Original Research
Women's Imaging
April 1, 2019

Can Digital Breast Tomosynthesis Replace Full-Field Digital Mammography? A Multireader, Multicase Study of Wide-Angle Tomosynthesis

Abstract

OBJECTIVE. The purpose of this study was to test the hypothesis whether two-view wide-angle digital breast tomosynthesis (DBT) can replace full-field digital mammography (FFDM) for breast cancer detection.
SUBJECTS AND METHODS. In a multireader multicase study, bilateral two-view FFDM and bilateral two-view wide-angle DBT images were independently viewed for breast cancer detection in two reading sessions separated by more than 1 month. From a pool of 764 patients undergoing screening and diagnostic mammography, 330 patient-cases were selected. The endpoints were the mean ROC AUC for the reader per breast (breast level), ROC AUC per patient (subject level), noncancer recall rates, sensitivity, and specificity.
RESULTS. Twenty-nine of 31 readers performed better with DBT than FFDM regardless of breast density. There was a statistically significant improvement in readers' mean diagnostic accuracy with DBT. The subject-level AUC increased from 0.765 (standard error [SE], 0.027) for FFDM to 0.835 (SE, 0.027) for DBT (p = 0.002). Breast-level AUC increased from 0.818 (SE, 0.019) for FFDM to 0.861 (SE, 0.019) for DBT (p = 0.011). The noncancer recall rate per patient was reduced by 19% with DBT (p < 0.001). Masses and architectural distortions were detected more with DBT (p < 0.001); calcifications trended lower (p = 0.136). Accuracy for detection of invasive cancers was significantly greater with DBT (p < 0.001).
CONCLUSION. Reader performance in breast cancer detection is significantly higher with wide-angle two-view DBT independent of FFDM, verifying the robustness of DBT as a sole view. However, results of perception studies in the vision sciences support the inclusion of an overview image.
The value digital breast tomosyn-thesis (DBT) adds to full-field digital mammography (FFDM) has been shown in multiple studies [17] and corroborated by the diffusion of DBT technology into clinical practice in the United States and Europe [8, 9]. The advent of DBT has improved both the sensitivity and the specificity of mammography, a remarkable feat for a test that is acknowledged for its sensitivity at the sacrifice of “unnecessary and anxiety-provoking” callbacks from screening mammography [10, 11].
Mammographic screening has reduced mortality from breast cancer, a true change in course since epidemiologists first began tracking this disease after World War II [12, 13]. Earlier detection of breast cancer relative to clinical presentation is the hallmark of mammography. Unfortunately, some breast malignancies are not visible with 2D FFDM. Therefore, research studies and anecdotal accounts of breast cancer detection with DBT when FFDM findings are negative are important [2, 1416]. Whether these additional detections of breast cancer further reduce breast cancer mortality will need to be determined.
One concern raised by radiologists is the additional time required to read the many images of DBT versus the single 2D image per view of FFDM [17, 18]. The best way to integrate DBT into clinical practice, maximize cancer detection, and keep dose as low as possible must be determined. The current work flow and approach to mammographic reading, whether FFDM alone, FFDM plus DBT, or DBT alone, and the technical design of DBT systems are being explored. Most clinical studies reported to date have been conducted with narrow-angle DBT systems. The population-based Malmö Breast Tomosynthesis Screening Trial [19], however, showed a fairlylarge increase in cancer detection rate with the use of one-view-only (mediolateral oblique) wide-angle DBT versus two-view (mediolateral oblique and craniocaudal) FFDM—8.7 per 1000 screens for one-view DBT versus 6.5 per 1000 screens for two-view FFDM (p < 0.0001)—but with a minimal increase in recall rate from 2.6% to 3.6%.
The objective of this study was to evaluate whether two-view DBT, without 2D synthesized views or FFDM, in which images are acquired with a wide-angle DBT system is superior to two-view FFDM only. We report the results of a multireader multicase study comparing the diagnostic accuracy of readers in a mixed population without symptoms and an enriched population with malignancies. We did not measure the effect on reading time.

Subjects and Methods

Institutional review board approval was obtained at seven clinical sites in the United States to recruit women for DBT with the overall goal of obtaining premarket approval by the U.S. Food and Drug Administration for a wide-angle DBT unit [20]. All subjects provided written informed consent to undergo a bilateral mediolateral oblique and craniocaudal DBT examination in addition to screening or clinically indicated diagnostic FFDM, including patients undergoing uni-lateral imaging for symptoms or screening recalls. The inclusion criteria were age 40 years or older if without symptoms and undergoing screening or 30 years or older if undergoing a diagnostic mammographic workup. Exclusion criteria were pregnancy; mastectomy, because bilateral DBT was required; lumpectomy within 5 years before study entry; and recently diagnosed breast cancer, because readers were not to be provided with information on clinical history or symptoms.
Subjects were followed for ground-truth documentation by the following: pathology reports to determine benign or malignant core and surgical biopsy results; 1 year of negative FFDM follow-up for BI-RADS categories 1 and 2; 2 years of stable follow-up findings for BI-RADS category 3; and 6 or 12 months of follow-up to confirm benign core biopsy results.
A multireader multicase study was conducted to evaluate the diagnostic accuracy of readers using the DBT images, without synthesized composite views, as the sole imaging modality (replacement paradigm) relative to their accuracy with FFDM.
The radiologists for this reader study were from sites other than those that acquired the cases. The radiologists were Mammography Quality Standards Act qualified from academic hospitals with variable experiences in reading DBT images (see later, Reader experience). These readers were not involved in the recruitment of subjects and had no knowledge of subject identity, history, or other diagnostic information. The trial was compliant with HIPAA and was approved by the institutional review board at each clinical site.

