Diagnostic Performance of a Dedicated 1.5-T Breast MR Imaging System 1

Transcription

1 Note: This copy is for your personal non-commercial use only. To order presentation-ready copies for distribution to your colleagues or clients, contact us at Bruce J. Hillman, MD Steven E. Harms, MD Gary Stevens, PhD Rebecca G. Stough, MD Alan B. Hollingsworth, MD Kamilia F. Kozlowski, MD Lawrence J. Moss, MD Diagnostic Performance of a Dedicated 1.5-T Breast MR Imaging System 1 Purpose: Materials and Methods: To assess diagnostic performance of dedicated breast magnetic resonance (MR) imaging at breast imaging centers by using a dedicated 1.5-T breast MR system that used high-spatial-resolution, high-contrast-resolution spiral trajectory acquisitions. The study was institutional review board approved and HIPAA compliant, with waiver of informed consent. Diagnostic performance was retrospectively assessed for 934 consecutive screening (n = 347) and diagnostic (n = 587) examinations performed from April 2006 to December 2007 in women aged years old from four sites for which dedicated breast MR imaging reports and ground truth (biopsy for cancer cases, 1-year follow-up with negative results for cases with negative findings) were available. The sensitivity, specificity, and receiver operating characteristic (ROC) for breast MR imaging were determined. Original Research n Breast Imaging Results: Conclusion: The sensitivity and specificity for the dedicated breast MR imaging system were 92% (92 of 100) and 88.8% (741 of 834). For all cases, the negative predictive value (NPV) was 98.9% (741 of 749). The NPV for screening cases was 100% (326 of 326). The area under the ROC curve was Of the 93 cases with false-positive findings seen at dedicated breast MR imaging, 25 (27%) were high-risk histologic findings for which excision is often recommended. The false-positive rate was 93 of 834 (11.2%) for all cases, but only 16 of 326 (4.9%) for the screening cohort. High accuracy was achieved by using dedicated breast MR imaging. q RSNA, From ACR Image Metrix, Philadelphia, Pa (B.J.H.); Department of Radiology, University of Virginia, Charlottesville, Va (B.J.H.); the Breast Center of Northwest Arkansas, 55 W Sunbridge Dr, Fayetteville, AR (S.E.H.); Dynastat Consulting, Bastrop, Tex (G.S.); Mercy Women s Center, Oklahoma City, Okla (R.G.S., A.B.H.); Knoxville Comprehensive Breast Center, Knoxville, Tenn (K.F.K.); and Department of Radiology, University of Massachusetts Medical School, Worcester, Mass (L.J.M.). Received March 28, 2011; revision requested May 10; final revision received September 29; accepted October 7; final version accepted May 3, Address correspondence to S.E.H. ( seharms@earthlink.net). Supplemental material: /suppl/doi: /radiol /-/dc1 q RSNA, 2012 Radiology: Volume 265: Number 1 October 2012 n radiology.rsna.org 51

2 Breast magnetic resonance (MR) imaging has been shown to be a highly sensitive (71% 100%) method for the detection of breast cancer (1 10). However, examinations with false-negative findings do occur. The lack of a discernible mass, as with ductal carcinoma in situ (DCIS) and lobular carcinoma, is the most common reason for MR examinations with falsenegative findings (11 20). In addition, the relatively poor specificity of breast MR imaging (68% for the American College of Radiology Imaging Network [ACRIN] 6883 trial), compared with the specificity of mammography (93% 100%), has been a major concern (1 10). The false-positive rates cited by researchers in prior studies are 32.2% (1), 41% (22), and 35% (23). Studies with false-positive findings lead to further testing and unnecessary biopsies, thereby increasing costs, morbidity, and patient anxiety. False-positive diagnoses may result from the incapability of MR to distinguish subtle cancers from benign findings that overlap in morphologic characteristics and flow dynamics. We hypothesized that improvements in spatial and contrast resolution would improve morphologic characterization, thus resulting in lower numbers of breast MR examinations with false-positive and false-negative findings than historically reported for breast MR imaging with the use of whole-body imagers (24). Our purpose was to assess diagnostic performance of dedicated breast MR imaging at breast imaging centers by using a dedicated 1.5-T breast MR system that used Advance in Knowledge nn The use of a dedicated 1.5-T breast MR imaging system can result in a high negative predictive value of 98.9% while also having a relative low