Introduction

Conventional digital mammography (CDM) is widely used to screen for breast cancer and examine breast lesions [1,2,3,4,5,6,7]. However, CDM is limited by its inability to accurately distinguish suspicious lesions from adjacent overlapping tissue. Specifically, diagnosis becomes more challenging in instances of dense or heterogeneously dense breasts in terms of sensitivity and specificity [8, 9]. In Asian countries, including Japan, women’s breasts are relatively small and highly dense [10, 11].

Digital breast tomosynthesis (DBT) is a three-dimensional imaging technique developed to overcome some of the limitations of CDM. During DBT, an X-ray tube moves through a limited arc angle and rebuilds the tissue in a series of thin slices to minimize the influences of breast tissue overlapping and structural noise [12]. Several studies have shown that DBT is a promising tool for breast cancer screening, as it is associated with decreased screening recall rates, increased cancer detection rates, and positive predictive values [13, 14].

While DBT has been extensively investigated, most previous studies have evaluated its use in screening programs [15, 16], while only a few have investigated its role in examining actual breast lesions. If DBT can resolve diagnostic difficulties of CDM caused by breast tissue overlapping structures and structural noise, then DBT would be useful for both screening and examinations. Therefore, the aim of this study was to determine if the diagnostic performance of breast lesion examinations could be improved by jointly using DBT and CDM for breast lesion examinations.

Materials and methods

Study population

The study protocol was approved by our institutional review board, and the opt-out model (via website) was used. Patients were consecutively recruited between June 2015 and March 2016. Patients who had not undergone previous mammography at our hospital were enrolled in this study. Data were interpreted retrospectively. Patients with suspected malignant findings underwent fine-needle or core biopsy followed by surgery and histopathological examination of the specimens. Patients without suspected malignant findings were followed for at least 12 months to establish the absence of cancer. If patients had a previous mastectomy, breast implants (including in the opposite breast), or previous breast cancer treatment, they were further excluded from analysis because of the effects these procedures can have on breast architecture. Men were also excluded. Ultimately, 1164 breasts (628 patients; age range 22–91 years; mean age 50.2 years) were evaluated.

Image acquisition

Both CDM and DBT images were obtained using a commercially available system (Senographe Essential; GE Healthcare Japan, Tokyo, Japan). The detector used in this system was an amorphous silicon flat-panel detector. The CDM and DBT images were acquired using a tube anode/filter combination (Mo/Mo, Mo/Rh, Rh/Rh) as determined by the automatic exposure control of the unit. Nine projection images were obtained with a total tomosynthesis angle of 25°, acquired in step-and-shoot mode while the breast was compressed in the fixed position. Using this machine, the mean radiation dose for DBT in both breasts in a single view was approximately 1.47 mGy. All patients concurrently underwent DBT in one view (medial lateral oblique) and CDM in two views (craniocaudal and medial lateral oblique). DBT examination was performed immediately after CDM by the same designated technician. CDM and DBT images were obtained in the same compression mode using automatic exposure control.

Image review

DBT images were reconstructed using the successive approximation method and divided into thin 0.5-mm slices and thick 10-mm slabs for viewing on a workstation. Three radiologists, who read at least 1000 CDM studies and 300 DBT studies per year combined, interpreted both the CDM and DBT images on 8 MG monitors. Each radiologist first evaluated the CDM image while blinded to the DBT image and the patient’s clinical information, and assigned a Breast Imaging Reporting and Database System (BI-RADS) category [17]. The radiologist then evaluated the DBT image and assigned a BI-RADS category to that as well. When the assigned BI-RADS categories differed among the radiologists, a consensus was reached through discussion.

BI-RADS categories 1, 2, and 3 were identified as negative, while categories 4 and 5 were identified as positive. The pathological findings from surgery and biopsy were used as the gold standard for the diagnosis of breast cancer. Lesions such as fibroadenoma and breast cysts that did not appear to change on ultrasonography during an observation period of at least 12 months were identified as benign lesions.

