Institutional review board approval was obtained for this retrospective study, and the requirement for written informed patient consent was waived. At Bergonié Institute, approximately 6000 DBT examinations are performed per year. The local PACS was searched to identify consecutively registered patients who underwent DBT with a spot compression view for the characterization of an indeterminate or suspicious finding on DBT performed at the institute from December 14, 2018, through December 18, 2019. The search was not constructed to identify patients with suspicious microcalcifications given that the study aim was to evaluate DBT spot compression imaging and not to evaluate magnification imaging, which is not feasible on DBT. Initially identified patients were excluded for the following reasons: presence of a postbiopsy marker clip in the area of the lesion, no clear abnormality on retrospective review of DBT images for the lesion for which the DBT spot compression view was obtained, DBT and DBT spot compression view not performed in the same projection (i.e., craniocaudal [CC] or mediolateral oblique [MLO]), or 1-year follow-up not available for a lesion not evaluated by percutaneous biopsy. Patient age, personal history of ipsilateral or contralateral breast cancer, synchronous ipsilateral or contralateral breast cancer, presence of clinical findings such as a palpable mass in the area of concern, and the clinical indication for performing DBT were recorded.

All mammograms were acquired with a commercial DBT system (Selenia Dimensions, Hologic). At Bergonié Institute during the study period, DBT was routinely performed in two projections (CC and MLO) for screening and for workup before cancer treatment and in one or two projections (CC, MLO, or both) for diagnosis. A DBT spot compression view was obtained concurrently with DBT at the radiologist's discretion for further assessment of equivocal DBT findings on CC or MLO images. The projection of the DBT spot compression image (CC or MLO) matched the projection of the standard DBT image depicting the equivocal finding. Two-dimensional synthetic mammograms were reconstructed from both DBT and DBT spot compression images by use of commercial software (C-View, Hologic). The mean glandular dose (MGD) of radiation for the DBT spot compression view and for ipsilateral DBT views was recorded.

Two fellowship-trained breast radiologists (reader 1 [F.C.], 10 years of posttraining experience in breast imaging and 5 in DBT; reader 2 [M.B.], 30 and 10 years of posttraining experience) and a fellow in breast imaging (reader 3 [P.A.L.], 1 year of experience in both breast imaging and DBT) reviewed all examinations independently. An additional investigator serving as a study coordinator (F.D., a 5th-year radiology resident), who had knowledge of patients' clinical data, displayed the images to the readers and indicated the location of the lesion to be evaluated at a dedicated workstation (SecureView, Hologic). Readers did not have access to digital mammography when performed in the given patient, nor were readers informed of the findings of other breast imaging examinations (e.g., ultrasound or MRI) or of the pathology findings. Readers were informed whether the patient had a personal history of conservative surgery on the breast ipsilateral to the lesion.

The coordinator first displayed the available DBT images, and the readers recorded their evaluation of the lesion. Immediately after the readers completed the lesion assessment on standard DBT images, the coordinator additionally displayed the DBT spot compression image, and the readers repeated the evaluation. The coordinator conducted a separate review session with each of the three readers. To evaluate intrareader agreement for one observer, reader 1 repeated the readings 4 months after the initial session. The cases were displayed in a different order in this second session, to reduce recall bias.

Readers assigned each lesion to one of six BI-RADS categories according to the BI-RADS atlas, 5th edition [18]: category 2 (benign), category 3 (probably benign), category 4A (low suspicion), category 4B (moderate suspicion), category 4C (high suspicion), and category 5 (highly suggestive of malignancy). In addition, to facilitate ROC analysis, readers evaluated the level of suspicion of cancer using a 5-point Likert scale (1, not suspicious; 5, highly suspicious), as previously reported [11]. Readers also estimated, using a 5-point Likert scale (1, no confidence; 5, confident), their level of confidence that the provided images were sufficient for final diagnostic evaluation as opposed to requiring additional diagnostic imaging, as previously described [12].

After completion of the interpretation sessions, one reader (reader 1) evaluated breast density for each patient according to the BI-RADS, 5th edition, classification and measured the maximum diameter of each lesion on DBT images.

All patients with suspicious breast findings on initial workup (including DBT, DBT with spot compression view, ultrasound, and MRI) underwent either ultrasound-guided, DBT-guided, or MRI-guided percutaneous biopsy. Patients with malignant lesions and lesions of uncertain malignant potential on percutaneous biopsy underwent subsequent surgery. Lesions of uncertain malignant potential (i.e., high-risk lesions such as radial scars and atypia) that were not upgraded to malignancy at surgery were considered benign for purposes of analysis. Lesions that were not biopsied were classified as benign on the basis of at least 1 year of imaging follow-up.

