologist interpretation time will increase with DBT and may not improve even with extensive training [12]. This blinded, multireader study compared DBT with conventional imaging in the diagnostic workup of abnormalities recalled from screening mammography. If shown to be adequate, DBT might more quickly be incorporated into diagnostic breast imaging practices. To evaluate DBT in an environment similar to clinical diagnostic imaging, comparison mammograms were available to study readers and breast ultrasound results were included in the final DBT assessment. The results of this study showed that DBT has potential to replace conventional diagnostic mammography for evaluating noncalcified asymmetries, areas of distortion, and masses recalled from screening mammography. There was no significant difference between DBT and diagnostic mammography with respect to sensitivity, specificity, or accuracy for any of the readers. Agreement between DBT BI- RADS and the clinical diagnostic mammography BI-RADS was excellent for readers 1 and 2 and good for reader 3. Although not statistically significant, reader 3 had lower specificity (89%) and lower accuracy (89%) using DBT as compared with diagnostic mammography (94%). The specificity of DBT was affected by the method of incorporating ultrasound results into the final DBT BI-RADS. Reader 3 showed more reliance on ultrasound than the other readers. Eight findings that reader 3 categorized as DBT BI-RADS 4 or 5 were not biopsied. In half of these cases, ultrasound was not performed during the clinical workup but was requested by reader 3. These cases were classified as BI-RADS 1 or 2 by the other readers and with conventional imaging. In real practice, reader 3 would have ordered ultrasound and, if the ultrasound findings were negative or benign, changed the BI-RADS category and improved specificity. One cancer that was diagnosed as a result of the conventional workup would not have been biopsied by reader 3 using DBT. This was a case of subareolar distortion that was perceived using DBT but dismissed because of stability. Agreement between readers in this study, weighted kappa values ranging from 0.66 to 0.87, indicated that the variability in radiologist interpretation and practice preferences that are present in conventional breast imag- 296 AJR:200, February 2013

7 DBT for Screening Recalls Without Calcifications TABLE 3: Sensitivity, Specificity, and Accuracy for Each Digital Breast Tomosynthesis (DBT) Reader and for the Conventional Imaging Results Modality Sensitivity (%) Specificity (%) Accuracy (%) DBT with or without ultrasound Reader (8/8) [68 100] 94 (115/123) [88 97] 94 (123/131) [88 97] Reader (8/8) [68 100] 93 (114/123) [87 96] 93 (122/131) [87 96] Reader 3 88 (7/8) [53 98] 89 (110/123) [83 94] 89 (117/131) [83 94] Diagnostic mammography with or without ultrasound (clinical) 100 (8/8) [68 100] 94 (115/123) [88 97] 94 (123/131) [88 97] Note Data in parentheses are raw data used to calculate performance values and data shown in brackets are 95% CIs. TABLE 4: Analysis of Outcomes for Digital Breast Tomosynthesis (DBT) BI-RADS Category 4 or 5 and Additional Findings With DBT for Each Reader Cancer or High-Risk Lesion Biopsies Reader DBT BI-RADS 4 or 5 Benign Biopsies No. Follow-Up on These Subjects Ultrasound Requested But Not Performed No. Follow-Up , no follow-up 0 5 1, no follow-up; 1, cyst; ing were also present with DBT. Compared with the conventional imaging results, none of the study readers showed improved sensitivity or specificity with DBT. Based on available follow-up at our institution, there was no evidence that additional cancers were detected with DBT either at the site that generated the callback or elsewhere in the same breast. When two-view DBT was used for evaluation of the screening mammography finding, readers considered DBT adequate mammographic evaluation in 93 99% of the cases. After reviewing the CC and MLO DBT images, study readers indicated that conventional extra views were still needed for complete mammographic assessment for 1 7% of the findings. Incomplete visualization of the finding on DBT and continued uncertainty were reasons for needing conventional mammography views after DBT. During the conventional diagnostic mammography workup, a median of three additional mammographic views (range, 1 6 views) were obtained per screening-detected abnormality. Positioning for additional mammographic views, such as spot compression views, can be challenging Additional Biopsy Recommended With DBT, Not Biopsied Clinically for technologists. Up to six extra views were obtained in some subjects to ensure that the area in question was included in the smaller FOV or that overlapping tissue had been adequately compressed. Positioning for DBT is much simpler, being identical to positioning for routine mammography, and the problem with tissue superimposition is greatly reduced. Thus, DBT has potential to reduce the number of additional mammographic views obtained during a diagnostic workup and improve technologist efficiency. Results from all readers indicated continued reliance on ultrasound with DBT. Ultrasound would have been requested by study readers to further evaluate 33 55% of the screening mammography findings after reviewing the DBT images. Ultrasound was performed during the clinical evaluation after diagnostic mammography for further evaluation of 49.4% of the findings. A larger decline in cases requiring ultrasound was expected because DBT should reduce concern that a true lesion was obscured by dense tissue. Hakim et al. [5], in a smaller study of known masses, found a 12% decline in the use of ultrasound. Also, the presence of prior Additional Finding With DBT (Ipsilateral Breast) 3, negative findings on follow-up (1 mastectomy, 21 mo, and 24 mo) , no follow-up; 1 3 1, cyst (same as reader 1); 2, negative findings on follow-up (16 and 21 mo) , no follow-up; 6, negative findings on follow-up (16 38 mo) Note NA = not applicable. 4 0 NA 2, negative findings on follow-up (36 and 37 mo) mammograms for comparison would be expected to reduce the need for further evaluation. At the time of our study, DBT was an emerging technology, one that had not even been approved by the FDA, and our study readers had limited clinical experience with DBT interpretation. This learning curve phenomenon may partially explain the higher-than-expected use of ultrasound. As radiologists become more familiar with DBT, we anticipate that ultrasound will be used less frequently to confirm normal tissue. A persistent mass or distortion detected with DBT will likely still require ultrasound for further evaluation. In contrast to most published studies [5, 6, 11], our study did not compare DBT with conventional mammography with regard to image quality, lesion visibility, or lesion characterization. Neither the conventional diagnostic images nor the final clinical results were available to study readers. Thus, the DBT assessment was determined without bias from the conventional diagnostic mammography evaluation. Study readers had exactly the same imaging information, including comparison mammograms, that was available to the diagnostic radiologist AJR:200, February

