The added value of digital breast tomosynthesis in improving diagnostic performance of BI-RADS categorization of mammographically indeterminate breast lesions

Breast cancer is the most common cancer among women in the world, accounting for about 12% of all new cancers and 27% of all female cancers [1]. Early detection becomes a critical job to reduce the morbidity and mortality associated with breast cancer [2]. Digital mammography (DM) is the primary breast imaging modality for early detection and diagnosis of breast cancer. However, some limitations persist [3]. One of the substantial limitations of DM is its use in dense breasts [4]. DM has low sensitivity and specificity in women with radiographically dense breast due to decrease contrast between a possible tumor and surrounding breast tissue and summation of tissues may obscure lesions [5]. Digital breast tomosynthesis (DBT) can be expected to overcome this problem by reducing or eliminating the tissue overlap. DBT technology is a modification of a DM unit to allow the acquisition of a three-dimensional (3D) volume of thin section data [4]. The role of DBT for ruling out suspected abnormalities that are identified during screening may be considered an essential diagnostic application [6]. It also allows visualization of cancers not apparent by DM [7]. The more explicit depiction with DBT should allow easier differentiation between benign and malignant lesions [4].

Breast Imaging Reporting and Data System (BI-RADS) was initially developed to allow radiologists to report their level of concern that breast lesions may be missed on DM due to dense tissue but has been widely used in breast cancer research and DM performance research [8, 9].

Several previous studies have shown the high benefits of the addition of DBT in screening programs and the diagnostic setting [10‐12]. Mammographic findings are seen more clearly in tomographic images with the consequent improvement of BI-RADS categorization. This fact is reflected, among other things, by the upgrade on BI-RADS classification of malignant lesions not correctly assessed by DM [13], in a better diagnostic performance in dense breasts with BI-RADS 0 findings [14] and in the demonstration of indeterminate lesions (BI-RADS 3 and 4) that are characterized on DM [15]. Consequently, we performed this prospective study to evaluate the added value of DBT to BI-RADS classification in categorization of indeterminate breast lesions (BI-RADS 0, 3, and 4) after DM as an initial approach. Additionally, we made a simple comparison between DBT and DM to test their diagnostic performance in this context.

In this study, we mainly focused on indeterminate BI-RADS categories, because reducing BI-RADS 0, 3, and 4 has pivotal implications for patient care. We aimed to display the added value of DBT to the BI-RADS classification. Few studies evaluated this topic; most of them have investigated category 3 [18, 19], and some have also included category 0 [20]. However, this study focused exclusively on BI-RADS 0, 3, and 4 categories. The overall results of our study confirmed the high diagnostic performance of DBT in the evaluation of indeterminate BI-RADS categories. We found that combined DBT and DM in BI-RADS resulted in a superior sensitivity (98.7%), specificity (96.6%), and accuracy (97.5%) for indeterminate breast lesion categorization than DM or DBT alone; the sensitivity, specificity, and accuracy declined to 66.9%, 67.6%, and 67.3%, respectively, for the DM assessment and 89.2%, 90.3%, and 90%, respectively, for the DBT assessment. Moreover, we found that DBT had a significantly higher sensitivity, specificity, and accuracy than DM in the diagnosis of indeterminate breast lesions. The above findings are matching with that of many previous studies [10, 12, 19, 21‐25], which have established that DBT increases the sensitivity and specificity of DM. Consequently, in light of our data, and considering the high diagnostic performance of DBT, we recommend the use of DBT as an additional imaging modality to improve diagnostic accuracy in detecting and characterizing indeterminate breast lesions.

We found that DBT produced a significant change of BI-RADS category in 66.7% lesions with an upgrade in 31.5% lesions (83% were malignant) and a downgrade in 35.2% lesions (94% were benign) in comparison to the DM. This finding agrees with the recent study published by Raghu et al. [12] who have proved that the addition of DBT has been found to change rates of BI-RADS final assessment over time. Similarly, Michell et al. [26] showed a reduction of probably benign cases in 57.8% by an additional DBT.

