Screening Outcomes of Supplemental Automated Breast Ultrasound in Women With Nondense Breasts Undergoing Mammography.

INTRODUCTION
Digital mammography (DM) is the primary screening modality for breast cancer and has been shown to significantly reduce breast cancer mortality. However, its sensitivity is limited in women with dense breast tissue [123]. Breast ultrasound (US) is widely used as a supplementary screening modality to address this limitation. Breast US has the advantage of not exposing the breast to radiation and is not affected by mammographic density. It can effectively detect mammographically occult, early-stage invasive cancers and help reduce interval cancer rates [45678].
Numerous studies have demonstrated that supplemental US increases the cancer detection rate (CDR) and sensitivity, albeit with reduced specificity and positive predictive value (PPV) [569101112]. In the large-scale multicenter Japan Strategic Anticancer Randomized Trial (J-START), the combination of mammography and handheld US (HHUS) improved sensitivity and reduced interval cancer rates, although this combination showed significantly lower specificity [5]. Similarly, a large prospective multicenter study showed that adding automated breast US (ABUS) to DM detected 1.9 additional breast cancers per 1,000 women with dense breasts, most of which were small invasive cancers [11].
Most studies on supplemental US have focused on women with dense breasts [91011], whereas evidence for its use in nondense breasts remains limited and controversial [13141516]. The secondary analysis from J-START showed improved CDR with supplemental US in women with nondense breasts [14]. In contrast, two large cohort studies conducted in the United States and Austria found no significant differences in CDRs between mammography alone and mammography with supplemental US in nondense populations [151617]. However, all of these studies relied on HHUS and were therefore subject to operator dependence. ABUS has been introduced to overcome this limitation. A recent subgroup analysis [13] of ABUS showed improved CDR in both dense and nondense breasts; however, the increase in sensitivity in nondense breasts did not reach statistical significance, likely due to the limited sample size. Taken together, previous studies included women with nondense breasts only as secondary subgroups, creating uncertainty regarding the true performance of supplemental US in this population.
Therefore, this study aimed to evaluate the screening performance of supplemental ABUS combined with DM in a larger cohort of women with nondense breasts from two institutions.

MATERIALS AND METHODS

Study Population
This retrospective study was approved by the Institutional Review Boards of two tertiary institutions (Samsung Medical Center [IRB No. SMC IRB 2025-03-066] and Kangbuk Samsung Hospital [IRB No. KBSMC 2024-05-026]), and the requirement for written informed consent was waived. To ensure adequate statistical power, the required sample size was calculated as at least 1,368 women. A search of institutional radiology databases across two institutions initially identified 9,874 asymptomatic consecutive women aged ≥20 years who underwent both DM and supplemental ABUS for breast cancer screening between January 2020 and December 2023. Among them, only women with mammographically nondense breasts, classified as almost entirely fatty or scattered fibroglandular density based on the breast density reported in the DM radiological report, were included in this analysis [18]. To ensure adequate confirmation of the outcome assessment, we excluded women without sufficient follow-up data or an established final diagnosis as follows: 1) less than 1 year follow-up for Breast Imaging Reporting and Data System (BI-RADS) category 1 or 2 assessments, 2) less than 2 years follow-up for BI-RADS 3 category lesions, 3) BI-RADS category 0 lesions without established final diagnoses, and 4) BI-RADS category 4 or 5 lesions without pathologic diagnoses (Fig. 1). Detailed information regarding sample size estimation is provided in the Supplement.

DM and ABUS Examinations
All women underwent standard two-view full-field DM of both breasts using mammography units (Selenia Dimensions, Hologic, Marlborough, MA, USA; MAMMOMAT Revelation, Siemens, Erlangen, Germany; Senographe Pristina, GE Healthcare, Chicago, IL, USA). The mean interval between DM and ABUS examinations was 2.3 days (range, 0–90 days)
At both institutions, all ABUS examinations were performed using the same system (Invenia ABUS; GE Healthcare) by trained technologists, with a wide-field-of-view transducer (Reverse Curve™ ultra-broadband transducer, GE Healthcare; frequency range: 6–15 MHz; aperture length, 15.4 cm; transducer bandwidth: 85%; imaging depth: up to 6.0 cm). At least three views (anteroposterior, lateral, and medial) were obtained for each breast. Detailed information regarding DM and ABUS image acquisition is provided in the Supplement.

