RSNA 2016

Abstract Archives of the RSNA, 2016


RC215

Breast Series: Hot Topics in Breast Imaging

Monday, Nov. 28 8:30AM - 12:00PM Room: Arie Crown Theater

BR

AMA PRA Category 1 Credits ™: 3.25
ARRT Category A+ Credits: 4.00

FDA Discussions may include off-label uses.

Christiane K. Kuhl, MD, Bonn, Germany (Moderator) Nothing to Disclose
Linda Moy, MD, New York, NY (Moderator) Nothing to Disclose
Edward A. Sickles, MD, San Francisco, CA (Moderator) Nothing to Disclose
LEARNING OBJECTIVES

ABSTRACT

Sub-Events
RC215-01
Jennifer A. Harvey, MD, Charlottesville, VA, (jharvey@virginia.edu) (Presenter) Research Grant, Hologic, Inc; Stockholder, Hologic, Inc; Research Grant, Volpara Health Technologies Limited; Stockholder, Volpara Health Technologies Limited;
LEARNING OBJECTIVES

1) Describe current political environment surrounding breast density. 2) State degree of decreased sensitivity due to dense tissue. 3) State risk of density and breast cancer, relative to traditional risk factors.

ABSTRACT

RC215-02
Kathleen R. Brandt, MD, Rochester, MN (Presenter) Nothing to Disclose
Meng-Kang Hsieh, Philadelphia, PA (Abstract Co-Author) Nothing to Disclose
Christopher G. Scott, MS, Rochester, MN (Abstract Co-Author) Nothing to Disclose
Lauren Pantalone, BS, Philadelphia, PA (Abstract Co-Author) Nothing to Disclose
Matthew Jensen, Rochester, MN (Abstract Co-Author) Nothing to Disclose
Stacey Winham, PhD, Rochester, MN (Abstract Co-Author) Nothing to Disclose
Dana H. Whaley, MD, Rochester, MN (Abstract Co-Author) Nothing to Disclose
Carrie B. Hruska, PhD, Rochester, MN (Abstract Co-Author) Institutional license agreement, Gamma Medica, Inc
Fang Fang Wu, Rochester, MN (Abstract Co-Author) Nothing to Disclose
Aaron Norman, Rochester, MN (Abstract Co-Author) Nothing to Disclose
Vernon S. Pankratz, Albuquerque, NM (Abstract Co-Author) Nothing to Disclose
Andrew Oustimov, Philadelphia, PA (Abstract Co-Author) Nothing to Disclose
Emily F. Conant, MD, Philadelphia, PA (Abstract Co-Author) Consultant, Hologic, Inc; Consultant, Siemens AG
Karla Kerlikowske, MD, San Francisco, CA (Abstract Co-Author) Nothing to Disclose
Despina Kontos, PhD, Philadelphia, PA (Abstract Co-Author) Nothing to Disclose
Celine M. Vachon, Rochester, MN (Abstract Co-Author) Nothing to Disclose
PURPOSE

Area and volumetric breast density measures are strong risk factors for breast cancer (BC).  We compared breast density estimates from a publically available, fully-automated software tool, the Laboratory for Individualized Breast Radiodensity Assessment (LIBRA), which can be run both on “For Processing” and “For Presentation” digital mammogram formats, to those from commercial software, which are only run on “For Processing” format.

METHOD AND MATERIALS

Digital mammograms in both formats were obtained prior to diagnosis on 437 incident BC cases and 1225 age-matched controls from a large screening mammography practice. LIBRA estimates included dense area (DA) and percent density (PD) averaged from four mammogram views of both digital mammogram formats.  Volumetric percent density (VPD) and dense volume (DV) estimates were also obtained on four views of “For Processing” formats only, using Volpara (Matakina Ltd.) and Quantra (Hologic Inc.) software.  We compared density measures using Pearson correlations (R) among controls, and odds ratios and 95% confidence intervals (OR (95%CI)) for BC per standard deviation (SD) density measure from conditional logistic regression, adjusting for age and body mass index.

RESULTS

LIBRA PD showed strong correlation with Volpara VPD (R=0.80-0.87), but moderate correlation with Quantra VPD (R=0.53-0.60).  LIBRA DA was low to moderately correlated with Quantra DV (R=0.28-0.52) and Volpara DV (R=0.52-0.65).  The strongest associations of LIBRA with BC were seen with “For Presentation” density measures, OR=1.3 (1.1-1.5) per SD of PD and OR=1.2 (1.1-1.4) per SD of DA, while estimates from “For Processing” images were attenuated: OR=1.1 (1.0-1.3) and OR=1.1 (0.97-1.2), per SD of PD and DA respectively.  For commercial measures, risk estimates for VPD with BC were slightly larger (OR=1.4 (1.2-1.6) and OR=1.3 (1.1-1.4) per SD VPD) than DV (OR=1.2 (1.1-1.4) and OR=1.2 (1.0-1.3) per SD DV) for Volpara and Quantra respectively, but not significantly different.

CONCLUSION

Our results confirm prior smaller studies showing that LIBRA, a publically available, fully-automated breast density estimation software run on readily available “For Presentation” mammograms, has similar BC associations as commercial software.  

CLINICAL RELEVANCE/APPLICATION

A publically available, fully-automated software utilizing “For Presentation” images could further enable research on quantitative density measures in personalized screening and cancer risk assessment.