Maximum lesion diameters on DBT, MGD, and confidence scores were summarized as median and interquartile range (IQR). Intrareader agreement on BI-RADS categorization was evaluated with weighted Cohen kappa coefficients, and interreader agreement was evaluated with Fleiss kappa coefficient. The classifications were as follows [19]: < 0.00, less than chance agreement; 0.00–0.20, slight agreement; 0.21–0.40, fair agreement; 0.41–0.60, moderate agreement; 0.61–0.80, substantial agreement; and 0.81–1.00, almost perfect agreement. Confidence scores were compared for DBT with and without DBT spot compression views by Wilcoxon matched pairs test. Accuracy, sensitivity, spec-ificity, NPV, and PPV were calculated. Lesions assessed BI-RADS category 2 or 3 were considered negative, and lesions assessed BI-RADS category 4A or higher were considered positive. The Mc-Nemar test was used to compare the diagnostic performance of the BI-RADS category assessments between DBT with and DBT without spot compression views.

Evaluation of diagnostic performance was also stratified by breast density according to the BI-RADS classification: not dense (types A and B) versus dense (types C and D), patient age (< 50 years vs ≥ 50 years), and presence versus absence of personal history of ipsilateral breast cancer (given the potential for postoperative distortion to affect ease of diagnostic evaluation). Diagnostic performance of the assigned probabilities of malignancy on a 5-point Likert scale was evaluated by ROC analysis with calculation of AUC. AUC was compared between DBT with and without spot compression views and among the three readers for DBT with and without spot compression views by Hanley-McNeil test.

MGD was compared between DBT spot compression views and the corresponding DBT views (i.e., CC DBT if DBT spot compression view was obtained in CC projection and MLO DBT if DBT spot compression view was obtained in MLO projection) by Wilcoxon matched pairs test. Values of p < .05 were considered statistically significant. Statistical analysis was performed with SAS software (version V9.4, SAS Institute).

A DBT spot compression image was obtained for an equivocal DBT finding in 144 patients during the study period. Of these, 42 patients were excluded for the following reasons: presence of a postbiopsy marker clip on DBT (n = 2), no clear abnormality on retrospective review of DBT images for lesion for which DBT spot compression view was used (n = 17), DBT and DBT with spot compression view not performed in the same projection (n = 7), and 1-year follow-up not available for a lesion that was not biopsied (n = 16). After these exclusions, the final study sample included 102 lesions in 102 patients (all women; mean age, 60 years). Figure 1 shows the flow of patient selection.

A total of 38% (39/102) of patients had a personal history of breast cancer (19 ipsilateral, 20 contralateral), and 15% (15/102) had synchronous breast cancer (eight ipsilateral, seven contralateral). DBT was initially performed as a screening or annual follow-up examination for a personal history of breast cancer in 52 patients, as diagnostic imaging for a palpable finding in three patients, as diagnostic imaging for an abnormal mammogram in 32 patients, and as a staging evaluation for synchronous breast cancer in 15 patients. Aside from the three patients who underwent DBT because of a palpable finding, no patient had clinical symptoms. Twelve lesions were described on DBT as masses (12%), 24 as architectural distortions (24%), and 66 (65%) as asymmetries. The median maximum lesion diameter on DBT was 11 mm (IQR, 7–15 mm).

Among the 102 patients, 98 underwent subsequent evaluation by breast ultrasound, which showed a correlate in 29 patients. The ultrasound correlate showed typically benign findings in eight patients (one intramammary lymph node, two cysts, five benign post-surgical changes) and suspicious findings in 21 patients, all of whom underwent ultrasound-guided percutaneous core biopsy. In 69 patients, no ultrasound correlate for the equivocal finding was identified. Among those without a correlate on ultrasound, seven patients underwent subsequent percutaneous biopsy because of the level of suspicion for the lesion, performed by tomosynthesis guidance in six patients and by MRI guidance in one patient. Thus, 28 (27%) patients underwent percutaneous biopsy. Histopathologic analysis revealed malignant findings in 18 of 28 (64%) lesions, benign findings in 3 of 28 (11%) lesions, and findings of uncertain malignant potential (i.e., high-risk lesions) in 7 of 28 (25%) lesions. No lesion of uncertain malignant potential was upgraded after surgical excision; all such lesions were thus categorized as benign.

Malignancy was found in 9 of 12 (75%) masses, 8 of 24 (33%) architectural distortions, and 1 of 66 (2%) asymmetries. Among the 74 patients who did not undergo percutaneous biopsy, 63 (85%) underwent follow-up imaging for at least 2 years, and 11 (15%) underwent follow-up for 1–2 years. All of the lesions in these patients were classified as benign on the basis of absence of suspicious findings on follow-up imaging. The 11 lesions classified as benign based on 1–2 years of imaging follow-up were all asymmetries. Three of the patients with these lesions also underwent breast MRI after initial workup showed no suspicious enhancement in the area of concern; one had an ultrasound correlate with typically benign findings; and the other seven had no ultrasound correlate. According to the reference standard, a total of 84 of 102 (82%) lesions were benign, and 18 of 102 (18%) lesions were malignant.