A remarkable observation of our study was the higher number of lesions identified with DBT than with DM [37 (10.4%)], with BI-RADS 2 lesion representing the greatest number of these missed lesions on DM [19 (51.4%)]. We found that the main cause for missing a lesion on DM was poor visibility due to dense breast parenchyma, tissue overlap, and a radiographically non-conspicuous lesion. In contrast, the DBT decreased interference from overlapping breast tissue and improved lesion conspicuity. These missed lesions on DM cause a significant upgrade of the BI-RADS categories between DM and DBT and subsequently increased diagnostic performance of DBT over DM. This finding indicates that DBT is more accurate than DM in the identification of breast lesion, which is comparable to the previous findings [10, 15, 19]. The increased number of lesions detected on DBT is most probably due to the use of rebuilt images in DBT, as stated by Andersson et al. [13]. These images are obtained from different angles from the breast in a short scanning process and allow the assessment of breast parenchyma where lesions may go unnoticed or less evident due to tissue overlap or increased breast density.

The reduction in the number of BI-RADS 3 and 4 lesions is one of the potential advantages of DBT as some lesions that were categorized as BI-RADS 3 and 4 on DM was upgraded to BI-RADS 5 or downgraded to BI-RADS 1 and 2 based on DBT. This increase in the identification of BI-RADS 3 and 4 lesions by DBT likely result in reduced follow-up of lesions that would not have been identified by DM alone and diminished the requirement for biopsy. These results are comparable to those reported by previous studies [13, 27, 28].

Although DBT has better diagnostic performance than DM, still some breast lesions could not be determined on DBT. On DBT images, improved visualization of a partially or totally smooth boundary in some malignant masses may potentially lead to a misdiagnosis that is false benign diagnosis [29]. Thirty-six of the breast lesions in this study were misdiagnosed on DBT (20 false-positives and 16 false-negatives). Sixteen masses were described as probably benign masses on DBT, but histopathology revealed breast cancer. However, the misdiagnosed lesions on DBT were less than that on DM (67 false-positives and 49 false-negatives). The combined DM and DBT decreased misdiagnosed lesions (seven false-positives and two false-negatives) when compared to DM or DBT alone. Accordingly, our study recommends the use of BI-RADS with a combined DM and DBT protocol, as it improved the BI-RADS performance for diagnosis of indeterminate breast lesions with subsequent potentially better disease management. Similar findings have been seen in various other studies [30‐32], in which the addition of DBT decrease in the number of false cases.

When comparing the ROC areas, it was found that DBT is significantly superior to DM in breast cancer diagnosis (AUC = 0.883 vs 0.619; p < 0.0001), and the BI-RADS with combined DM and DBT was significantly superior to BI-RADS with DM or BI-RADS with DBT alone (AUC: 0.971; p < 0.0001). Similarly, Cai et al. [33] analyzed 79 cases with pathologic results by using ROC curve and showed that the AUC of the combined DM and DBT was greater than that of DM alone (0.914 vs. 0.805). Also, Thomassin et al. [34] calculated AUC by averaging the ROC from four readers; the mean AUC for BI-RADS with combined DM and DBT was higher than that calculated for BI-RADS with DM alone (0.809 vs. 0.685; p < 0.01). In contrast, Gennaro et al. [35] concluded that the overall clinical performance with DBT and DM for malignant versus all other cases was not significantly different (AUCs = 0.851 vs. 0.836, p = 0.645).

Finally, our study demonstrates that BI-RADS could diagnose the same patient with breast lesions by DBT but not by DM. Thus, we should use DBT in BI-RADS categorization of breast lesions as if we used DM alone as decisive for breast cancer diagnosis; a significant number of breast cancer lesions would be missed by the BI-RADS. The suggested approach for the evaluation of breast cancer lesions according to our results is as follows: first, we perform the DM as a primary imaging modality; if the BI-RAD category provides a confident diagnosis of being benign or malignant (i.e., BI-RADS 1, 2, or 5), the patients go on to the management without further additional imaging. In cases of indeterminate diagnosis (i.e., BI-RADS 0, 3, or 4), we perform DBT, and the BI-RADS category is determined by the combination of DM and DBT findings. We perform the DM initially and not DBT as the former’s lower cost and wide availability.

There are some limitations to our study. First, we focused on indeterminate breast lesions (BI-RADS 0, 3, and 4) and did not consider other BI-RADS categories (BI-RADS 1, 2, and 5). Second, no trial was conducted to analyze the inter-reader agreement in the classification of breast lesions. Third, we did not address the DBT performance in each breast density category. Fourth, the cost-effectiveness and the added radiation dose of combined DM and DBT protocol may be disadvantages of this protocol. Thus, we suggest that the combined protocol be limited to doubted lesions when there remains uncertainty in the BI-RADS category after conducting DM alone.

Adding DBT to BI-RADS classification improves its diagnostic performance in the detection and characterization of lesions categorized as BI-RADS 0, 3, and 4 by DM.