false-positive rate (positive predictive value of 49.7%); the false-positive rate for all cases was only 11.2%, and the false-positive rate for the screening cohort was only 4.9%. high-spatial-resolution, high-contrastresolution spiral trajectory acquisitions. Materials and Methods The design, conduct, and analysis of the study were performed by ACR Image Metrix (Philadelphia, Pa), an independent imaging contract research organization, which also oversaw the writing of this manuscript under contract to Aurora Imaging Technology (North Andover, Mass). Grant support for the study was provided by Aurora Imaging Technology. Two authors (B.J.H. and G.S.), who had control of the data, are independent contractors for ACR Image Metrix and had no conflicts of interest. One author (S.E.H.) is a stockholder and medical director of Aurora Imaging Technology and is the trial principal investigator. Another author (B.J.H.) is the chief medical officer of ACR Image Metrix. Still another author (L.J.M.) is the medical director of Aurora Breast MRI of Central Massachusetts (Worcester, Mass). Subject Eligibility and Recruitment The project was approved by the institutional review board for each site and was Health Insurance Portability and Accountability Act compliant. Informed consent was waived, as this study was a retrospective study. All subjects were retrospectively identified from the records of consecutive examinations performed from April 2006 to December 2007 at four breast centers that each had at least 1 year of experience with the dedicated breast MR imaging system. Sites were selected for their differences in locations and practice variations: (a) academic (University of Massachusetts, Worcester, Mass), (b) hospital based (Mercy, Oklahoma City, Implication for Patient Care nn The low frequency of false-positive and false-negative findings associated with the dedicated breast MR imaging system allows for use of breast MR imaging with low risk of patient harm in both the screening and diagnostic environments. Okla), and (c) private clinics (Knoxville Comprehensive Breast Center, Knoxville, Tenn, and the Breast Center of Northwest Arkansas, Fayetteville, Ark). The breast MR volume at the sites varied from three to six cases per day, with a mean of 4.75 cases per day. In the mammography comparison cohort, all images were digital (three sites used Hologic [Bedford, Mass] and one used GE Healthcare [Waukesha, Wis] equipment). For the ultrasonography (US) comparison cohort, the US examinations were performed at all sites with transducer frequencies of MHz. All US examinations were performed by and findings were interpreted by breast imagers, including 12 radiologists, among whom were four authors (S.E.H., R.G.S., K.F.K., and L.J.M.). To be eligible for accrual, 934 women between the ages of 25 and 89 years underwent a breast MR imaging examination with the dedicated breast MR imaging equipment for either screening (n = 347) or diagnostic (n = 587) purposes. Screening examinations were indicated on the basis of high risk factors, such as having a family history of breast cancer, being a carrier of or a relative of women Published online before print /radiol Content code: Radiology 2012; 265:51 58 Abbreviations: ACRIN = American College of Radiology Imaging Network A z = area under the ROC curve BI-RADS = Breast Imaging Reporting and Data System CAD = computer-aided detection CI = confidence interval DCIS = ductal carcinoma in situ NPV = negative predictive value PPV = positive predictive value ROC = receiver operating characteristic Author contributions: Guarantors of integrity of entire study, B.J.H., S.E.H.; study concepts/study design or data acquisition or data analysis/ interpretation, all authors; manuscript drafting or manuscript revision for important intellectual content, all authors; approval of final version of submitted manuscript, all authors; literature research, S.E.H.; clinical studies, B.J.H., S.E.H., G.S., R.G.S., K.F.K., L.J.M.; statistical analysis, B.J.H., S.E.H., G.S., K.F.K.; and manuscript editing, B.J.H., S.E.H., G.S., R.G.S., A.B.H. Potential conflicts of interest are listed at the end of this article. 52 radiology.rsna.org n Radiology: Volume 265: Number 1 October 2012

