Background: The aim of this study was to evaluate the added value of digital breast tomosynthesis (DBT) when combined with digital mammography (DM) in BI-RADS assessment and follow-up management. Methods: From February 2014 to January 2015, 214 patients underwent DM and DBT, acquired with a Siemens Mammomat Inspiration unit. 2 expert readers independently reviewed the studies in 2 steps: DM and DM+DBT, according to BI-RADS rate. Patients with BI-RADS 0, 3, 4, and 5 were recalled for work-up. Inter-reader agreement for BI-RADS rate and work-up rate were evaluated using Cohen's kappa. Results: Inter-reader agreement (κ value) for BI-RADS classification was 0.58 for DM and 0.8 for DM+DBT. DM+DBT increased the number of BI-RADS 1, 2, 4, 5 and reduced the number of BI-RADS 0 and 3 for both readers compared to DM alone. Regarding work-up rate agreement, κ was poor for DM and substantial (0.7) for DM+DBT. DM+DBT also reduced the work-up rate for both Reader 1 and Reader 2. Conclusion: DM+DBT increased the number of negative and benign cases (BI-RADS 1 and 2) and suspicious and malignant cases (BI-RADS 4 and 5), while it reduced the number of BI-RADS 0 and 3. DM+DBT also improved inter-reader agreement and reduced the overall recall for additional imaging or short-interval follow-up.
Breast cancer is the most common female invasive cancer in Western countries. It represents almost 25% of female cancers, affecting 1 in 10 women. The incidence is ever increasing, mainly due to 3 factors: actual rise in disease rate (30%), increase in the number of early diagnoses due to the use of mammography (60%), and increase in life expectancy with a higher probability of cancer developing (10%). In recent years, in Italy, the breast cancer mortality of under 49-year-old women decreased by almost 11%. This reduction demonstrates the relevance of early diagnosis through breast imaging in order to improve treatment strategies and outcomes [1,2].
The limits of digital mammography (DM) are well known and are partly related to the 2-dimensional (2D) format of conventional imaging of the 3-dimensional (3D) breast volume. Since some breast cancers have the same X-ray attenuation as normal breast tissue, clinically relevant malignancies may be obscured by overlapping normal tissue. In contrast, complex areas of normal tissue may be perceived as suspicious. The limitations of 2D breast imaging lead to low sensitivity in detecting certain cancers and high false-positive recall rates. Although mammographic screening has been shown to reduce breast cancer deaths by approximately 30%, controversy exists over when and how often screening mammography should occur , .
Digital breast tomosynthesis (DBT) is an emerging diagnostic technique which provides a 3D reconstruction of the breast, minimizing the effect of overlapping breast tissue and improving the detection of any lesions , . At present, DBT is a promising technique, but it is mainly used in women with a positive screening mammography or clinical symptoms. Large-scale studies are still needed to confirm its role in screening and to introduce this tool into daily practice , .
The purpose of our study is to assess the advantages of DBT in the assignment of BI-RADS categories  and in the management and follow-up of patients compared to mammography alone. Also, an inter-reader agreement study between 2 readers was conducted in order to assess whether the use of DBT could make readings more homogeneous, thus improving the management of patients in clinical practice.
Patients and Methods
The study was approved by the ethics committee of our institution. Objectives and methodology were clearly explained to the patients, and all patients provided written informed consent. Each recruited woman could withdraw from the study at any time. The study was performed independently of any type of industry support (provision of equipment, financial support), and the authors had full control of all data and information submitted for publication.
From February 2014 to January 2015, we enrolled 214 patients aged 36-85 years (mean 51.47 years) from approximately 5,332 women presenting for spontaneous screening mammography.
The study population included symptomatic women, women with dense breasts (American College of Radiology (ACR) BI-RADS c, d ), and women who had previously undergone breast surgery for malignant disease. Each patient underwent clinical breast examination, DM, and DBT; the medical history was also evaluated. We decided to exclude from our study asymptomatic patients with predominantly fatty breasts (ACR BI-RADS a, b ), patients who carry mutations in BRCA 1/2 due to the increased radiosensitivity , , and those with a possible pregnancy.
Patients with suspicious findings underwent fine needle or core biopsy followed by surgery and histopathological examination of the specimens. Patients without suspicious findings were followed for at least 1 year ,  to establish absence of cancer.
Participants underwent DM imaging of both breasts in the mediolateral oblique (MLO) and craniocaudal (CC) position, while DBT imaging was acquired in bilateral MLO projections for patients with dense breast, and in MLO and CC projections on operated breasts only for patients with former breast cancer or symptomatic women.