Imaging System

DBT images were obtained with a Mammomat Inspiration system (Siemens Healthcare) which acquires 25 projections over an angle of 50° in 25 seconds [21] with a tungsten-rhodium anode-filter combination. The tube voltages were automatically selected as a function of compressed breast thickness in the same manner as for 2D FFDM on the same system; exposure of one tomosynthesis image was set to 1.5–2 times that of a single 2D FFDM image. The tomosynthesis slices, with 1-mm slice separation, were reconstructed with a filtered back projection algorithm [22]. The 2D FFDM images were acquired either with the Mammomat Inspiration system in 2D mode or with various other commercial FFDM systems, with subsequent DBT. Only a minority of images were obtained in one compression for both FFDM and DBT. During the multireader multicase study, FFDM images and the tomosynthesis slices were read at a Syngo Mammoreport workstation (Siemens Healthcare).

Study Population and Case Selection

Between May 2011 and April 2014, 764 women were recruited to undergo wide-angle DBT. Because sufficient power was achieved with a smaller cohort, the study statistician selected 330 cases using a stratified random sample: 210 (64%) patients without cancer, 105 (32%) patients with pathologically proven malignancy, and 15 (4%) patients with unknown truth at the time of selection (see later).
The nonmalignant cases were collected from cases of benign biopsy results (n = 72) among cases assessed BI-RADS category 0 at screening and determined to be negative by the previously described ground-truth algorithm (n = 32) and from randomly selected negative cases (n = 106).
Pathologically proven malignant cases comprised 105 (32%) cases in the 330-case multireader multi-case set. To ensure an adequate range of types and subtlety of malignant lesions, we included all cases of findings of architectural distortion or asymmetric density and all cases of one or more sites of suspicious microcalcifications. We then took a random sample of patients with only mass-type lesions.
Last, we sampled 15 cases with unknown truth status at multireader multicase case selection to minimize verification bias. Delayed follow-up of the special cases showed 26 breasts to have negative findings, one breast with a benign biopsy result, and three breasts remaining unverified.
Cases were selected to reflect the range of breast density encountered in clinical practice. The densities of the tissues (fatty, scattered fibroglandular densities, heterogeneously dense, extremely dense) between noncancer and malignant cases were similar (Fig. 1). Thus, the reader case set (n = 330) included 105 malignancies, 72 biopsy-proven benign lesions, 138 negative cases, and 15 special cases in which follow-up findings were not available at the time of the case selection. The mean age of the subjects was 56.3 ± 9.8 (SD) years. The race distribution was as follows: 73.6% (n = 243) white, 11.2% (n = 37) African American, 4.2% (n = 14) Hispanic/Latino, 1% (n = 3) Asian, less than 1% (n = 1) American Indian, and 9.7% (n = 32) other.
Fig. 1A —Distribution of breast densities.
A, Chart shows distribution in noncancer cases (n = 224). Percentages do not total 100 owing to rounding.
Fig. 1B —Distribution of breast densities.
B, Chart shows distribution in cancer cases (n = 104).
Readers were shown both breasts in all cases and asked to score each breast for the presence of all findings suspicious for breast cancer. At the time of analysis, however, truth in some of the breasts, as previously defined, was unknown. Therefore, analyses were conducted on 218 bilateral and 112 unilateral cases, yielding 548 verified breasts: 110 breasts with malignant lesions, 85 with biopsy-proven benign lesions, and 353 with negative findings. Two patients were excluded from final analyses because erroneous data were entered in one cancer case, and an erroneous non-cancer case, not selected by the statistician, was shown to the readers, resulting in 104 subjects with malignancy and 224 without cancer.
In the 104 selected cancer cases, there were 134 mammographic findings of cancer, 80.6% invasive (n = 108) and 19.4% (n = 26) ductal carcinoma in situ. Most were masses and calcifications (Table 1). The others were architectural distortions and asymmetric densities. Lesion sizes, which were determined in pathologic analysis, were distributed similarly over four quartiles (Table 1): < 10 mm, 10–19 mm, 20–29 mm, and ≥ 30 mm, which best simulated clinical practice and degrees of subtlety of detection.
TABLE 1: Characteristics of the Cancer Lesions in 104 Patients: Mammographic Findings and Sizes
CharacteristicValue
Total no. of cancers134
Lesion type 
 Mass85 (63.4)
 Calcification29 (21.6)
 Architectural distortion12 (9.0)
 Asymmetric density8 (6.0)
Lesion size (mm) 
 < 1029 (21.6)
 10–1940 (29.9)
 20–2933 (24.6)
 ≥ 3027 (20.1)
 Missing size data5 (3.7)

Note—Values in parentheses are percentages.