Statistical analysis

The overall comparison of clinical performance was derived from the differences between the mean areas under the receiver operating characteristic (ROC) curves. Fisher’s exact test was used to compare the sensitivity, specificity, positive predictive value, and negative predictive value. A p value < 0.05 was considered statistically significant.

Results

In this study, of the 1164 breasts analyzed, 100 breasts were found to have cancer. Biopsy was performed in 226 of 1164 cases. Of the cancers, 74 were invasive ductal carcinomas, 19 were ductal carcinomas in situ, 4 were invasive lobular carcinomas, 1 was a mucinous carcinoma, and 1 was an apocrine carcinoma. Histological diagnosis was unknown in 1 case because the patient was diagnosed with breast cancer via biopsy but switched hospitals afterwards; therefore, her postoperative histological results were not available. Comparison of categories of all breasts and all carcinoma cases is presented in Table 1. The percentage of dense breasts (i.e., extremely dense or heterogeneously dense) was 85.1% (990/1164), and that of scattered or fatty breasts was 14.9% (174/1164).

Table 1 Comparison of BI-RADS categories using CDM alone vs. CDM plus DBT for all cases (a) and all carcinoma cases (b)
Full size table

ROC curves of CDM alone and CDM plus DBT

The areas under the ROC curves were 0.916 (95% confidence interval 0.878–0.954) for CDM alone and 0.938 (95% confidence interval 0.902–0.973) for CDM plus DBT. Using BI-RADS category 4 as a cut-off for diagnosing breast cancer in both CDM alone and CDM plus DBT, breast tumors were diagnosed with sensitivities of 61.0% (61/100) and 83.0% (83/100), specificities of 99.1% (1054/1064) and 98.9% (1052/1064), positive predictive values of 85.9% (61/71) and 87.4% (83/95), and negative predictive values of 96.4% (1054/1093) and 98.4% (1052/1069), respectively (Fig. 1). Sensitivity was significantly higher in CDM plus DBT than in CDM alone (p = 0.0009). No significant differences were observed in the specificity, positive predictive value, or negative predictive value between CDM plus DBT and CDM alone. There were no adverse events.

Fig. 1
figure 1

Receiver operating characteristics curves for a conventional digital mammography alone and b conventional digital mammography plus digital breast tomosynthesis

Full size image

Improvement in diagnostic performance by adding DBT

CDM alone yielded 39 false-negative diagnoses. With the inclusion of DBT, breast cancer was diagnosed in 56.4% of these cases (22/39). Of the 22 lesions, 15 were invasive ductal carcinomas, 5 were ductal carcinoma in situ lesions, 1 was invasive lobular lesion, and 1 was mucinous carcinoma. The descriptions of the findings are shown in Table 2 and Figs. 2, 3 and 4. Regarding breast density, 86.4% (19/22) of the breasts were dense and 13.6% (3/22) were scattered or fatty.

Table 2 Improvement in diagnostic performance by including DBT
Full size table
Fig. 2
figure 2

A 37-year-old woman with invasive ductal carcinoma. a A CDM image shows focal asymmetries in the upper portion of the right breast in the MLO view; and b a DBT image shows an irregular mass with a spiculated margin in the corresponding area of right breast. CDM conventional digital mammography, DBT digital breast tomosynthesis, MLO mediolateral oblique

Full size image
Fig. 3
figure 3

A 54-year-old woman with invasive ductal carcinoma. a No lesion is detected on the CDM image. b DBT shows irregular mass with architectural distortion in the middle portion of the right breast in the MLO view. CDM conventional digital mammography, DBT digital breast tomosynthesis, MLO mediolateral oblique

Full size image
Fig. 4
figure 4

A 68-year-old woman with ductal carcinoma in situ. a A CDM image shows grouped microcalcifications in the lower portion of the left breast in the MLO view. b A DBT image shows microcalcifications arrayed in a line. CDM conventional digital mammography, DBT digital breast tomosynthesis, MLO mediolateral oblique