RC215-03
Hui Li, MD, PhD, Chicago, IL (Presenter) Nothing to Disclose
Benjamin Q. Huynh, Chicago, IL (Abstract Co-Author) Nothing to Disclose
Maryellen L. Giger, PhD, Chicago, IL (Abstract Co-Author) Stockholder, Hologic, Inc; Stockholder, Quantitative Insights, Inc; Co-founder, Quantitative Insights, Inc; Royalties, Hologic, Inc; Royalties, General Electric Company; Royalties, MEDIAN Technologies; Royalties, Riverain Technologies, LLC; Royalties, Mitsubishi Corporation; Royalties, Toshiba Corporation;
Natalia O. Antropova, Chicago, IL (Abstract Co-Author) Nothing to Disclose
Li Lan, Chicago, IL (Abstract Co-Author) Nothing to Disclose
PURPOSE

We evaluated the potential of deep learning in the assessment of breast cancer risk in which convolutional neural networks (CNNs) extract features directly from FFDM images instead of measuring breast density and parenchymal textures.

METHOD AND MATERIALS

The study included 456 clinical FFDM cases from two high-risk datasets - BRCA1/2 gene-mutation carriers (53 cases) and unilateral cancer patients (75 cases), and a low-risk dataset (328 cases). All FFDM images [12-bit quantization and 100 micron pixels] had been acquired with a GE Senographe 2000D system and were retrospectively collected under an IRB-approved, HIPAA-compliant protocol. Regions of interest (ROIs) of 256x256 pixels were selected from the central breast region behind the nipple in the craniocaudal projection of the FFDMs.  We compared the use of direct image features, which were automatically extracted using transfer learning and pre-trained CNNs, and the use of features from radiographic texture analysis (RTA).  Each feature set was input to a support vector machine classifier and underwent leave-one-case-out cross validation. Area under the ROC curve (AUC) served as the figure of merit in the task of distinguishing between high-risk and low-risk subjects.

RESULTS

In the task of distinguishing between the BRCA1/2 gene-mutation carriers and low-risk women, comparable classification performance was obtained using features extracted from CNNs (AUC=0.83; SE=0.03) and from RTA (AUC=0.82; SE=0.03). However, in the task of distinguishing between unilateral cancer and low-risk women, the performance was significantly improved with the CNNs (AUC=0.82; SE=0.03) compared to RTA (AUC=0.73; SE=0.03) with a p-value of 0.009.

CONCLUSION

Deep learning with CNNs appears to be able to extract textural characteristics directly from FFDMs as well as, or better than, conventional texture analysis in the task of distinguishing between cancer risk populations.

CLINICAL RELEVANCE/APPLICATION

Deep learning has potential to help clinicians in assessing mammographic parenchymal patterns for breast cancer risk assessment.

RC215-04
James G. Mainprize, PhD, Toronto, ON (Presenter) Institutional research agreement, General Electric Company
Olivier Alonzo-Proulx, Toronto, ON (Abstract Co-Author) Institutional research agreement, General Electric Company
Roberta A. Jong, MD, Toronto, ON (Abstract Co-Author) Nothing to Disclose
Heba M. Hussein, MD, toronto, ON (Abstract Co-Author) Nothing to Disclose
Jennifer A. Harvey, MD, Charlottesville, VA (Abstract Co-Author) Research Grant, Hologic, Inc; Stockholder, Hologic, Inc; Research Grant, Volpara Health Technologies Limited; Stockholder, Volpara Health Technologies Limited;
Martin J. Yaffe, PhD, Toronto, ON (Abstract Co-Author) Research collaboration, General Electric Company Founder, Matakina International Ltd Shareholder, Matakina International Ltd Co-founder, Mammographic Physics Inc
PURPOSE

The sensitivity of screening mammography is reduced for dense breasts.  BI-RADS density assessment emphasizes the masking potential of breast density, but the assessment is qualitative and achieves only moderate agreement between radiologists. We are refining a masking index based on the local “detectability map”, that predicts the probability of missing a cancer (if present) due to the amount and patterns of dense tissue in the breast.  The maps are automatically calculated from the image and the DICOM header contents. 

METHOD AND MATERIALS

Simulated breast cancer lesions are sequentially “inserted” into digital mammograms (contralateral to actual cancer), one location at a time. An automated computer “observer” is used to combine measured image features (contrast, noise power, texture) to create a detectability map whose pixels represent the probability of detecting a possible cancer at each location in the mammogram. From the map, a masking index, giving the probability of missing a cancer due to reduced contrast or clutter caused by superposition of dense breast tissue, is calculated for the entire mammogram. Under IRB approval, for initial development, we analyzed a set of de-identified mammograms consisting of 8 cancers missed on mammography and 40 screen-detected cancers. Results from a  larger set consisting of 106 interval cancers and 596 screen-detected cancers are currently under analysis. The masking index predictions are compared against truth status (cancer missed vs. cancer found).

RESULTS

In our preliminary ROC analysis, the ability to predict detected/missed status was found to have an AUC of 0.74 (0.54-0.87, 95% confidence interval).  Interestingly, volumetric breast density alone was not as informative in predicting missed cancer status, AUC 0.67 (0.4-0.84 CI).

CONCLUSION

A masking index is being developed that has shown promise in predicting the probability of masking in a mammogram.  It is based on detectability maps that indicated regions of high/ low detectability and have been shown to correlate with radiologists’ impressions of mammograms and screening performance.

CLINICAL RELEVANCE/APPLICATION

A quantitative, objective measure of masking could become a key tool in determining when the performance of screening mammography in an individual woman is compromised due to breast density.