Multireader Multicase Study

Reader experience—All 31 readers were Mammography Quality Standards Act–qualified radiologists with practice experience ranging from 4 to 38 years (mean, 14.2 years). Twenty-six (84%) had more than 5 years of experience with FFDM. The radiologists' experience with DBT was markedly less: 13 had no experience; 10, 1–6 months of experience; three, 6–12 months; two, 1–2 years; and three, more than 2 years. There was a mix of breast specialists and general radiologists; only 19 (61%) were breast imaging fellowship trained.
Reader training—Before reading the study cases, all readers took part in a 10-hour training session, combined 4-hour lecture, 1.5 hours at the workstation to learn how to enter the data, and 4.5 hours of case review. The training cases consisted of on-site guided review in 2D FFDM and 3D DBT case interpretation (n = 10), approach to reading DBT only (n = 10), independent review of DBT only (n = 25), and review of case examples (n = 25). This provided the radiologists with an overview of DBT physics, examples of lesion features in wide-angle DBT, interpretation approaches to DBT studies, and instruction on how to use the image review workstation. Training cases separate from the test set were used to demonstrate how interpretation data should be recorded on the electronic case report form. Training case data were not used for the reader study.
Reader tasks—The 31 readers were asked to mark the locations of all suspicious lesions on FFDM and DBT images. All cases were bilateral mediolateral oblique and craniocaudal images. BI-RADS category 0, 1, or 2 was assigned to each breast. For each detected lesion, readers assigned two types of scores: a probability of malignancy (POM) confidence score on a scale of 1–100 and a BI-RADS assessment based on lesion appearance without the benefit of additional diagnostic imaging and the assumption that the lesion was the only finding (forced BI-RADS 3, 4, or 5).
Reader sessions—Each reader participated in two reading sessions separated by 4 weeks. Cases were randomized to which modality would be interpreted at the first versus second session. During each reading session, one-half of the cases were interpreted with FFDM and the other half with DBT. After the 4 weeks, the readers interpreted the opposite modality for each case. All cases were read without prior images for comparison, clinical information, or other diagnostic imaging.

Statistical Methods

The electronic data capture system, image database, and reader interpretation database were designed and maintained by the independent data and image management company BioClinica. The data were transferred to an independent statistician for analysis.
The endpoints were lesion detection per patient (subject level) and per breast (breast level), and ROC AUCs were averaged over the readers. The null hypothesis was that the mean breast-level AUC for FFDM is equal to the mean breast-level AUC for DBT versus the alternative hypothesis that the mean AUCs are not equivalent. The ROC curves were constructed from the POM scores. For breasts with cancer, the POM score assigned to a correctly located lesion was used (true-positive); otherwise, a score of 0 was assigned (false-negative). For breasts without cancer, the highest POM score assigned to the breast was used (false-positive); otherwise, a score of 0 was assigned (true-negative). Nonparametric methods [23] were used to estimate the readers' mean AUC. The method of Obuchowski and Rockette [24] adjusted to handle the clustered data [25] was used to test the null hypothesis with the adjustment to the degrees of freedom by Hillis et al. [26]. A two-tailed test with a significance level of 0.05 was used. The sample size and the number of readers were selected to provide at least 80% power to detect a difference in AUCs of 0.04 or larger. The data analysis was performed with SAS/STAT software (version 9, SAS Institute). ROC analyses were performed with OBUMRM (version SAS 9.4, Cleveland Clinic Lerner Research Institute) and CORROC2 (version SAS 9.4, University of Chicago) software.
After the primary study analysis, a hierarchic analysis plan was followed (Fig. 2). If superiority of the DBT AUC was found, then a noninferiority test of the subject-level recall rate was planned with a noninferiority margin of 20% of the FFDM recall rate. If noninferiority in recall rate was found, superiority tests of the DBT recall rate and DBT subject-level sensitivity were planned. To control type I error at 0.05 for these latter comparisons, Holm correction [27] was applied.
Fig. 2 —Chart shows hierarchic analysis plan. If superiority of digital breast tomosynthesis (DBT) over full-field digital mammography (FFDM) was found, recall rates were compared in test of noninferiority of DBT. If noninferiority was found, superiority of DBT recall rate and sensitivity was tested.
To assess the robustness of the analyses with respect to verification bias, multiple imputation was performed at the breast level. Because reader confidence scores were obtained for all breasts of all study patients, the truth status was imputed only for breasts for which truth verification was not available. Available information about the subject was used, and the following assumptions were made: for cancer patients, we assumed that the probability of cancer in the contralateral breast was 0.02 [28]; for a breast assessed BI-RADS category 3 at follow-up and followed for less than 2 years, we assumed that the probability of cancer was 0.02 [29]; for a breast assessed BI-RADS 4 at follow-up, we assumed that the probability of cancer was 0.20; and for breasts with no follow-up information, we assumed that the probability of cancer was 0.005. We imputed 10 datasets (m = 10) [30, 31]. The overall estimate of the effect size was the mean of the 10 individual estimates of the effect size. The overall total variance was calculated as the between-imputation variance times (1 + [1 / m]) plus the within-imputation variance [32].