3 with genetic mutations associated with breast cancer risk, having high-risk syndromes, having high-risk histologic findings, having dense breasts, or having a personal history of breast cancer. Diagnostic studies were indicated on the basis of local staging; presence of a palpable mass; presence of an axillary mass with no known primary cancer; presence of nipple discharge, retraction, or inversion; presence of skin findings; presence of breast enlargement; and inconclusive results at mammography or US. We eliminated from the analysis subjects who lacked ground truth data (n = 50), subjects who had missing MR data (Breast Imaging Reporting and Data System [BI-RADS] category 0 cases) (n = 46), subjects who had neither ground truth nor MR data (n = 26), subjects who represented duplicated cases (n = 40), or subjects who had BI-RADS category 6 lesions (n = 166). Of the 50 patients excluded for lack of ground truth data, the BI-RADS categories for lesions and corresponding numbers of patients were as follows: BI-RADS category 1, n = 18; BI-RADS category 2, n = 24; BI-RADS category 3, n = 2; BI-RADS category 4, n = 1; and BI-RADS category 5, n = 1. Data Collection and Management Specific data elements recorded by the record extractors for breast MR imaging, mammography, and breast US performed within 3 months of the MR imaging (when available) included whether the imaging was performed as a screening or diagnostic study, the BI- RADS category for the imaging examination results, the outcome of breast biopsy or surgery, and the results of recent mammography and/or US and other follow-up imaging examinations. MR Imaging All studies were performed by using a 1.5-T dedicated breast MR imaging system (Aurora Imaging Technology). This system employs a spiral-trajectory k-space acquisition method and the rotating delivery of excitation off resonance water excitation pulse sequence, which samples reconstruction space with a spiral trajectory to improve the speed and efficiency of MR data collection, resulting in better image contrast and spatial resolution (25 27). As opposed to typical MR acquisitions with the use of three-dimensional Fourier transform where only one projection is taken from the contrast-dependent portion of reconstruction space, all 5120 spiral projections originate in the center of reconstruction space. Therefore, spiral trajectory acquisitions oversample the image contrast dependent portion of reconstruction space, resulting in improved contrast resolution (25). This is especially advantageous for breast imaging because of the potential temporal variability of the peak of contrast enhancement. For each case, a nonenhanced nonspoiled rotating delivery of excitation off resonance acquisition was followed by four contrast material enhanced spoiled rotating delivery of excitation off resonance acquisitions performed immediately after intravenous bolus administration of gadolinium-based contrast material (0.1 mmol/kg) (28). Each acquisition was 3 minutes in length. The image matrix was , with an in-plane resolution of 1 mm and a section thickness of 1.1 mm. The display matrix was All images were interpreted at a dedicated workstation where dynamic computer-aided detection (CAD) was performed, and dynamic curves were generated. In addition to dynamics, the CAD also depicted pure fluid and edema as distinct color codes overlaid on the image. The workstation provides maximum intensity projection and multiplanar reformation as part of the visualization package. Original interpretations were performed by a total of 12 specialized breast radiologists with either extensive experience in breast imaging (six radiologisits) or experience and fellowship training (six radiologists). The breast MR reading experience of the radiologists varyied from 1 to 17 years, with a mean of 3.3 years. The interpreters had access to whatever prior clinical and/or imaging information was available at the time of the original reading. Analysis In the analysis, the MR imaging diagnoses were compared with ground truth to calculate sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) on a per-subject basis. For cases with positive findings, ground truth was defined by either a surgical and/or biopsy result reporting histologic evidence of breast cancer. For cases with negative findings, ground truth was defined as a breast MR imaging examination, mammography, or breast US performed at least 1 year following the index breast MR imaging examination with negative findings. For our primary calculation of sensitivity and specificity, cases with findings classified as BI- RADS categories 4 5 were considered cases with positive findings; cases with findings classified as BI-RADS categories 1 3 were counted as cases with negative findings. We did not