RC215-05
Bennett Battle, MD, Little Rock, AR (Presenter) Nothing to Disclose
Sharp F. Malak, MD, MPH, Little Rock, AR (Abstract Co-Author) Nothing to Disclose
Ishwori Dhakal, MPH, Little Rock, AR (Abstract Co-Author) Nothing to Disclose
Jeannette Lee, PhD, Little Rock, AR (Abstract Co-Author) Nothing to Disclose
Noel Keith, Little Rock, AR (Abstract Co-Author) Nothing to Disclose
Barbara Fuhrman, PhD, Little Rock, AR (Abstract Co-Author) Nothing to Disclose
PURPOSE

Previously, methods for measurement of mammographic density (MD) have been either time-consuming, or subjective and only modestly reliable. Because MD offers information both about individual risk and about the efficacy of screening, a standardized and robust method for measurement of this prevalent risk factor has been a goal of many researchers.

METHOD AND MATERIALS

Using Volpara software (Matakina, New Zealand), we measured volumetric mammographic density on images from 42,527 screening mammograms done on 13,942 women seen at our institution between 2006-2012. Billing data and data collected by the cancer registrar was gathered to document person-level and procedure-level factors on 25,034 women seen for mammography assessment in the same period. A retrospective cohort was defined using women with both imaging and billing data who were initially seen for a screening mammogram, had no previous history of breast cancer, and were followed at our institution for at least 6 months following their first captured visit.

RESULTS

Among 5746 eligible women, 92 registry-confirmed breast cancers and a total of 121 registry-confirmed or treated breast cancer cases were ascertained. We observed monotonically increasing risks of registry-verified incident breast cancers by quartile of VMD% with HR= 1.0, 1.2 (0.7-2.2), 1.2 (0.7-2.2), and 2.2 (1.2-4.1), Ptrend=0.02; a similar trend was seen by quartile of DV, with HR across quartiles = 1.0, 1.5 (0.8-2.9), 1.8 (0.9-3.4), and 2.9 (1.5-5.4) (Ptrend=0.0009). Among women without a breast cancer diagnosis, changes in MD were significantly modified by birth cohort; while both VMD% and DV increased significantly in serial images taken over time in women born before 1940, VMD% declined significantly over time (P<0.0001) and DV decreased on average, but not statistically significantly (p=0.46), in women born between 1950-1959.

CONCLUSION

Automated measurement of Volumetric MD allows for assessment of this important breast cancer risk factor in a large number of women and on repeated mammographic assessments. In this cohort, assembled using the Enterprise Data Warehouse, mammographic density was associated with increased risk of subsequent breast cancer as expected, supporting the validity of the automated measures.

CLINICAL RELEVANCE/APPLICATION

Automated volumetric mammographic density measurement allows for the identification of women at increased risk of developing breast cancer.

RC215-06
Lin Chen, Philadelphia, PA (Abstract Co-Author) Nothing to Disclose
Lauren Pantalone, BS, Philadelphia, PA (Abstract Co-Author) Nothing to Disclose
Andrew Oustimov, Philadelphia, PA (Abstract Co-Author) Nothing to Disclose
Meng-Kang Hsieh, Philadelphia, PA (Abstract Co-Author) Nothing to Disclose
Emily F. Conant, MD, Philadelphia, PA (Abstract Co-Author) Consultant, Hologic, Inc; Consultant, Siemens AG
Despina Kontos, PhD, Philadelphia, PA (Presenter) Nothing to Disclose
PURPOSE

We evaluate the association of breast parenchymal texture features to breast cancer risk with full-field digital mammography (FFDM) versus digital breast tomosynthesis (DBT) central source projections (CSP) images. 

METHOD AND MATERIALS

We retrospectively analyzed images from women who underwent routine breast cancer screening at our institution using a combined FFDM and DBT protocol (Selenia Dimensions, Hologic Inc.), during 9/2011-6/2014.  Each DBT source projection was acquired at approximately 90% dose reduction, for a total of 15 source projection images per breast and view. A total of 86 women were diagnosed with unilateral invasive breast cancer. From these, 72 had “For Processing” DBT-CSP images available for analysis that were of sufficient image quality (no artifacts, implants, pace-makes), and were used as cases. A total of 360 controls were randomly selected from women who had negative screening exams and confirmed one-year negative follow-up, which were age-, race- and side-matched to cases at 1:5 ratio. The mediolateral oblique (MLO) view of the “For Processing” FFDM and CSP-DBT images were used for parenchymal pattern analysis; for cases the contralateral (unaffected images) were analyzed. Multiple texture descriptors were extracted, including gray-level histogram, co-occurrence, run-length, and fractal dimension features,  using a previously validated, fully-automated, lattice-based texture analysis pipeline. The discriminatory capacity of the texture features based on FFDM versus DBT-CSP was tested via a logistic regression classifier and the area under curve (AUC) of the receiver operating characteristic (ROC). The ROC AUCs were compared using DeLong’s test.

RESULTS

The model with FFDM texture features had an AUC=0.75 (95% CI:0.69-0.82). Texture features from the corresponding low-dose DBT-CSP images had an AUC=0.76 (95% CI:0.70-0.82). No significant difference was found between FFDM and DBT-CSP based on the performance of the lattice texture features (p=0.87).

CONCLUSION

Our study suggests that parenchymal texture analysis from DBT-CSP images, acquired at substantially lower x-ray dose, is feasible and may result to similar associations to breast cancer risk compared to standard-dose FFDM images.