DBT examinations were performed immediately after the DM examinations, both on the dedicated full-field DM system Siemens MAMMOMAT Inspiration (Siemens AG Healthcare, Erlangen, Germany). This unit has an a-Se detector with 85 µm pitch and a selenium layer thickness of 200 µm. The A/F combination was W/Rh and the tube voltage range was 26-32 kV as determined by the automatic exposure control device of the DM unit. The nominal focal spot dimension was 0.3 mm. The source-to-image receptor distance (SID) was 655.5 mm, while the source-to-breast support table distance (STD) was 637 mm. The system used a linear grid with a ratio of 5:1, lead septa, and strip density of 31 lines/cm . The average dose to the glandular tissues within the breast (average glandular dose, AGD) for a single DBT view was on average factor 1.5 higher than those for a single 2D mammogram. The significance of this dose increase must be assessed in relation to the clinical performance and future role of DBT systems.
Siemens' breast tomosynthesis unit acquires 25 images during a continuous scan, while the X-ray source rotates along a predefined arc of ± 25° relative to the 0° position around the compressed breast, with a 2° angle increment per image. Images are reconstructed using filtered back projection in order to provide sections parallel to the breast support.
Two breast imaging specialists, with respectively 15 and 10 years' experience in breast imaging, blindly interpreted the DM and DBT images using a dedicated workstation on 2 Nio 5-megapixel MDNG-6121 Barco monitors (Barco, Kortrijk, Belgium). DBT reconstructions were reviewed in cine or manual scroll modes on the dedicated workstation.
Each reader evaluated the images in 2 steps: DM first (at this time, DBT examinations were not available to the radiologist to avoid clinical decision making being affected), and after a period of 2-4 weeks the combined evaluation of DM and DBT views. At the end of 1 reading step, each radiologist gave a probability of breast malignancy using the BI-RADS classification (0-5)  and determined the need for further diagnostic investigations (score of 0 no work-up, 1 work-up). Patients with BI-RADS 0, 3, 4, and 5 were recalled for work-up (table 1).
We calculated the percentage of single-reader BI-RADS-assigned categories and the need for further investigation values as determined after DM image reading and after combined DM/DBT view reading. Inter-reader agreement of BI-RADS rate and work-up rate were evaluated using Cohen's kappa in order to verify the reproducibility of the techniques.
Cohen's kappa is a statistic coefficient that measures inter-reader agreement for qualitative (categorical) items. It is generally thought to be a more robust measure than simple percentage agreement calculation, since κ takes into account the agreement occurring by chance. It is calculated from the observed and expected frequencies on the diagonal of a square contingency table. Kappa can range from 0 to +1, where 0 represents the amount of agreement that can be expected from random chance, and 1 represents perfect agreement between the readers.
Of 214 patients, 37 were proven to have cancer, whereas after at least 1 year of follow-up, 177 cases were established to be benign or healthy. Of the 37 cancer patients, 25 had a newly diagnosed breast cancer and 12 had breast cancer recurrence. 21 patients had a single lesion, 7 had multifocal disease, and 9 had multicentric disease, with 59 overall breast lesions. The 59 malignant breast lesions detected in the histologic evaluation were classified as invasive ductal carcinoma (n = 38), ductal carcinoma in situ (n = 7), invasive lobular carcinoma (n = 7), lobular carcinoma in situ (n = 2), mucinous carcinoma (n = 2), and papillary carcinoma (n = 3).
The combined reading of DM with DBT compared to DM alone significantly increased the number of BI-RADS 1 and 2 for both Reader 1 (p < 0.001) and Reader 2 (p < 0.001). The DM+DBT combination also increased the number of BI-RADS 4 and 5 for Reader 1 (p < 0.001) and Reader 2 (p < 0.001), while it reduced the number of BI-RADS 0 and 3 for Reader 1 (p < 0.001) and Reader 2 (p < 0.001) (table 2). The request for further investigations was significantly (p < 0.001) reduced with DM+DBT versus DM alone for both Reader 1 and Reader 2 (table 2).
With regard to inter-reader agreement, the BI-RADS classification kappa value was 0.58 for DM (moderate agreement) and 0.8 for DM+DBT (substantial agreement) (table 3). Call-back inter-reader agreement proved to be slight for DM (κ = 0.1) and substantial for DM+DBT (κ = 0.7) (table 3).
DBT involves 3D imaging via mathematically reconstructed sections of breast volume with high spatial resolution; 3D cross-sectional images improve the visibility of breast lesions by reducing superimposition of breast structures, with a resultant increase in cancer detection and reduction in false-positive findings .