Results

Using DBT, 29 of 31 readers improved their performance over that with FFDM in terms of AUC (Fig. 3). The subject-level parametric ROC AUC increased from 0.765 (standard error [SE], 0.027) for FFDM to 0.835 (SE, 0.027) for DBT (p = 0.002; 95% CI, 0.026–0.115) (Fig. 4).
Fig. 3 —Chart shows estimated per-breast AUC for 31 readers using full-field digital mammography (FFDM) (blue) and improvement using digital breast tomosynthesis (DBT) alone (red). Order of readers is from lowest to highest AUC using FFDM.
Fig. 4 —Graph shows summary parametric subject-level ROC curves. AUC for full-field digital mammography (FFDM) (blue) is 0.765; AUC for digital breast tomosynthesis (DBT) (red) is 0.835.
Subject-level (per patient) performance (Table 2) showed the readers' mean sensitivity with bilateral two-view DBT, with BI-RADS category 3, 4, or 5 as a positive result, was 0.819 (SE, 0.030) compared with the readers' mean sensitivity with FFDM of 0.774 (SE, 0.032) (p = 0.122; 95% CI for improvement, −0.012 to 0.102). For 28 of 31 readers, the recall rate with DBT was lower than their recall rate with FFDM. The mean recall rate for noncancers decreased from 0.479 (SE, 0.027) with FFDM to 0.386 (SE, 0.025) with DBT, a reduction of 0.093, or 19%, which was statistically significant (p < 0.001; 95% CI, 0.049–0.136 [10.1–28.5%]).
TABLE 2: ROC AUCs, Recall Rates, Sensitivity, and Specificity of Two-View Digital Breast Tomosynthesis (DBT) Compared With Two-View Full-Field Digital Mammography (FFDM)
ResultFFDMDBTΔp
Breast-level nonparametric AUC0.818 (0.019)0.861 (0.019)0.0430.011a
Subject-level parametric AUC0.765 (0.027)0.835 (0.027)0.0700.002a
Subject-level noncancer recall rate0.479 (0.027)0.386 (0.025)−0.093< 0.001a
Subject-level sensitivity BI-RADS 3, 4, 5b0.774 (0.032)0.819 (0.030)0.0450.122
Breast-level sensitivity BI-RADS 4, 5c0.715 (0.034)0.772 (0.032)0.0570.047a
Breast-level specificity BI-RADS 4, 5c0.807 (0.017)0.848 (0.015)0.041< 0.001a
Breast-level sensitivity BI-RADS 3, 4, 5b0.762 (0.031)0.807 (0.030)0.0450.109
Breast-level specificity BI-RADS 3, 4, 5b0.718 (0.019)0.773 (0.017)0.055< 0.001a

Note—Values in parentheses are standard error.

a
Statistically significant at 0.05 level.
b
Sensitivity and specificity counting BI-RADS categories 3, 4, and 5 as positive test result.
c
Sensitivity and specificity counting BI-RADS categories 4 and 5 as positive test result.
Breast-level (per breast) performance also improved with the use of two-view DBT alone versus FFDM (Table 2). The readers' mean breast-level ROC AUC 0.861 (SE, 0.019) using DBT was significantly greater than their mean breast-level ROC AUC 0.818 (SE, 0.019) using FFDM (p = 0.011). The 95% CI for the mean improvement with DBT was 0.010–0.076. The sensitivity and specificity of lesion detection for BI-RADS 4 and 5 cases or BI-RADS 3, 4, and 5 cases counted as positive per breast are shown in Table 2. The sensitivity of DBT improved, but the difference was not statistically significant. The specificity improved significantly (p < 0.001).
The effects on performance with FFDM and DBT according to breast density and detection of specific mammographic findings are shown in Table 3. Lesion detection of not only dense tissues but also fatty tissue improved, although not statistically significantly, with DBT compared with FFDM. Therefore, all women regardless of breast density could receive benefit from DBT. Detection of both masses and architectural distortion increases significantly with DBT, in spite of the small number of cases of architectural distortion (n = 12). Although calcifications were less apparent on DBT than on FFDM images, the difference was not statistically significant.
TABLE 3: Subanalyses by Breast Density, Mammographic Findings, and Invasive Cancers
FindingTwo-View FFDMTwo-View DBTΔp
Dense breasts0.802 (0.027)0.844 (0.026)0.0430.106
Nondense breasts0.826 (0.026)0.873 (0.026)0.0470.024a
Masses0.858 (0.018)0.923 (0.018)0.065< 0.001a
Calcifications0.796 (0.042)0.749 (0.041)−0.0470.136
Architectural distortion0.773 (0.038)0.930 (0.035)0.157< 0.001a
Invasive cancers0.836 (0.018)0.912 (0.018)0.076< 0.001a