CLINICAL RELEVANCE/APPLICATION

Parenchymal texture analysis can provide robust indicators of cancer risk from low-dose DBT-CSP images, which may become important as DBT is increasingly replacing FFDM in breast cancer screening.  

RC215-07
Mark A. Helvie, MD, Ann Arbor, MI (Presenter) Institutional Grant, General Electric Company
LEARNING OBJECTIVES

1) Define overdiagnosis of breast cancer. 2) Differentiate overdiagnosis from overtreatment. 3) Estimate magnitude of overdiagnosis from screening.

ABSTRACT

RC215-08
Nico Karssemeijer, PhD, Nijmegen, Netherlands (Presenter) Shareholder, Matakina Technology Limited Consultant, QView Medical, Inc Shareholder, QView Medical, Inc Director, ScreenPoint Medical BV Shareholder, ScreenPoint Medical BV
Katharina Holland, Nijmegen, Netherlands (Abstract Co-Author) Nothing to Disclose
Ioannis Sechopoulos, PhD, Atlanta, GA (Abstract Co-Author) Research agreement, Siemens AG; Research agreement, Toshiba Medical Systems Corporation; Speaking agreement, Siemens AG
Ritse M. Mann, MD, PhD, Nijmegen, Netherlands (Abstract Co-Author) Research agreement; Siemens AG; Research agreement, Seno Medical Instruments, Inc
Gerard J. den Heeten, MD, PhD, Nijmegen, Netherlands (Abstract Co-Author) Founder, SigmaScreening BV
Carla H. van Gils, PhD, Utrecht, Netherlands (Abstract Co-Author) Software support, Matakina Technology Limited
PURPOSE

While firm breast compression is generally thought to be required for high quality mammograms the relationship between the amount of compression and screening performance has not been studied systematically. The aim of this study is to determine breast cancer screening outcomes in relation to the compression pressure applied during mammography.

METHOD AND MATERIALS

A consecutive series of 111,870 digital screening examinations performed in 53,684 women between July 2003 and December 2011 was collected from a screening centre operating within a nationwide breast cancer screening program. A total of 662 screen-detected cancers were included in this series, while 280 interval cancers corresponding to the selected exams were identified by linkage to the Dutch Cancer Registry. Using a research version of Volpara Density software (Volpara Solutions, Wellington, NZ) breast volume (V), dense tissue volume (VD), and volumetric density grade (VDG), were estimated for each exam, while compression pressure was estimated for medio-lateral oblique (MLO) view by dividing the compression force by the area of contact surface between the breast and the compression paddle. We calculated frequencies of recalls, screen-detected cancers, and interval cancers stratified by compression pressure in five groups and derived program sensitivity, specificity, and positive predictive value (PPV). In addition, for each group we computed mean values of V, VD, and VDG. For statistical analysis Pearson's Chi-squared test was used. 

RESULTS

Screening outcomes were different in the five compression pressure groups (p=0.004). Program sensitivity decreased with increasing pressure (77.0%, 69.7%, 74.5%, 63.2%, 66.7%) (p=0.02), specificity was similar, and PPV was highest in the midrange of pressure (28.5%, 31.0%, 34.2%, 26.7%, 25.7%) (p=0.03). Cutoff points for pressure dividing the data in groups of 20% were 7.7, 9.2, 10.7, 12.8 kPa. V and VD both decreased with increasing pressure. Mean VDG moderately increased (1.75, 2.0, 2.2, 2.4, 2.8). 

CONCLUSION

Results suggest that if too much pressure is applied during mammography this may increase interval cancer rates and decrease PPV.  

CLINICAL RELEVANCE/APPLICATION

Controlling pressure during mammography is important to decrease the discomfort experienced by women, but it may also be required to optimize screening outcomes.  

RC215-09
Alana A. Lewin, MD, New York, NY (Presenter) Nothing to Disclose
Yiming Gao, MD, New York, NY (Abstract Co-Author) Nothing to Disclose
Leng Leng Young Lin, MD, New York, NY (Abstract Co-Author) Nothing to Disclose
Marissa L. Albert, MD, MSc, New York, NY (Abstract Co-Author) Nothing to Disclose
James S. Babb, PhD, New York, NY (Abstract Co-Author) Nothing to Disclose
Hildegard B. Toth, MD, New York, NY (Abstract Co-Author) Nothing to Disclose
Linda Moy, MD, New York, NY (Abstract Co-Author) Nothing to Disclose
Samantha L. Heller, MD, PhD, New York, NY (Abstract Co-Author) Nothing to Disclose
PURPOSE

Critics of screening mammography argue that the harms of screening include unnecessary recalls and biopsies. The purpose of our study is to evaluate whether false positive biopsy affects subsequent mammographic screening compliance.

METHOD AND MATERIALS

This was an IRB approved, HIPAA compliant retrospective review of women with stereotactic-guided core biopsies performed between 2012-2014. Patient age, clinical history, biopsy pathology, short-term follow-up, and first post-biopsy screening mammogram were reviewed. Statistical analyses were performed using Fisher exact, Mann-Whitney, and Chi-square tests.