include cases with findings classified as BI-RADS category 0. We also evaluated the effect on the accuracy metrics of sensitivity, specificity, PPV, and NPV of using BI-RADS categories 3 5 to classify a case as one with positive findings. In a subanalysis, we also compared performance according to site. We used the BI-RADS category assignments to plot a receiver operating characteristic (ROC) curve as a depiction of overall diagnostic performance. We constructed ROC curves, as well as 95% confidence intervals (CIs), and calculated the area under the ROC curve (A z ). Because sites had access to the results of US and mammography while interpreting breast MR images, we simply present data on true- and false-positive diagnoses but make no formal comparisons with the performance of breast MR imaging. All summary statistics were analyzed by using software (SAS version 9.3; SAS, Cary, NC). Results Description of the Sample The sites accrued 1262 subjects. After exclusions, there were a total of 934 subjects who had undergone 347 screening examinations and 587 diagnostic examinations. These 934 subjects included 834 subjects with ground truth negative (benign) cases and 100 subjects with cancer. A total of 624 subjects also had qualifying mammograms (529 with Radiology: Volume 265: Number 1 October 2012 n radiology.rsna.org 53

4 negative results), and of 624 subjects 95 had cancer, 86 had true-positive results, and 443 had true-negative results; 436 subjects had qualifying breast sonograms (352 with negative results), and of 436 subjects, 84 had cancer, 62 had true-positive results, and 290 had true-negative results. By using BI-RADS categories 4 5 as positive interpretations, the sensitivity was 92% (92 of 100; 95% CI: 84.4%, 96.8%), and the specificity was 88.8% (741 of 834; 95% CI: 86.5%, 90.9%). The PPV was 49.7% (92 of 185), and NPV was 98.9% (741 of 749). The A z for MR imaging was An example of positive MR findings is depicted in Figure 1. For the analysis that included BI-RADS category 3 interpretations as positive, the sensitivity increased to 94% (94 of 100; 95% CI: 87.4%, 97.7%), but the specificity decreased to 77.2% (644 of 834; 95% CI: 74.2%, 80.0%). When BI-RADS category 3 interpretations were included as positive, the PPV was 66.9% (190 of 284), and the NPV was 99.1% (644 of 650). Two cases of cancer were classified as BI-RADS category 3 and were DCIS. The other cases that were classified as BI-RADS category 3 were benign and were considered false-negative results. For the five screening cases that contained a cancer, MR imaging was sensitive (five of five [100%]) (Table 1). The specificity for MR imaging screening studies was 95.3% (326 of 342). In regard to MR imaging diagnostic studies, the results indicated that they were sensitive (92%, 87 of 95) and specific (84.3%, 415 of 492). In a total of 25 of 93 (27%) MR examinations with false-positive findings (Table 2), women were considered to have high-risk histologic findings for which excision or chemoprevention is commonly recommended, and these findings included papilloma (n = 18), lobular carcinoma in situ (n = 3), atypical ductal hyperplasia (n = 2), and atypical lobular hyperplasia (n = 2) (Table 3). The examinations with false-negative findings consisted of those with subtle DCIS (two of six, 33%) and infiltrating ductal carcinoma (four of six, 67%). An example of false-negative MR findings is shown in Figure 2. Discussion The dedicated breast MR system led to better diagnostic performance for all case metrics (sensitivity, specificity, NPV, PPV, and A z ) than has been historically reported for breast MR imaging that employs conventional imaging units (1 10,18 23). The ACRIN 6883 trial is an especially apt comparator because investigators in that study evaluated the diagnostic capability of breast MR for patients with highly suspicious lesions prior to biopsy. The sensitivity (96.6%, 225 of 233), specificity (80.3%, 416 of 518), and NPV (98.1%, 416 of 424) achieved by using the dedicated breast MR in this study are all much higher than reported in the ACRIN 6883 trial (sensitivity, 88.1% [356 of 404]; specificity, 67.4% [281 of 417]; and NPV, 86.5% [281 of 325]) (1). It may be argued that the ACRIN 6883 study was conducted when interpretation was not as sophisticated. However, reanalysis of data in 2006 by using combined architectural and dynamic criteria in multivariate models of masses showed an A z of only 0.88, with univariate criteria having significantly lower performance (A