Note—Breast-level AUC stratified for dense (BI-RADS c or d) and nondense (BI-RADS a or b) breasts and for masses and calcifications. Values in parentheses are standard error. FFDM = full-field digital mammography, DBT = digital breast tomosynthesis.

a
Statistically significant at 0.05 level.
Most important, the detection of cases of invasive carcinoma was independently significantly greater with two-view DBT than with FFDM, in spite of readers' minimal experience with DBT relative to FFDM. After multiple imputation was performed for breasts not subjected to ground-truth verification, the study results were unchanged. The primary analysis showed an estimated mean AUC improvement of 0.038 with DBT (p = 0.021; 95% CI for improvement, 0.006–0.071). The recall rate was significantly reduced (19%) with DBT (p < 0.001). The improvement in subject-level sensitivity was 0.045 (p = 0.139).

Discussion

The objective of this study was to evaluate whether two-view DBT with a wide-angle DBT system is superior to two-view FFDM. This reader study tested the replacement paradigm, that is, breast cancer detection performance of DBT alone without seeing a 2D FFDM or synthetic 2D image.
Bilateral images and images obtained in two views per breast with each modality were presented to 31 readers who had little experience with DBT in two reading sessions separated by at least 1 month. The results showed that the performance of two-view wide-angle DBT alone over FFDM in both subject-level and breast-level analyses improved significantly (subject level, 0.835 vs 0.765, p = 0.002; breast-level, 0.861 vs 0.818, p = 0.011). In the evaluation of data per reader, cancer detection by 29 of 31 readers improved when they used DBT. Of the two readers who did not have a higher AUC with DBT, one had incorrectly marked the locations of the lesions in one of the reading sessions, yet this reader was still counted in the final analyses (reader T, Fig. 3). The other radiologist had the highest performance with FFDM compared with the others, suggesting that this reader was already working at an exceptional level (reader E′, Fig. 3). Improved diagnostic accuracy with DBT was also measured by the 19% reduction in callback rate of patients without cancer, thereby increasing specificity. The most important finding is the improvement in detection of invasive cancers with DBT, likely affected by greater mass and architectural distortion conspicuity and despite the trend toward poorer visibility of calcifications.
Our results with wide-angle DBT are concordant with prior reports of reader- and population-based studies—even though the methods (DBT alone, one vs two views, adjunctive to FFDM) and the hardware and software construction of the DBT units may have differed [17]—and have been corroborated in clinical practice with units sold in the United States and Europe [8, 9].
One difference between DBT systems is the angular span over which the x-ray source images the breast. To date, commercial DBT systems [21] have an angular range between 15° (narrow) and 50° (wide). Improvements in AUC with wide-angle DBT in a multi-reader multicase study [20] were greater than those reported with a narrow-angle DBT system in a similar cohort [6]. Tomosynthesis with sampling from a wide angular range acquires more information on the object than does narrow-angle sampling. Reconstruction theory indicates that there is not only superior depth resolution, and thereby potentially greater discrimination of lesions, but also better contrast of subtle masses with wider angular range [33, 34]. A disadvantage of wide-angle acquisition may be patient motion due to longer imaging duration.
Our study showed superior detection of masses and distortion with DBT alone compared with FFDM and a statistically nonsignificant trend toward poorer performance in the detection of calcifications. Similar results have been reported in the literature [35, 36]. Phantom experiments have shown that use of a wide angle may improve mass detection [36], whereas narrow-angle DBT may have higher sensitivity and afford greater conspicuity for subtle groups of microcalcifications [35] than wide-angle DBT does. On the other hand, in the multireader study by Rafferty et al. [6], improvement in detection of calcifications was not statistically significant for adjunct narrow-angle DBT compared with FFDM. Clauser et al. [37] found noninferior detection and characterization of microcalcifications with a wide-angle DBT system compared with FFDM. The capability of DBT to show subtle calcifications may depend on other factors, such as the reconstruction algorithm, acquisition technique, and detector noise. Evaluation of wide- versus narrow-angle DBT remains the subject of current research [38].
Our methods in this study speak to the robustness of the data showing greater diagnostic accuracy of wide-angle DBT because it was tested independently from FFDM. Our results add to the body of knowledge on improved performance of DBT for detecting breast cancers and reducing recall rates (false-positive results). Our reader study is in line with the strong guidelines evoked by Ko-pans [39] for evaluating the contributions of two modalities independently. The data show that DBT alone is superior to FFDM in both sensitivity and specificity and detection of invasive cancer, notwithstanding the trend in poorer calcification detection. Therefore, the replacement paradigm supports the hypothesis that two-view DBT can be used as the sole interpretation reading display.
The theoretic question of how mammographic images should be displayed is fascinating from a perceptual standpoint. Using only DBT slices would be a marked change in the work flow of most radiologists. We know from eye-tracking studies on the perception of lesion detection that there are two fundamental patterns, scanning and drilling [40]. Scanning is used in evaluating a 2D single image. The eye quickly takes in a rapid, unconscious overview of the entire breast, a holistic approach, from a distance set back from the monitor followed by a more focused look at a smaller area closer to the monitor. This viewing is analogous to facial recognition in which one rapidly determines the identity of a person without having to analyze each structure of the face. Drilling, in contrast, is used for a stack of images, as in DBT. The eye stays focused on an area smaller than the entire breast as the computer quickly pages through the stack. At least one study [41] has shown that mammographers use a combination of scanning and drilling when interpreting DBT images.
A body of literature exists on the visual perception called the gist [42], the ability to rapidly extract information from a scene on first glimpse on the basis of a signal that is not always correct but is also not random. Expert radiologists can differentiate normal from abnormal mammograms at above-chance levels when viewing a 2D FFDM image for as little as one-fourth of a second. The signal is a global one, and radiologists can detect it but cannot localize it under the same rapid viewing conditions [43]. The theory is that visual search passes through two stages [44, 45]. The first is an initial global processing component in which perception may direct or filter the subsequent search. A selective pathway recognizes objects, and rapid, global extraction arises from a nonselective pathway. The theories of visual perception support an observer's need for an image that allows global overview, particularly when symmetry between the left and right sides affect the observer's perception of abnormalities. Fortunately, the development of synthesized FFDM images, formed by the computerized summing of all of the exposures obtained for DBT, provides that holistic image. These composite FFDM views are generated without an additional radiation exposure. Both patient and radiologist benefit.
There were limitations to this study, which was a reader study with an enriched population of positive cases. Symptoms and clinical history were not provided. Results may have been affected by the two separate compression events between FFDM and DBT, which would change the appearance of the breast. This study was not powered to show statistical significance for small differences in findings, such as calcifications in only 29 cases. The study was powered for the primary objective, which was breast-level ROC, to detect a difference in AUC of 0.04 with a power of 80% or greater (134 total cancers). We acknowledge that no prior comparison mammograms were provided. Nevertheless, because it was a detection study, prior images do not indicate the presence or absence of a finding on a current image. A mass, calcifications, architectural distortions, or focal asymmetries are present or not present for the reader to see. The effect of prior images is on the observer's ability to detect change in the perception of developing asymme-tries and in assessing probability of malignancy based on stability. An additional point is that synthetic composite views, a simulated FFDM, were not available and therefore not used in the DBT-only algorithm. It is unknown how the synthetic 2D views would have affected the results, and this topic warrants future study. The effects of different reading algorithms on reading time and efficiency were speculated as reasons to conduct our study but were not directly tested. Further investigation of reading time efficiency paradigms is warranted, particularly in an era of increased pressures on radiologists to read more studies within a limited number of hours. Regarding the generalization of our results to other DBT units, particularly those with a narrower sweep angle, those outcomes have yet to be determined. Studies are currently being conducted [46].
In summary, this reader study showed that wide-angle two-view DBT alone has greater diagnostic accuracy than FFDM for most radiologists, even for those inexperienced with DBT technology. The results of this study strongly support the robustness of DBT as a sole view. Nevertheless, it is likely that an overview image will be useful in the clinical work flow because other factors related to viewing and perception affect one's ability to detect breast cancer.