RESULTS

913 stereotactic vacuum assisted biopsies (SVAB) performed with a 9 Gauge Suros needle were performed in 2012-2014. Women with malignant or high-risk lesions or biopsies resulting in a recommendation of surgical excision were excluded. 458 SVABs yielded benign pathology in 436 women (average age is 53.7 years, range 24-85 years). Findings were matched with 29,774 patients who had a BI-RADS 1 or 2 screening mammogram in the same time period. 226/458 (49%) women who had a biopsy returned for annual follow-up compared to 20,256/29,774 (68%) women without biopsy who returned for follow-up (p<0.001). 228/458(63%) women who had a biopsy returned for follow-up within 2 years compared to 21,677/29,774 (73%) women without biopsy who returned (p<0.001). Women with a past history of cancer or atypia who had benign SVAB were more likely to return to screening (p=0.027 and p=0.049, respectively). Women who had unilateral short-term follow-up for evaluation of biopsy (30.6% [140/458]) were also more likely to return than women with no such follow-up (p<0.001). There was no association between pathology type or multi-site biopsy and return to subsequent screening mammography.

CONCLUSION

A significantly greater percentage of patients who did not undergo stereotactic-biopsy returned to screen compared with benign biopsy patients, suggesting that benign SVAB may have a negative impact on screening compliance. Biopsied patients with a history of cancer/atypia and those who had a post-biopsy diagnostic unilateral follow-up were more likely to return to screen.

CLINICAL RELEVANCE/APPLICATION

Benign breast biopsies may affect screening compliance. Additional education and discussion may be warranted when discussing future screening recommendations with patients after benign biopsy.

RC215-10

Awards
Student Travel Stipend Award

Krystal C. Buchanan, MD, New Haven, CT (Presenter) Nothing to Disclose
Patricia C. Barrett, New Haven, CT (Abstract Co-Author) Nothing to Disclose
Paul H. Levesque, MD, Madison, CT (Abstract Co-Author) Nothing to Disclose
Jaime L. Geisel, MD, New Haven, CT (Abstract Co-Author) Consultant, QView Medical, Inc Consultant, Siemens AG
Regina J. Hooley, MD, New Haven, CT (Abstract Co-Author) Consultant, FUJIFILM Holdings Corporation; Consultant, Siemens AG
Liane E. Philpotts, MD, New Haven, CT (Abstract Co-Author) Nothing to Disclose
PURPOSE

Breast density notification legislation is being passed in more states every year. Density classification on mammography is primarily achieved by subjective means. With such laws in effect, there is the possibility of radiologists overtly or subconsciously changing density, particularly downgrading such that supplemental tests will not be required. The purpose of this study was to determine the effect of density classification over time and by patient age.

METHOD AND MATERIALS

A search of the electronic breast imaging database (PenRad, MN) was performed to determine the density classifications (BI-RADS categories a,b,c,d) reported on digital screening mammograms over a 10 year period (2006 – 2015). Our state density notification law went into effect in 2009. Prior to 2011 these were FFDM and after 2011, the majority were tomosynthesis exams (all Hologic, MA). The combined data was assessed and additionally, the data were subdivided by patient age by decade: 40-49, 50-59, 60-69, 70 and up. 

RESULTS

A total of 76,924 screening exams were assessed. For all age groups, there was a small decrease in dense breast categories and corresponding increase in non-dense of 5% in 2009, which returned to usual the following year. However, there has been a consistent trend of increasing percentage of heterogeneously dense since that time, from 23% to 34%. When assessed by age, this trend is found mostly in women in the 50’s and 60’s decade. No specific pattern change was noted in 2011 with the conversion to tomosynthesis. 

CONCLUSION

The patterns of density reporting appear to be initially affected by state legislation, yet the pattern did not return to previous rates, but actually shows increase towards more women being reported as dense, particularly women in the 50-69. 

CLINICAL RELEVANCE/APPLICATION

Density reporting appears to be affected by legislation, but such trends may change over time, with increase towards more women being reported as dense. This may be a reflection of radiologists not downgrading density as women age, or leaning towards allowing more women the possibility of supplemental screening.

RC215-11
Jin You Kim, MD, Busan, Korea, Republic Of (Presenter) Nothing to Disclose
Hyun Jung Kang, MD, Busan, Korea, Republic Of (Abstract Co-Author) Nothing to Disclose
Seung Hyun Lee, Busan, Korea, Republic Of (Abstract Co-Author) Nothing to Disclose
Tae Hong Lee, Busan, Korea, Republic Of (Abstract Co-Author) Nothing to Disclose
Suk Kim, MD, Pusan, Korea, Republic Of (Abstract Co-Author) Nothing to Disclose
PURPOSE

To analyze the clinicopathologic and immunohistochemical features of invasive breast cancers visible only on digital breast tomosynthesis (DBT) compared to those visible on both DBT and full-field digital mammography (FFDM).

METHOD AND MATERIALS

The medical records of 205 women with invasive breast cancer who underwent FFDM and DBT prior to surgery between April 2015 and February 2016 were retrospectively reviewed. For women with multifocal or bilateral cancer, the largest tumor was included. To assess the visibility of each lesion, two radiologists first reviewed the FFDM data alone and then the DBT combined with FFDM data; consensus was attained. The clinicopathologic and immunohistochemical features of tumors visible on DBT only and both DBT and FFDM were compared.