z, ), especially for nonmasslike enhancement (2). Other studies show similar findings. Wedegartner et al (17) showed an A z of 0.90 for masses with irregular contour, an A z of 0.70 for those with inhomogeneous enhancement, an A z of 0.81 for those with irregular shape, and an A z of 0.64 for those with ring enhancement. Nonmasslike enhancement typically results in lower diagnostic performance than does enhancement for masses. In our study, MR imaging performed better, with an A z of for all lesion morphologic appearances including nonmasslike enhancement. One-quarter of the cases with positive findings were DCIS. Investigators in many MR studies report a lower diagnostic capability for in situ carcinoma, which is most often depicted as nonmasslike enhancement. In the ACRIN 6883 trial, the sensitivity for detecting DCIS was 73% (46 of 63) (1). Some studies show similar difficulties. Liberman et al (19) detailed the highest PPV for spiculated masses (80%, four of five), somewhat lower PPV for irregular masses (32%, 12 of 37), and considerably lower PPV for nonmasslike features (25%, 10 of 40). Guttierez et al (20) looked at morphologic characteristics as predictors of cancer and observed PPVs for masses of up to 46%, but the best nonmasslike enhancement predictor was only 31%. Baltzer et al (21) reported a PPV for masses of 86.1% (105 of 122), but they observed a low PPV for nonmasses (52%, 15 of 29). In this study, 27.7% (64 of 231) of the cancers were DCIS (21). Low-grade DCIS is often missed at breast MR imaging; however, in our study, six of 23 (26%) of DCIS cases were low grade (grade 1), and five of 23 (22%) DCIS cases were intermediate grade (grade 2). Examinations with false-negative findings may result in missing a potentially curable cancer. The ACRIN 6883 trial cautions that an MR examination with negative results should not be used as evidence to deny a biopsy for a mammographically or ultrasonographically suspicious finding (1). The much greater NPV in our study (98.9%, 741 of 749) could have profound clinical benefits, in some cases obviating biopsy. The NPV for lack of enhancement in the ACRIN 6883 trial was 85.4% (281 of 329) in the original report (1). If lack of enhancement was used as an additional criterion for a negative finding, the NPV improved only slightly to 88.0% (183 of 208) (2). The lack of specificity of breast MR in previous studies increases the ultimate cost of patient care by encouraging redundant testing and unnecessary biopsies, as well as potentially increasing morbidity and anxiety for the patient. Incidental lesions seen at MR imaging for which biopsy was recommended occurred in 103 of 423 patients (24.3%) in the ACRIN 6883 trial. The PPV for these findings was 13.2% (56 of 423) (29). In the Magnetic Resonance Imaging in Breast Screening trial, the recall rate for MR imaging was about three times higher than for mammography (10.7% vs 3.9%) (7). In the original Kuhl et al study (14), 18% of patients required follow-up MR imaging to decrease false-positive interpretations and improve the PPV. If 54 radiology.rsna.org n Radiology: Volume 265: Number 1 October 2012

5 Figure 1 Figure 1: Enhancing lesion on breast MR images in a 47-year-old woman referred for a suspicious mass in the left breast at mammography and US. (a) Nonenhanced nonspoiled T1- and T2-weighted oblique maximum intensity projection shows extensive hyperintense cysts. (b) Contrast-enhanced spoiled T1-weighted without T2 weighting oblique maximum intensity projection shows an occult enhancing mass (arrow) in the right breast but this is difficult to distinguish from multiple hyperintense cysts that have a short T1 because of proteinaceous fluid. (c) Immediately after contrast agent administration, subtraction axial section shows the enhancing cancer (arrow). Cysts and tumor are easily discriminated by using positive- and negative-scale subtractions. Cysts appear black on the negative scale of the subtraction because cysts are hyperintense on nonspoiled images and hypointense on spoiled images. Biopsy showed an infiltrating ductal carcinoma. The CAD program looks not only for dynamic characteristics but also for fluid content. (d) CAD overlay shows the cancer with washout (red), cysts (blue), and edema (green). Mammographic, US, and additional MR images are shown in Figure E1 (online). the initial MR impression was used, the PPV decreased to 64.2% (nine of 14), and the specificity decreased to 75% (72 of 96) (4). The high PPV (62.5%, 15 of 24) for the Sardanelli et al (10) trial was achieved by repeating breast MR and then deciding on biopsy on the basis of the results of the follow-up examination. The Sardanelli et al article (10) did not provide the PPV on the basis of the results of the initial MR imaging evaluation. Warner et al (6) applied Radiology: Volume 265: Number 1 October 2012 n radiology.rsna.org 55