Footnote

N. A. Obuchowski reports that her institution has a contract with Siemens Healthcare. J. Y. Lo has a licensing agreement with Gammex Inc. J. Y. Lo and J. A. Baker have research grants from Siemens Healthcare. R. F. Brem was a consultant for Siemens Healthcare during development of the study design and throughout the acquisition and reader studies. P. R. Fisher is coinvestigator of two current research grants from Siemens Healthcare. W. Zhao has a research grant from and has a speaker arrangement with Siemens Healthcare. L. L. Fajardo receives research support from Hologic, Inc, and is a consultant for FujiFilm. T. Mertelmeier is an employee of Siemens Healthcare GmbH and is a stockholder of Siemens AG and Siemens Healthineers AG.

References

1.
Poplack SP, Tosteson TD, Kogel CA, Nagy HM. Digital breast tomosynthesis: initial experience in 98 women with abnormal digital screening mammography. AJR 2007; 189:616–623
2.
Andersson I, Ikeda DM, Zackrisson S, et al. Breast tomosynthesis and digital mammography: a comparison of breast cancer visibility and BIRADS classification in a population of cancers with subtle mammographic findings. Eur Radiol 2008; 18:2817–2825
3.
Skaane P, Bandos AI, Gullien R, et al. Comparison of digital mammography alone and digital mammography plus tomosynthesis in a population-based screening program. Radiology 2013; 267:47–56
4.
Ciatto S, Houssami N, Bernardi D, et al. Integration of 3D digital mammography with tomosyn-thesis for population breast-cancer screening (STORM): a prospective comparison study. Lancet Oncol 2013; 14:583–589
5.
Rose SL, Tidwell AL, Bujnoch LJ, Kushwaha AC, Nordmann AS, Sexton R Jr. Implementation of breast tomosynthesis in a routine screening practice: an observational study. AJR 2013; 200:1401–1408
6.
Rafferty EA, Park JM, Philpotts LE, et al. Assessing radiologist performance using combined digital mammography and breast tomosynthesis compared with digital mammography alone: results of a multicenter, multireader trial. Radiology 2013; 266:104–113
7.
Giess CS, Pourjabbar S, Ip IK, Lacson R, Alper E, Khorasani R. Comparing diagnostic performance of digital breast tomosynthesis and full-field digital mammography in a hybrid screening environment. AJR 2017; 209:929–934
8.
Hardesty LA, Kreidler SM, Glueck DH. Digital breast tomosynthesis utilization in the United States: a survey of physician members of the Society of Breast Imaging. J Am Coll Radiol 2014; 11:594–599
9.
Gao Y, Babb JS, Toth HK, Moy L, Heller SL. Digital breast tomosynthesis practice patterns following 2011 FDA approval. Acad Radiol 2017; 24:947–953
10.
Myers ER, Moorman P, Gierisch JM, et al. Benefits and harms of breast cancer screening: a systematic review. JAMA 2015; 314:1615–1634
11.
Segel JE, Balkrishnan R, Hirth RA. The effect of false-positive mammograms on antidepressant and anxiolytic initiation. Med Care 2017; 55:752–758
12.
Howlader N, Noone AM, Krapcho M, et al. SEER cancer statistics review, 1975–2014, National Cancer Institute website. seer.cancer.gov/csr/1975_2014/. Published April 2017. Accessed January 15, 2019
13.
American Cancer Society. Breast cancer facts & figures 2017–2018. Atlanta, GA: American Cancer Society, 2017
14.
Taskin F, Durum Y, Soyder A, Unsal A. Review and management of breast lesions detected with breast tomosynthesis but not visible on mammography and ultrasonography. Acta Radiol 2017; 58:1442–1447
15.
Lång K, Andersson I, Zackrisson S. Breast cancer detection in digital breast tomosynthesis and digital mammography—a side-by-side review of discrepant cases. Br J Radiol 2014; 87:20140080
16.
Chan HP, Helvie MA, Hadjiiski L, et al. Characterization of breast masses in digital breast tomosynthesis and digital mammograms: an observer performance study. Acad Radiol 2017; 24:1372–1379
17.
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; 17:450–455
18.
Benedikt RA, Boatsman JE, Swann CA, Kirkpatrick AD, Toledano AY. Concurrent computer-aided detection improves reading time of digital breast tomosynthesis and maintains interpretation performance in a multireader multicase study. AJR 2018; 210:685–694
19.
Zackrisson S., Lång K, Rosso A, et al. One-view breast tomosynthesis versus two-view mammography in the Malmö Breast Tomosynthesis Screening Trial (MBTST): a prospective, population-based, diagnostic accuracy study. Lancet Oncol 2018; 19: 1493–1503
20.
U.S. Food and Drug Administration website. Pre-market Approval (PMA). Mammomat Inspiration with tomosynthesis option. www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfpma/pmacfm?id=320583. Published 2015. Accessed February 8, 2016