RESULTS

Of 205 cancers, 175 (85.4%) were visible on both DBT and FFDM (“both-visible” group). Twenty cancers (9.6%) not visible on FFDM were recognized by DBT as a mass (55.0%), an asymmetric density (35.0%), or an architectural distortion (10.0%) (“DBT-only” group). The remaining 10 tumors (4.9%) were not evident on either DBT or FFDM (“both-occult” group). The mean tumor size of the DBT-only group was significantly smaller than that of the both-visible group (1.53 ± 0.79 vs. 2.35 ± 1.26 cm, P=0.027) but did not differ significantly from that of the both-occult group (1.53 ± 0.79 vs. 1.89 ± 1.09 cm, P=0.310). Tumors of the DBT-only group had more lower histological grade (45.0% vs. 14.9%, P=0.001), estrogen receptor positivity (100.0% vs. 76.0%, P=0.013), progesterone receptor positivity (95.0% vs. 68.6%, P=0.013), human epidermal growth factor receptor-2 negativity (95.0% vs. 71.4%, P=0.023), and lower expression levels of Ki-67 (45.0% vs. 20.6%, P=0.014), compared to the both-visible group. All tumors in the DBT-only group were luminal subtype. Dense breast parenchyma was more common in the DBT-only than the both-visible group (90.0% vs. 64.0%, P=0.019). No DBT-only tumor exhibited calcification as the only mammographic finding.

CONCLUSION

In patients with invasive breast cancer, tumors visible only on DBT were less histologically aggressive than tumors visible on both DBT and FFDM.

CLINICAL RELEVANCE/APPLICATION

The use of digital breast tomosynthesis in addition to conventional mammography increases the detection of less aggressive subtype of invasive breast cancer.

RC215-12
Valentina Iotti, MD, Reggio Emilia, Italy (Presenter) Nothing to Disclose
Cinzia Campari, Reggio Emilia, Italy (Abstract Co-Author) Nothing to Disclose
Andrea Nitrosi, PhD, Reggio Emilia, Italy (Abstract Co-Author) Nothing to Disclose
Rita Vacondio, Reggio Emilia, Italy (Abstract Co-Author) Nothing to Disclose
Paolo Giorgi Rossi, Reggio Emilia, Italy (Abstract Co-Author) Nothing to Disclose
Pierpaolo Pattacini, Reggio Emilia, Italy (Abstract Co-Author) Nothing to Disclose
PURPOSE

To evaluate the impact of screening with digital breast tomosynthesis (DBT) compared with digital mammography (DM) on breast cancer prognosis and mortality in a randomized test and treat study. Will be presented interim data at baseline on detection rate, recall rate, and positive predictive value on recalls.

METHOD AND MATERIALS

Consenting women attending the population-based (45-70 age range) mammography screening of our provincial program and presenting for a new screening round, were invited to participate in the study and randomized to DM (standard of care) or DBT + DM two-view bi-lateral examinations, both with double independent reading. Results of DBT only were recorded separately for analyses, and women in the investigational arm were managed based on the combined evaluation of DBT and DM. Interval between screening rounds is one and two years for women 45-49 and 50-70 age groups, respectively. The planned sample size is 20.000 women per arm (NCT02698202).

RESULTS

From March 2014 to March 2016, 38762 women have been invited. Among the 79% accepting the screening call, 50.6% consented to participate and were enrolled in the study.
Data for 9776 women were available at the interim analysis: 4832 in the DM-arm and 4944 in the DBT-arm. Recall rate was 3.6% and 3.5% with DM and DBT, respectively (RR 0.97; 95%CI:0.79-1.20); detection rate was 5.4/1000 and 7.7/1000 (RR 1.43; 95%CI:0.87-2.35), respectively; positive predictive value was 15.5% and 23.3 (RR 1.50; 95%CI:0.96-2.33). Out of 38 cancers identified in the investigational arm, 8 were detectable only in DBT. Reading time was 37'' vs 60'' in women at first round, (+62%, p<0.05) and 32'' vs 56'' at second round (+75%, p<0.05); the increase was not significant in recalled women: 99'' vs 108'' and 93'' vs 108'' at first and second round, respectively.

CONCLUSION

Initial data from this randomized two-arm study confirmed the results of higher detection rate without increasing recall rate with DBT screening. By August 2016, the second year interim results on about 20,000 women will be available.

CLINICAL RELEVANCE/APPLICATION

In this randomized two-arm study in a screening setting, tomosynthesis confimed a higher detection rate compared to digital mammography, without increasing the recall rate. 

RC215-13
Robert A. Smith, PhD, Atlanta, GA, (robert.smith@cancer.org ) (Presenter) Nothing to Disclose
LEARNING OBJECTIVES

1) Describe current screening guidelines and factors influencing differences. 2) State global evidence for effectiveness of breast cancer screening. 3) Describe opportunities for improved effectiveness of breast cancer screening in the U.S. 

RC215-14

Awards
Student Travel Stipend Award

Christine Chen, MD, New Haven, CT (Presenter) Nothing to Disclose
Melissa A. Durand, MD, New Haven, CT (Abstract Co-Author) Research Grant, Hologic, Inc
Madhavi Raghu, MD, New Haven, CT (Abstract Co-Author) Nothing to Disclose
Regina J. Hooley, MD, New Haven, CT (Abstract Co-Author) Consultant, FUJIFILM Holdings Corporation; Consultant, Siemens AG
Liane E. Philpotts, MD, New Haven, CT (Abstract Co-Author) Nothing to Disclose
PURPOSE

Digital breast tomosynthesis (DBT) has been shown to reduce recall rate and increase cancer detection in screening compared to 2D-mammography during its initial implementation; however, outcome metrics during subsequent rounds of interpretation are uncertain. The purpose of this study was to determine whether the initial improved outcomes can be maintained and to assess trends in these metrics over time.

METHOD AND MATERIALS

A HIPPA-compliant retrospective review of the electronic database (PenRad, MN) at five clinical sites was performed over 4 consecutive years since introduction of DBT at our facility (8/1/2011-7/31/2015). DBT screens interpreted by breast-subspecialized radiologists were identified. Recall rate (RR), cancer detection rate (CDR) and positive predictive value for recall (PPV1) for each one-year study period (DBT1-4) were analyzed and compared with those of 2D digital mammography control from 8/1/2008 to 7/31/2010 (2D-DM). Differences in outcome metrics between consecutive DBT years were also assessed. Percentage of screen-detected in-situ vs invasive cancers for each study period was calculated.