Full size image

False-negative cases by CDM plus DBT

CDM plus DBT failed to diagnose cancer in 17 cases, of which 7 were invasive ductal carcinomas, 9 were ductal carcinomas in situ, and 1 was an invasive lobular carcinoma. Regarding breast density, 82.4% (14/17) of the breasts were dense and 17.6% (3/17) were scattered or fatty. The pathological types of carcinoma are listed in Table 3. Eight cases were miscategorized as BI-RADS category 3 by both CDM alone and CDM plus DBT: 7 as calcifications and 1 as a mass. These lesions were suspected of being malignant by ultrasonography after CDM, follow-up CDM, or ultrasonography, and were subsequently verified as such.

Table 3 Pathological type of false-negative cases by CDM + DBT (n = 17)
Full size table

False-positive cases

CDM plus DBT yielded false-positive diagnoses in 12 cases. In 2 of these cases, the CDM findings were considered negative, although the DBT images showed indistinct masses and were considered positive. Ultimately, these lesions were deemed to be benign on follow-up ultrasonography or fine-needle aspiration. As for the other 10 cases that were diagnosed as false-positive on both CDM alone and CDM plus DBT, calcification observed in 9 cases was diagnosed as mastopathy or benign lesions upon follow-up ultrasonography or biopsy. One case with an indistinct mass was proven to be a benign phyllodes tumor following surgery.

Discussion

This study demonstrated an improvement in breast cancer diagnostic performance with the combined use of CDM and DBT. When CDM and DBT were combined, the false-negative rate decreased and sensitivity increased as compared to using only CDM (61.0–83.0%). However, the area under the ROC curve for CDM alone was already high (0.916), and was therefore not significantly different from that for CDM plus DBT (0.938). DBT is particularly useful for detecting breast cancer with mass formation because DBT shows the tumor contour more precisely and increases the tumor’s contrast in comparison to normal mammary tissue [18]. Since DBT provides a definite depiction of a lesion, the method should be able to distinguish between benign and malignant lesions. In the current study, DBT contributed to the discovery of 22 carcinomas, suggesting DBT is a helpful tool for discriminating between tumors and normal mammary gland tissue in dense breasts [19]. Consequently, we expect that adding DBT is beneficial for Japanese women who have dense breasts or a symptom of breast mass. Furthermore, including DBT facilitates the detection of focal asymmetries and lesions with architectural distortions, which are often observed in breast cancer.

The ability of DBT to assess calcification has not been established [20,21,22]. In the current study, DBT was not inferior to CDM for the detection of calcified lesions. In addition, DBT exhibited a clearer spatial distribution of calcifications in the mammary ducts as compared to CDM (Fig. 4). Therefore, this study showed that DBT might also be useful for the assessment of calcifications in some cases. These advantages contribute to a reduction in unnecessary biopsies and subsequently alleviate patients’ mental and economic burdens. Further, these advantages improve the confidence of the radiologist’s diagnosis. Previous studies have indicated that DBT reduces the number of category 3 cases, which comprise the majority of cases with focal asymmetries [22]. In contrast, in the present study, adding DBT increased the number of cases with category 3 (Table 1), including 58 mass lesions with a clear margin, suggesting benign lesions. Finally, these were diagnosed as cysts (38), fibroadenomas or mastopathies (18), and intraductal papillomas (2) by ultrasonography and biopsy. This is considered a reason for the increase of category 3 cases.

This study had several limitations. First, since our hospital is not a screening facility, but a facility that conducts scrutiny and treatment, most patients that were included in this study presented with some symptoms or abnormalities in screening mammography. Further, only the patients who underwent mammography for the first time at our hospital were selected. Thus, some degree of selection bias could not be avoided. Finally, the number of patients was relatively small in comparison to other screening studies [23, 24].

Conclusion

This study demonstrated that the combination of DBT and CDM for the diagnosis of breast cancer in women with abnormal screening findings or clinical symptoms proved effective and should be used to improve the diagnostic performance of breast cancer examination, and investigate patients with abnormal findings or those who present with clinical symptoms.