21.
Sechopoulos I. A review of breast tomosynthesis. Part 1. The image acquisition process. Med Phys 2013; 40:014301
22.
Mertelmeier T, Orman J, Haerer W, Kumar MK. Optimizing filtered backprojection reconstruction for a breast tomosynthesis prototype device. Proc SPIE 2006; 6142:61420F1
23.
DeLong ER, DeLong DM, Clarke-Pearson DL. Comparing the areas under two or more correlated receiver operating characteristic curves: a nonpara-metric approach. Biometrics 1988; 44:837–845
24.
Obuchowski NA, Rockette HE. Hypothesis testing of the diagnostic accuracy for multiple diagnostic tests: an ANOVA approach with dependent observations. Commun Stat Simul Comput 1995; 24:285–308
25.
Obuchowski NA. Nonparametric analysis of clustered ROC curve data. Biometrics 1997; 53:567–578
26.
Hillis SL, Obuchowski NA, Schartz KM, Berbaum KS. A comparison of the Dorfman-Berbaum-Metz and Obuchowski-Rockette methods for receiver operating characteristic (ROC) data. Stat Med 2005; 24:1579–1607
27.
Holm S. A simple sequentially rejective multiple test procedure. Scand J Stat 1979; 6:65–70
28.
Donovan AJ. Bilateral breast cancer. Surg Clin North Am 1990; 70:1141–1149
29.
Sickles EA. Periodic mammographic follow-up of probably benign lesions: results in 3,184 consecutive cases. Radiology 1991; 179:463–468
30.
Pisano ED, Gatsonis C, Hendrick E, et al; Digital Mammographic Imaging Screening Trial (DMIST) Investigators Group. Diagnostic performance of digital versus film mammography for breast-cancer screening. N Engl J Med 2005; 353:1773–1783
31.
Rosenberg RD, Yankaskas BC, Abraham LA, et al. Performance benchmarks for screening mammography. Radiology 2006; 241:55–66
32.
Rubin DB. Multiple imputation for nonresponse in surveys. New York, NY: Wiley, 1987
33.
Zhao B, Zhou J, Hu YH, Mertelmeier T, Ludwig J, Zhao W. Experimental validation of a three-dimensional linear system model for breast tomosynthesis. Med Phys 2009; 36:240–251
34.
Samei E, Thompson J, Richard S, Bowsher J. A case for wide-angle breast tomosynthesis. Acad Radiol 2015; 22:860–869
35.
Goodsitt MM, Chan HP, Schmitz A, et al. Digital breast tomosynthesis: studies of the effects of acquisition geometry on contrast-to-noise ratio and observer preference of low-contrast objects in breast phantom images. Phys Med Biol 2014; 59:5883–5902
36.
Chan HP, Goodsitt MM, Helvie MA, et al. Digital breast tomosynthesis: observer performance of clustered microcalcification detection on breast phantom images acquired with an experimental system using variable scan angles, angular increments, and number of projection views. Radiology 2014; 273:675–685
37.
Clauser P, Nagl G, Helbich TH, et al. Diagnostic performance of digital breast tomosynthesis with a wide scan angle compared to full-field digital mammography for the detection and characterization of microcalcifications. Eur J Radiol 2016; 85:2161–2168
38.
Glick S. System design and acquisitions parameters for breast tomosynthesis. In: Reiser I, Glick S, eds. Tomosynthesis imaging. Boca Raton, FL: Taylor and Francis, 2014
39.
Kopans DB. Recent issues in breast cancer detection and the premarket approval by the Food and Drug Administration of a US system for breast lesion evaluation: what happened to science? Radiology 1997; 202:315–316
40.
Drew T, Vo ML, Olwal A, Jacobson F, Seltzer SE, Wolfe JM. Scanners and drillers: characterizing expert visual search through volumetric images. J Vis 2013; 13:3
41.
Aizenman A, Drew T, Ehinger KA, Georgian-Smith D, Wolfe JM. Comparing search patterns in digital breast tomosynthesis and full-field digital mammography: an eye tracking study. J Med Imaging (Bellingham) 2017; 4:045501
42.
Gandomkar Z, Ekpo EU, Lewis SJ, et al. Detection of the abnormal gist in the prior mammograms even with no overt sign of breast cancer. Proc SPIE 2018; 10718:4
43.
Evans KK, Georgian-Smith D, Tambouret R, Bird-well RL, Wolfe JM. The gist of the abnormal: above-chance medical decision making in the blink of an eye. Psychon Bull Rev 2013; 20:1170–1175
44.
Kundel HL, Nodine CF. A visual concept shapes image perception. Radiology 1983; 146:363–368
45.
Al Mousa DS, Brennan PC, Ryan EA, Lee WB, Tan J, Mello-Thoms C. How mammographic breast density affects radiologists' visual search patterns. Acad Radiol 2014; 21:1386–1393
46.
Scaduto DA, Huang H, Liu C, et al. Impact of angular range of digital breast tomosynthesis on mass detection in dense breasts. Proc SPIE 2018; 10718:1V