6 Table 1 Diagnostic Performance of Screening and Diagnostic Dedicated Breast MR Imaging Parameter and Modality Table 3 Histologic Findings of 92 Breast MR Imaging Examinations with True-Positive Findings Histologic Finding No. of Examinations* Mean Size (mm) Infiltrating ductal carcinoma 58/92 (63) 23.1 (1 75) Grade 1 17/58 (29) Grade 2 17/58 (29) Grade 3 20/58 (34) Unspecified 4/58 (7) DCIS 23/92 (25) 26.7 (1 52) Grade 1 6/23 (26) Grade 2 5/23 (22) Grade 3 10/23 (43) Unspecified 2/23 (9) Infiltrating lobular carcinoma 6/92 (7) 19.9 (10 40) Papillary carcinoma 1/92 (1) 30 Colloid or mucinous carcinoma 2/92 (2) 46 (40 52) Other not specified 11/92 (12) 20 (7 35) Overall (1 75) * Number in parentheses is the percentage. Numbers in parentheses are ranges. screening to a very high-risk population of BRCA-positive subjects and reported a higher PPV of 42% (11 of 26), but the sensitivity was also lower (85%, 11 of 13). In our trial, the high sensitivity (100%, five of five; 95% CI: 47.8%, No. of Cases for Screening and Diagnostic MR Percentage Estimate 95% CI Sensitivity Screening and diagnostic MR 92/ , 96.8 Screening MR 5/ , 100 Diagnostic MR 87/ , 96.3 Specificity Screening and diagnostic MR 741/ , 90.9 Screening MR 326/ , 97.3 Diagnostic MR 415/ , 87.5 Accuracy Screening and diagnostic MR 833/ , 91.1 Screening MR 331/ , 97.3 Diagnostic MR 502/ , 88.3 PPV Screening and diagnostic MR 92/ , 57.2 Screening MR 5/ , 47.2 Diagnostic MR 87/ , 60.9 NPV Screening and diagnostic MR 741/ , 99.5 Screening MR 326/ , 100 Diagnostic MR 416/ , %) achieved for the screening cohort was attained with a very acceptable PPV (23.8%, five of 21); however, given the small number of cancers in our screening population, the CI for sensitivity was quite large. Table 2 Histologic Findings of 93 Breast MR Imaging Examinations with False-Positive Findings Histologic Finding No or little increased 68 (73) risk association Benign breast tissue 26 (28) Fibrocystic changes 13 (14) Fibroadenoma 8 (9) Fibrosis 4 (4) Adenosis 4 (4) Sclerosing adenosis 4 (4) Apocrine metaplasia 2 (2) Fat necrosis 2 (2) Usual hyperplasia 2 (2) Duct ectasia 1 (1) Hamartoma 1 (1) Spindle cell lesion 1 (1) Associated with 25 (27) increased risk Papilloma 18 (19) Lobular carcinoma 3 (3) in situ Atypical lobular 2 (2) hyperplasia Atypical ductal 2 (2) hyperplasia Overall 93 No. of Examinations* * Numbers in parentheses are percentages of total examinations with false-positive findings. Percentages were rounded. This value represents 72% of 25 increased-risk lesions. Percentage was rounded. This value represents 12% of 25 increased-risk lesions. Percentage was rounded. This value represents 8% of 25 increased-risk lesions. Percentage was rounded. The reported false-positive rate for screening breast MR trials for initial screenings ranges from 5% to 29% (4 9). The false-positive rate for diagnostic breast MR studies is typically in the 32% 41% range (1,22,23). The false-positive rate for our study was only 11.2% (93 of 834). For screening examinations, the false-positive rate was only 4.9% (16 of 326). We attribute the lower false-positive rate in our trial to a better ability to discern benign enhancement from truly malignant enhancement as a result of better spatial and contrast resolution in the dedicated system (25). 56 radiology.rsna.org n Radiology: Volume 265: Number 1 October 2012

7 Figure 2 Figure 2: False-negative MR findings in a 46-year-old woman with mammographically detected microcalcifications. (a) Magnification cranial-caudal view shows cluster of microcalcifications (arrow) in the left breast for which this patient was recalled from screening mammography. The MR image was interpreted as negative. (b) Immediate contrast-enhanced subtracted maximum intensity projection and (c) reformatted sagittal image (not subtracted) retrospectively show branching enhancement (arrow) in a location similar to the location of the microcalcifications. (d) Dynamic CAD shows a red lesion (arrow), representing washout enhancement. The stereotactic biopsy revealed atypical ductal hyperplasia. DCIS was found on the subsequent excision. Additional MR images are shown in