8 Brandt et al. when he or she began the clinical diagnostic workup. In addition, breast ultrasound results, when available, were incorporated with DBT as they were with diagnostic mammography in the clinical assessment. Ultrasound is such an important tool in diagnostic breast imaging that to understand the impact of a new modality ultrasound must be taken into consideration. The method of incorporating ultrasound results in this study was a limitation when ultrasound had not been performed clinically but was requested by study readers. We acknowledge that asking readers to grade their level of confidence in the DBT diagnosis, as we did in this study, may not translate consistently to clinical practice where decisions affect patient care. However, study readers were explicitly instructed to make their decisions as if these cases were active clinical cases. Based on a literature review, this study was the first diagnostic DBT study that simulated clinical practice by incorporating comparison mammograms and breast ultrasound results into the study design. DBT did not show improved sensitivity or specificity compared with conventional diagnostic mammography in this study. The study design favored results from the conventional workup and false-negatives would be detected only by follow-up imaging. The lack of follow-up or biopsy in 27 of the 158 (17%) lesions was a limitation of this study. In a large tertiary medical center such as ours, the lack of follow-up is expected because many of our patients receive routine medical care closer to their homes. Another limitation of this study is the small number of significant lesions diagnosed in study subjects. The average screening recall rate in our practice is just under 10%. Thus, approximately 1500 screening mammograms were interpreted to yield the 146 patients that were recalled to yield eight significant lesions. This yield is congruent with the Mammography Quality Standards Act audit guidelines, which anticipate 2 10 cancers detected per 1000 screening mammograms [13]. Although we believe the number of positive findings was representative of the population, the small number of positive cases means that we cannot estimate sensitivity very precisely, as shown by wide CIs, and that our results are not well powered to detect differences in sensitivity. A final potential limitation of this study was that the study readers did not interpret all of the screening mammograms or conventional diagnostic examinations. These were interpreted by one of 18 breast radiologists in our practice. The threshold for recall, number of additional views, and use of ultrasound vary among radiologists. We also found variations between the three study readers using DBT. A larger study with more subjects and more study readers would improve precision and statistical power and allow us to better characterize heterogeneity among DBT readers. In conclusion, the results of this study suggested that DBT could replace conventional diagnostic mammography views for the evaluation of noncalcified findings recalled from screening mammography and could achieve similar sensitivity and specificity. Two-view DBT was considered adequate mammographic evaluation for more than 90% of the findings compared with an average of three extra mammographic views per finding with diagnostic mammography. There was minimal change in the use of ultrasound with DBT compared with diagnostic mammography. References 1. Berkowitz JE, Gatewood OM, Gayler BW. Equivocal mammographic findings: evaluation with spot compression. Radiology 1989; 171: Sickles EA. Combining spot-compression and other special views to maximize mammographic information. (letter) Radiology 1989; 173: Gur D, Abrams GS, Chough DM, et al. Digital breast tomosynthesis: observer performance study. AJR 2009; 193: Poplack SP, Tosteson TD, Kogel CA, Nagy HM. Digital breast tomosynthesis: initial experience in 98 women with abnormal digital screening mammography. AJR 2007; 189: Hakim CM, Chough DM, Ganott MA, Sumkin JH, Zuley ML, Gur D. Digital breast tomosynthesis in the diagnostic environment: a subjective side-by-side review. AJR 2010; 195:528; [web] W172 W Noroozian M, Hadjiiski L, Rahnama-Moghadam S, et al. Digital breast tomosynthesis is comparable to mammographic spot views for mass characterization. Radiology 2012; 262: Tagliafico A, Astengo D, Cavagnetto F, et al. Oneto-one comparison between digital spot compression view and digital breast tomosynthesis. Eur Radiol 2012; 22: Gennaro G, Toledano A, di Maggio C, et al. Digital breast tomosynthesis versus digital mammography: a clinical performance study. Eur Radiol 2010; 20: U.S. Food and Drug Administration website. P Hologic Selenia Dimensions 3D System: summary of safety and effectiveness data ( ). Published Committees/CommitteesMeetingMaterials/Medical Devices/MedicalDevicesAdvisoryCommittee/ RadiologicalDevicesPanel/UCM pdf. Accessed October 16, 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: Wallis MG, Moa E, Zanca F, et al. Two-view and single-view tomosynthesis versus full-field digital mammography: high resolution x-ray imaging observer study. Radiology 2012; 262: Zuley ML, Bandos AI, Abrams GS, et al. Time to diagnosis and performance levels during repeat interpretations of digital breast tomosynthesis: preliminary observations. Acad Radiol 2010; 17: Bassett LW, Hendrick RE, Bassford TL, et al. Quality determinants of mammography: clinical practice guideline no. 13. Rockville, MD: Agency for Health Care Policy and Research, Public Health Service, U.S. Department of Health and Human Services, October 1994: AHCPR publication no AJR:200, February 2013