Information & Authors

Information

Published In

American Journal of Roentgenology
Pages: 1393 - 1399
PubMed: 30933648

History

Submitted: June 24, 2018
Accepted: December 14, 2018
Version of record online: April 1, 2019

Keywords

  1. full-field digital mammography
  2. multireader multicase study
  3. vision sciences-perception
  4. wide-angle tomosynthesis

Authors

Affiliations

Dianne Georgian-Smith
Department of Radiology, Brigham and Women's Hospital, 75 Francis St, Boston, MA 02115.
Nancy A. Obuchowski
Quantitative Health Science, Cleveland Clinic Foundation, Cleveland, OH.
Joseph Y. Lo
Department of Radiology, Duke University School of Medicine, Durham, NC.
Rachel F. Brem
The George Washington Cancer Center, George Washington University, Washington, DC.
Jay A. Baker
Department of Radiology, Duke University School of Medicine, Durham, NC.
Paul R. Fisher
Departments of Radiology and Surgery, State University of New York at Stony Brook Health Science Center, Stony Brook, NY.
Alice Rim
Imaging Institute, Cleveland Clinic Foundation, Cleveland, OH.
Wei Zhao
Department of Radiology, Stony Brook Medicine, Stony Brook, NY.
Laurie L. Fajardo
Department of Radiology and Imaging Sciences, University of Utah School of Medicine, Salt Lake City, UT.
Thomas Mertelmeier
Diagnostic Imaging, X-Ray Products, Siemens Healthcare GmbH, Forchheim, Germany.

Notes

Address correspondence to D. Georgian-Smith ([email protected]).

Funding Information

Supported by a grant from Siemens Healthcare

Metrics & Citations

Metrics

Citations

Export Citations

To download the citation to this article, select your reference manager software.

Articles citing this article

View Options

View options

PDF

View PDF

PDF Download

Download PDF

Media

Figures

Other

Tables

Share

Share

Copy the content Link

Share on social media