RESULTS

A total of 42,470 screening DBT exams were performed during a 4-year period (vs. 20,553 2D-DM screening exams over 2-year period). Recall rate decreased slightly over 4 years of DBT (DBT1, 7.9%; DBT2, 8.7%; DBT3, 7.9%; DBT4, 7.5%) and remained significantly reduced compared with 2D-DM recall rate of 11.4% (p<.05). A trend toward increasing cancer detection rate per 1000 exams was observed over years 1 to 4 of DBT (DBT1, 5.7; DBT2, 5.1; DBT3, 5.9; DBT4, 6.4 vs. 2D-DM, 3.8), which was significantly different from 2D-DM in the third and fourth DBT years (p<.05). Similarly, there was a trend toward rising PPV1 over 4 years of DBT (DBT1, 7.2%; DBT2, 5.9%; DBT3, 7.4%; DBT4, 8.5% vs. 2D-DM, 3.4%). Significant increase in PPV1 with DBT versus 2D-DM was sustained across years 2-4 of DBT (p<.05). A higher percentage of invasive cancers was detected over time (DBT1, 50%; DBT2, 55%; DBT3, 59%; DBT4, 74% vs. 2D-DM, 59%).

CONCLUSION

Significant RR reduction and increase in CDR and PPV1 with DBT screening is sustainable. Our results showed a trend toward continued improvement in these outcome metrics over time.

CLINICAL RELEVANCE/APPLICATION

The sustainable, superior performance of DBT over 2D-DM illustrates the integral role of DBT in breast cancer screening, which has important implications for policy making in the future. 

RC215-15
Cindy S. Lee, MD, San Francisco, CA (Presenter) Nothing to Disclose
Debapriya Sengupta, MBBS,MPH, Reston, VA (Abstract Co-Author) Nothing to Disclose
Judy Burleson, Reston, VA (Abstract Co-Author) Nothing to Disclose
Mythreyi Bhargavan-Chatfield, PhD, Reston, VA (Abstract Co-Author) Nothing to Disclose
Edward A. Sickles, MD, San Francisco, CA (Abstract Co-Author) Nothing to Disclose
Elizabeth S. Burnside, MD, MPH, Madison, WI (Abstract Co-Author) Nothing to Disclose
Margarita L. Zuley, MD, Pittsburgh, PA (Abstract Co-Author) Research Grant, Hologic, Inc;
PURPOSE

Mammography is the standard imaging examination for breast cancer screening and has substantially reduced mortality from breast cancer. In the last decade, different interpretations of the evidence on outcomes have resulted in various screening guidelines and debate regarding the balance of benefits and risks of mammography screening. There is uncertainty about when to stop screening, as women ≥75 years were not included in randomized trials, limiting available data to mostly small observational studies. This knowledge gap may be informed by new large-scale evidence from the National Mammography Database (NMD), an, up-to-date mammography outcomes database with data representing a large proportion of US states. The purpose of our study is to evaluate the relationship between patient age and screening mammography performance metrics in women age ≥40 years.

METHOD AND MATERIALS

Our HIPAA Compliant and IRB approved project analyzed data from 218 mammography facilities in 39 states in the NMD registry. The NMD receives clinical practice data including self-reported demographics, clinical findings, screening mammography interpretation, and biopsy results (the reference standard). Performance metrics calculated were cancer detection rate, recall rate, and positive predictive values for biopsy recommended (PPV2) and biopsy performed (PPV3).

RESULTS

We analyzed data for 6,980,054 screening mammograms performed between January 2008 and December 2014 in 3,416,075 women. Overall, we found a mean cancer detection rate of 3.65 per 1000 (95% CI: 3.60-3.69), recall rate of 10% (95% CI: 10-10%), PPV2 of 20% (95% CI: 19-20%), and PPV3 of 28% (95% CI: 28-29%). Based on increasing age, performance metrics demonstrate a gradual upward trend for cancer detection rate, PPV2 and PPV3, and downward trend in recall rate, until age 90 years.

CONCLUSION

The NMD provides up-to-date nationwide benchmarks for screening performance metrics. According to these metrics demonstrating preserved cancer detection, recall rate, and PPV, our study suggests that there is no clear age cut-point to inform the decision when to stop screening.

CLINICAL RELEVANCE/APPLICATION

The stability of screening mammography performance metrics in women aged 75-90 years, does not provide evidence for age-based mammography cessation but rather adds support for guidelines that encourage screening decisions based on individual patient values, co-morbidities, and health status.

RC215-16
Constance D. Lehman, MD, PhD, Boston, MA (Presenter) Research Grant, General Electric Company; Medical Advisory Board, General Electric Company
Rob F. Arao, MPH, Seattle, WA (Abstract Co-Author) Nothing to Disclose
Brian L. Sprague, PhD, Burlington, VT (Abstract Co-Author) Nothing to Disclose
Janie M. Lee, MD, Bellevue, WA (Abstract Co-Author) Research Grant, General Electric Company
Diana S. Buist, PhD,MPH, Seattle, WA (Abstract Co-Author) Nothing to Disclose
Diana Miglioretti, PhD, Seattle, WA (Abstract Co-Author) Nothing to Disclose
Louise M. Henderson, Chapel Hill, NC (Abstract Co-Author) Nothing to Disclose
Tracy Onega, PhD,MS, Lebanon, NH (Abstract Co-Author) Nothing to Disclose
Anna N. Tosteson, Lebanon, NH (Abstract Co-Author) Nothing to Disclose
Garth H. Rauscher, PhD, Chicago, IL (Abstract Co-Author) Nothing to Disclose
Karla Kerlikowske, MD, San Francisco, CA (Abstract Co-Author) Nothing to Disclose
PURPOSE

To establish performance benchmarks for modern screening digital mammography and assess performance trends over time in U.S. community practice.