Figure E2 (online). The greatest limitation of this study was its generalizability to other settings. MR imaging was performed at dedicated breast centers with expertise in breast imaging. However, the comparison studies were all from subspecialized radiologists in academic centers who used rigorous diagnostic criteria in a research setting. It is unlikely that these findings are totally caused by interpretation skills alone. The patient sample is heavily weighted toward diagnostic, rather than screening, examinations. That this is the case means that many patients had positive findings at mammography or US prior to undergoing the MR imaging examinations. Because insurance companies often require a positive finding at biopsy before MR, in many women, diagnostic screening was performed after biopsy. Inclusion of these cases would bias results. This study is further limited by its retrospective design. However, the potential for selection bias is mitigated by our sampling of patients consecutively and by our use of the original clinical interpretation. As with most screening trials, the number of total cancers detected is small and the 95% CI for sensitivity is 47.8% to 100%. In conclusion, our study demonstrates results that led to better diagnostic performance for all case metrics (sensitivity, specificity, NPV, PPV, and A z ) than has been historically reported for breast MR imaging that employs conventional imaging units. Radiology: Volume 265: Number 1 October 2012 n radiology.rsna.org 57

8 Acknowledgments: We thank Derek Harms, Lindsey Collins, Lynne Meeks, Mindy Schultz-Fee, Kim Davis, and Michelle Yesalusky for their efforts in data entry for this study. We thank Brenda Young and Elena Kachanova from the American College of Radiology for their assistance in data management. We thank Rochelle Keen for her assistance in management of the project. Disclosures of Potential Conflicts of Interest: B.J.H. Financial activities related to the present article: institution received consulting fee or honorarium, support for travel to meetings for the study or other purposes, fees for participation in review activities such as data monitoring boards, statistical analysis, end point committees, and the like, payment for writing or reviewing the manuscript from ACR Image Metrix. Financial activities not related to the present article: received payment for board membership from Philips Healthcare, consultancy fees from ACR Image Metrix and American College of Radiology (journal editorship), royalties for The Sorcerer s Apprentice: How Medical Imaging is Changing Health Care, and payment for lectures. Other relationships: none to disclose. S.E.H. Financial activities related to the present article: none to disclose. Financial activities not related to the present article: received payment as medical director, stock or stock options, and travel, accommodates, or meeting expenses from Aurora Imaging Technology. Other relationships: none to disclose. G.S. Financial activities related to the present article: none to disclose. Financial activities not related to the present article: received consultancy fees for statistical analysis of data from Image Metrix. Other relationships: none to disclose. R.G.S. Financial activities related to the present article: institution received financial stipend for chart review services from Aurora Imaging Technologies. Financial activities not related to the present article: received stock options for serving on medical advisory board, payment for travel expenses, and payment for travel, accommodations, and meeting expenses as RSNA consultant from Aurora Imaging Technologies. Other relationships: none to disclose. A.B.H. Financial activities related to the present article: none to disclose. Financial activities not related to the present article: received payment as medical director and board member from Breast MRI of Oklahoma and payment for two presentations over the past 36 months from Aurora Imaging Technology. Other relationships: none to disclose. K.F.K. No potential conflicts of interest to disclose. L.J.M. No potential conflicts of interest to disclose. References 1. Bluemke DA, Gatsonis CA, Chen MH, et al. Magnetic resonance imaging of the breast prior to biopsy. JAMA 2004;292(22): Schnall MD, Blume J, Bluemke DA, et al. Diagnostic architectural and dynamic features at breast MR imaging: multicenter study. Radiology 2006;238(1): Kuhl CK, Schmutzler RK, Leutner CC, et al. Breast MR imaging screening in 192 women proved or suspected