Intrareader agreement on BI-RADS categorization between the two readings performed by reader 1 was moderate for DBT without (κ = 0.43 [95% CI, 0.31–0.55]) and good for DBT with (κ = 0.72 [95% CI, 0.63–0.82]) spot compression views. Interreader agreement among the three readers for BI-RADS categorization was fair for DBT without (κ = 0.21 [95% CI, 0.09–0.32]) and moderate for DBT with (κ = 0.45 [95% CI, 0.33–0.57]) spot compression views.

The median confidence score for reader 1 on DBT was 3 (IQR, 2–3), reader 2 was 3 (IQR, 2–3), and reader 3 was 2 (IQR, 2–3). The scores for DBT with the spot compression view were 4 (IQR, 3–4), 4 (IQR, 3–4), and 4 (IQR, 3–4). For all three readers, the confidence score was significantly higher (all p < .001) with the addition of the DBT spot compression view.

Table 1 shows the distribution of BI-RADS categorizations of benign and malignant lesions for DBT with and without spot compression views for all three readers. For the three readers, among the 84 benign lesions, the number classified BI-RADS category 2 on DBT without spot compression views increased from 13 to 59, 7 to 65, and 8 to 69 with spot compression; lesions classified BI-RADS category 3 decreased from 45 to 17, 56 to 14, and 49 to 9; and lesions classified BI-RADS category 4A and higher decreased from 26 to 8, 21 to 5, and 27 to 6. For the three readers, among the 18 malignant lesions, the number classified BI-RADS category 4A or higher for DBT without versus DBT with spot compression views for reader 1 decreased from 18 to 16, for reader 2 increased from 12 to 17, and for reader 3 increased from 16 to 18. Across the three readers, the cancers missed on DBT with spot compression views were a lymphoma assessed BI-RADS category 3 by readers 1 and 2 and an invasive ductal carcinoma assessed BI-RADS category 2 by reader 1.

Table 2 compares diagnostic performance between DBT and DBT with a spot compression view based on the BI-RADS categorizations. For all three readers, accuracy was significantly higher (p < .001) when DBT spot compression views were used (accuracy for DBT without vs DBT with spot compression views, 75% vs 90%, 74% vs 94%, and 72% vs 94%). For reader 2, sensitivity was significantly higher (p = .03) with spot compression views (sensitivity of DBT without vs with spot compression views, 67% vs 94%); for reader 3, sensitivity was higher for DBT with spot compression views (sensitivity of DBT without vs with spot compression views, 89% vs 100%), but this difference was not significant (p = .16). For reader 1, sensitivity was lower when DBT spot compression views were used (sensitivity of DBT without vs with spot compression views, 100% vs 89%), but this difference was not significant (p = .16). For all three readers, specificity was significantly higher (p < .001) with the use of DBT spot compression views (specificity of DBT without vs with DBT spot compression views, 69% vs 90%, 75% vs 94%, and 68% vs 93%).

Table 3 shows diagnostic performance stratified by breast density, patient age, and personal history of ipsilateral breast cancer. Accuracy was significantly higher (all p < .05) for DBT with versus DBT without spot compression views for nondense breasts for all three readers, for dense breasts for reader 3, for age younger than 50 years for reader 3, for age 50 years or older for all three readers, for no personal history of ipsilateral breast cancer for all three readers, and for personal history of ipsilateral breast cancer for readers 2 and 3. For all other stratifications, accuracy was higher for DBT with versus DBT without spot compression views, although the difference was not statistically significant (all p > .05).

Based on the assessment of probability of malignancy on a 5-point Likert scale, AUC was significantly greater with than without DBT spot compression views for reader 2 (0.93 [95% CI, 0.87–1.00] vs 0.69 [95% CI, 0.55–0.83]; p < .001) and for reader 3 (0.98 [95% CI, 0.96–1.00] vs 0.76 [95% CI, 0.64–0.88]; p < .001) but not for reader 1 (0.90 [95% CI, 0.80–0.99] vs 0.88 [95% CI, 0.80–0.97]; p = .78) (Fig. 2). AUCs varied significantly among readers not using (p = .03) but not among those using (p = .19) DBT spot compression views.

Figure 3 shows a patient in whom the DBT spot compression view was helpful in resolving a questionable asymmetry. Figure 4 shows a patient in whom the DBT spot compression view improved the conspicuity of a subtle architectural distortion.

Findings regarding MGD are summarized in Table 4. The median MGD was 1.97 mGy for DBT with a spot compression view, 1.78 mGy for DBT in the standard CC view, 1.81 mGy for DBT in the standard MLO view, and 1.75 mGy for DBT in the standard view (whether CC or MLO) corresponding to the DBT spot compression view in the given patient. Median MGD was significantly higher for the DBT spot compression view than for the corresponding DBT standard view (p = .03). The median difference in MGD between the DBT spot compression view and the corresponding DBT standard view was 0.06 mGy.