Evidence from large-scale studies in the US and Europe show that DBT+DM, compared to DM alone, yields higher invasive cancer detection rates, increasing the effectiveness of breast cancer screening [16,17,18,19]. Several studies also show that DBT reduces recall rates, thus reducing both costs and distress caused by false-negative recall [20,21,22].
Methods proposed for DBT imaging include the addition of 1-view tomosynthesis to DM, the addition of 2-view tomosynthesis to DM , or a mixture of CC DM and MLO tomosynthesis. Most studies performed 1-view tomosynthesis in MLO acquisition because it generally enables the inclusion of more breast tissue than the CC position, while Rafferty et al. , in a recent study, demonstrated a significant gain in diagnostic accuracy with 2-view tomosynthesis, even with an incremental radiation dose.
However, the optimal method for incorporating tomosynthesis into the mammographic workflow is still not defined; hence, we decided to acquire bilateral MLO views for patients with dense breasts, and MLO plus CC views for patients with previous breast cancer surgery and symptomatic patients. The aim of our study was to evaluate the diagnostic potential of DBT compared with DM in clinical practice. In particular, we assessed how DBT used with DM can influence the assignment of BI-RADS categories and the request for further investigations compared to DM alone, possibly reducing non-diagnostic or doubtful cases, such as BI-RADS 0 and 3, and therefore alleviating the need for additional diagnostic tests (ultrasound, magnetic resonance imaging, biopsy) and minimizing patients' anxiety associated with work-up. Moreover, test reproducibility and homogeneity of results are among the most important features in order to determine the value of a technique; hence, the second purpose of our study was to assess the inter-reader agreement for DM versus DM+DBT.
We particularly focused on BI-RADS category assignment, because reducing BI-RADS 0 and 3 has pivotal implications for patient care. Our findings showed that DBT+DM significantly reduced the number of cases in these 2 categories (from 58.1 to 8.1% for both readers) whilst increasing the number of cases in BI-RADS categories 1 and 2 (from 26.3 to 63.4% for both readers). This leads to lower work-up rates and fewer supplementary breast ultrasounds and unnecessary biopsies. The few studies pertaining to BI-RADS classification currently in the literature provided almost comparable results to our study [27,].
More cases were also classified BI-RADS category 4 and 5 with DBT+DM (about 13% more than with DM alone), which means higher cancer detection rates, consistent with the results of prior studies [29,]. There was also a significant reduction in false-positive and false-negative cases as also previously reported in the literature , . The work-up rate was significantly reduced with DBT+DM versus DM alone (33-42%), similar to previous studies in analogue settings (38.5% for 1-view tomosynthesis ).
The inter-reader agreement was substantial for DBT+DM both with regard to BI-RADS assessment and request for further investigation, and was shown to be higher than for DM alone. Our study showed higher agreement and more homogeneous reports, enforcing the reliability of tomosynthesis. To our knowledge, there are no previous studies that investigated this.
The benefits of DBT, according to the 2 readers, were better evaluation of lesion conspicuity due to reduced structure noise from normal breast tissue, margin feature analysis, and detection of additional findings. In the case of a suspicious lesion created by overlapping breast tissue, tomosynthesis was able differentiate normal tissue superimposition from real lesions (fig. 1 and 2) , . Concerning calcifications, there are discordant opinions in the literature , . Although calcifications are seen with equal or greater clarity on DBT images once they have been detected , a cluster of calcification may not be easily perceived with DBT: every single plane can reveal 1 or 2 calcification; consequently, the reader may not recognize the ‘cluster'. However, DBT was useful in identifying a mass under the calcifications (fig. 2).
Interpretation time was longer for DM+DBT than for DM alone, as also noted by others , . However, optimization of interpretation efficiency by means of improved hanging protocols or work-flow is beyond the scope of this investigation.
Some limitations of this study may have biased our results. We enrolled only symptomatic women, women with dense breasts, and women who had previously undergone surgery for breast cancer; therefore, our results may not be representative of the general population. In fact, we scheduled patients with difficult-to-read mammographies due to ‘anatomical noise' from overlapping glandular tissue or parenchymal distortion caused by surgery. Moreover, we included 214 patients, and, even if we obtained statistically significant results, large-scale studies are needed to confirm them.
Despite these limitations, our study demonstrates that tomosynthesis can reduce the ambiguity of the final diagnosis and possibly increase the rate of both true-negative and true-positive cases, confirming the results of previous studies , , . This leads to lower work-up rates and fewer supplementary breast ultrasounds and unnecessary biopsies, reducing the relative emotional, financial, and clinical costs , . Moreover, adding DBT to DM significantly increased inter-reader agreement, thus improving the performance of radiologists in routine clinical practice.
The authors declare that they have no conflict of interest related to the publication of this article.