METHOD AND MATERIALS

In this HIPAA compliant IRB approved study we measured performance of digital screening mammography interpreted by 359 radiologists across 95 facilities in six Breast Cancer Surveillance Consortium registries.  The study population included 1,682,504 digital screening mammograms performed between 2007 and 2013 in 792,808 women.  Performance measures were calculated according to the American College of Radiology BI-RADS 5th edition and compared to published prior benchmarks by the BCSC, the National Mammography Database (NMD) and published recommendations for performance by expert opinion.  Benchmarks were derived from the distribution of performance metrics across radiologists and presented as 50th (median), 10th, 25th, 75th and 90th percentiles with graphic presentations using smoothed curves.

RESULTS

Mean performance measures (95% Confidence Interval) were: abnormal interpretation rate (AIR)11.6% (11.5%-11.6%); cancers detected per 1000 screens (CDR) 5.1 (5.0-5.2); sensitivity 86.9% (86.3%-87.6%); specificity 88.9% (88.8%-88.9%); false negative rate per 1000 screens 0.8 (0.7-0.8); PPV-1 4.4% (4.3%-4.5%); PPV-2 25.6% (25.1%-26.1%); PPV-3 28.6% (28.0%-29.3%); 76.9% of screen detected cancers  were stage 0 or 1; 57.7% were minimal cancers; 79.4% were node negative invasive cancers. Recommended CDRs were achieved by 92.1% of radiologists in community practice and 97.1% achieved recommended ranges for sensitivity.  CDR was significantly higher than that reported by the NMD (3.4/1000).  Only 59.0% of radiologists achieved recommended AIR and 63.0% achieved recommended specificity. 

CONCLUSION

The majority of radiologists in the BCSC surpass performance recommendations for screening mammography; however, abnormal interpretation rates continue to be higher and specificity lower than the recommended rates for almost half of radiologists interpreting screening mammograms. 

CLINICAL RELEVANCE/APPLICATION

Efforts to implement advanced technology should be combined with effective educational programs to reduce false-positive rates without sacrificing high detection rates of invasive cancers.

RC215-17
Elizabeth S. Burnside, MD, MPH, Madison, WI (Presenter) Nothing to Disclose
Wendy B. Demartini, MD, Madison, WI (Abstract Co-Author) Nothing to Disclose
Sarina Schrager, MD,MS, Madison, WI (Abstract Co-Author) Nothing to Disclose
Amy Trentham-Dietz, Madison, WI (Abstract Co-Author) Nothing to Disclose
John Hampton, Madison, WI (Abstract Co-Author) Nothing to Disclose
Christina Shafer, PhD, Madison, WI (Abstract Co-Author) Nothing to Disclose
Lee G. Wilke, MD, Madison, WI (Abstract Co-Author) Nothing to Disclose
PURPOSE

Risk-based screening in women < 50 years old has been promoted to increase benefits and decrease harms of a mammography screening program, but has not been evaluated in practice. We compared the impact of risk-based screening to age-based screening in women <50 to determine screening program outcomes.

METHOD AND MATERIALS

We analyzed a database of consecutive screening mammograms (1/1/2006-12/31/2013) from an academic practice—starting at 40 without an upper age limit. To evaluate only “average risk” women, we excluded those with a personal history of breast cancer or with documented BRCA mutation.  We matched our population with a cancer registry as our reference standard. In women <50 we used clinical intake data at the time of each mammogram to estimate breast cancer risk using the BCSC risk calculator (https://tools.bcsc-scc.org/bc5yearrisk/calculator.htm). We emulated a risk-based screening strategy by excluding women <50 whose five-year breast cancer risk was less than the average risk for a 50-year old (≤ 1.253%). We emulated an age-based screening similar to the American Cancer Society guidelines by removing all women <45. We compared outcomes for the two strategies, including number of cancers detected, proportion of DCIS/all cancers), recalls, and biopsies using the chi-square statistical test, defining a p value of <0.05 as statistically significant.

RESULTS

In our actual clinical baseline practice, screening average risk women age ≥40, we performed 75,107 screening mammograms in 25,155 women and detected 344 cancers—230 invasive (183 local and 43 regional) and 91 DCIS, and had 4,816 recalls and 995 biopsy recommendations (Table). Clinical practice audit outcomes were cancer detection rate of 4.58/1000 and recall rate of 8.2%. Age-based screening starting at 45 detected more cancers than risk-based screening (p<0.05), while prompting more recalls (p<0.0001) and biopsies (p<0.01). There was no statistically significant difference in the proportion of DCIS (p=0.99).

CONCLUSION

Risk-based screening in women < 50 results in less recalls and biopsies but also detects fewer cancers than an age-based strategy that starts screening at age 45.

CLINICAL RELEVANCE/APPLICATION

Evaluating the potential effects of risk-based compared to age-based strategies can enumerate trade-offs: fewer cancers detected and false positives without altering the proportion of DCIS.