to be carriers of a breast cancer susceptibility gene: preliminary results. Radiology 2000;215(1): Kuhl CK, Weigel S, Schrading S, et al. Prospective multicenter cohort study to refine management recommendations for women at elevated familial risk of breast cancer: the EVA trial. J Clin Oncol 2010;28(9): Rijnsburger AJ, Obdeijn IM, Kaas R, et al. BRCA1-associated breast cancers present differently from BRCA2-associated and familial cases: long-term follow-up of the Dutch MRISC Screening Study. J Clin Oncol 2010;28(36): Warner E, Plewes DB, Hill KA, et al. Surveillance of BRCA1 and BRCA2 mutation carriers with magnetic resonance imaging, ultrasound, mammography, and clinical breast examination. JAMA 2004;292(11): Leach MO, Boggis CR, Dixon AK, et al. Screening with magnetic resonance imaging and mammography of a UK population at high familial risk of breast cancer: a prospective multicentre cohort study (MAR- IBS). Lancet 2005;365(9473): Lehman CD, Isaacs C, Schnall MD, et al. Cancer yield of mammography, MR, and US in high-risk women: prospective multiinstitution breast cancer screening study. Radiology 2007;244(2): Lehman CD, Blume JD, Weatherall P, et al. Screening women at high risk for breast cancer with mammography and magnetic resonance imaging. Cancer 2005;103(9): Sardanelli F, Podo F, D Agnolo G, et al. Multicenter comparative multimodality surveillance of women at genetic-familial high risk for breast cancer (HIBCRIT study): interim results. Radiology 2007;242(3): Soderstrom CE, Harms SE, Copit DS, et al. Three-dimensional RODEO breast MR imaging of lesions containing ductal carcinoma in situ. Radiology 1996;201(2): Orel SG, Mendonca MH, Reynolds C, Schnall MD, Solin LJ, Sullivan DC. MR imaging of ductal carcinoma in situ. Radiology 1997;202(2): Menell JH, Morris EA, Dershaw DD, Abramson AF, Brogi E, Liberman L. Determination of the presence and extent of pure ductal carcinoma in situ by mammography and magnetic resonance imaging. Breast J 2005;11(6): Kuhl CK, Mielcareck P, Klaschik S, et al. Dynamic breast MR imaging: are signal intensity time course data useful for differential diagnosis of enhancing lesions? Radiology 1999;211(1): Rodenko GN, Harms SE, Pruneda JM, et al. MR imaging in the management before surgery of lobular carcinoma of the breast: cor- relation with pathology. AJR Am J Roentgenol 1996;167(6): Weinstein SP, Orel SG, Heller R, et al. MR imaging of the breast in patients with invasive lobular carcinoma. AJR Am J Roentgenol 2001;176(2): Harms SE. The use of breast magnetic resonance imaging in ductal carcinoma in situ. Breast J 2005;11(6): Wedegärtner U, Bick U, Wörtler K, Rummeny E, Bongartz G. Differentiation between benign and malignant findings on MRmammography: usefulness of morphological criteria. Eur Radiol 2001;11(9): Liberman L, Morris EA, Lee MJ, et al. Breast lesions detected on MR imaging: features and positive predictive value. AJR Am J Roentgenol 2002;179(1): Gutierrez RL, DeMartini WB, Eby PR, Kurland BF, Peacock S, Lehman CD. BI-RADS lesion characteristics predict likelihood of malignancy in breast MRI for masses but not for nonmasslike enhancement. AJR Am J Roentgenol 2009;193(4): Baltzer PA, Benndorf M, Dietzel M, Gajda M, Runnebaum IB, Kaiser WA. False-positive findings at contrast-enhanced breast MRI: a BI-RADS descriptor study. AJR Am J Roentgenol 2010;194(6): Teifke A, Hlawatsch A, Beier T, et al. Undetected malignancies of the breast: dynamic contrast-enhanced MR imaging at 1.0 T. Radiology 2002;224(3): Fischer U, Kopka L, Grabbe E. Breast carcinoma: effect of preoperative contrastenhanced MR imaging on the therapeutic approach. Radiology 1999;213(3): Hollingsworth AB, Stough RG, O Dell CA, Brekke CE. Breast magnetic resonance imaging for preoperative locoregional staging. Am J Surg 2008;196(3): Harms SE, Harms SS. Making quality technical decisions: No free lunch in MRI. Semin Breast Dis 2008;11(2): Block KT, Frahm J. Spiral imaging: a critical appraisal. J Magn Reson Imaging 2005;21(6): Meyer CH, Hu BS, Nishimura DG, Macovski A. Fast spiral coronary artery imaging. Magn Reson Med 1992;28(2): Harms SE, Flamig DP, Hesley KL, et al. MR imaging of the breast with rotating delivery of excitation off resonance: clinical experience with pathologic correlation. Radiology 1993;187(2): Schnall MD, Blume J, Bluemke DA, et al. MRI detection of distinct incidental cancer in women with primary breast cancer studied in IBMC J Surg Oncol 2005;92(1): radiology.rsna.org n Radiology: Volume 265: Number 1 October 2012