Image Interpretation and Management
The DM images were reviewed independently by a board-certified breast radiologist at each participating institution, each with 2–33 years of experience in breast imaging. All interpretations were performed using a picture archiving and communication system without access to the ABUS information.
Using a dedicated ABUS workstation (Invenia ABUS; GE Healthcare), a board-certified breast radiologist with 5–33 years of experience in breast imaging interpreted the ABUS images at each institution. Radiologists interpreting the ABUS images were not blinded to the DM results and had access to prior US studies. However, ABUS readings were performed independently without integration of DM images. The background echotexture on ABUS was visually assessed and categorized as homogeneous–fat, homogeneous–fibroglandular, or heterogeneous by the radiologists [18].
DM and ABUS images were independently assigned a BI-RADS category ranging from 0 to 5 according to the BI-RADS guidelines [18]. The final clinical recommendation was made based on the more suspicious BI-RADS category assigned by either DM or ABUS. Detailed imaging management and follow-up protocols are provided in Supplement.

Data Collection
The reference standards included histopathological results from surgical excision (n = 4), US-guided core biopsy (n = 41), vacuum-assisted biopsy (n = 3), stereotactic biopsy (n = 9), and stability on follow-up imaging (n = 2,847).
Data on age, menopausal status, family history of breast cancer, BRCA mutations, and personal history of high-risk lesions were extracted from the electronic medical records. For patients with confirmed cancer, data on histologic type, pathologic tumor size, immunohistochemical status, and axillary lymph node status were collected from pathological reports (Supplement).
DM and ABUS imaging data, including BI-RADS categories, mammographic density, and background echotexture, were collected from radiologic reports. Two breast imaging radiologists (M.R.K. and J.S.C.) retrospectively reviewed the newly diagnosed cancers to determine whether imaging abnormalities were visible on the most recent ABUS or DM images. The cancers were categorized as those detected only by DM, those detected only by ABUS, and those detected by both ABUS and DM.

Outcome Measures
Screening-detected cancer was defined as cancer diagnosed at, or as a direct result of, an index screening examination (DM or ABUS plus DM) that was classified as BI-RADS categories 0, 3, 4, or 5, with a tissue diagnosis of cancer within 1 year of the screening examination; BI-RADS category 0 and 3 lesions were included only when malignancy was confirmed at the same site. Interval cancer was defined as cancer diagnosed due to clinical symptoms following a test-negative screening examination (BI-RADS category 1 or 2) and before the 12-month follow-up screening was performed. Each screening examination was treated as an independent unit of analysis, and the follow-up for cancer ascertainment was truncated at the next screening examination to avoid double-counting across rounds.
The primary endpoint was the CDR, defined as the number of cancers detected per 1,000 screening examinations. The secondary endpoints were sensitivity, specificity, PPV for recall (PPV1), PPV for biopsy (PPV3), abnormal interpretation rate (AIR), short-term follow-up rate, and biopsy rate. Diagnostic accuracy was also assessed using the area under the receiver operating characteristic curve (AUC). The additional exploratory endpoints included false-positive biopsy rates and cancer characteristics by detection method. Detailed definitions of the outcome measures are provided in the Supplement.

Statistical Analysis
Baseline characteristics of the study population by institution were compared using the independent two-sample t-test for continuous variables and the χ2 test or Fisher’s exact test for categorical variables.
The screening outcome measures of DM alone and ABUS combined with DM (ABUS plus DM) were analyzed for all examinations and subgroups. A generalized estimating equation with a logit link function and an independent working correlation matrix was used to model repeated examinations in the same woman. Differences in the outcome measures between DM alone and ABUS plus DM were estimated. The generated values and 95% confidence intervals (CIs) were calculated using a bootstrapped approach with 1,000 resamples [19]. Subgroup analyses were performed to compare the outcome measures based on 1) breast density (almost entirely fatty vs. scattered fibroglandular density) and 2) screening rounds (prevalent vs. incident screening based on prior supplemental breast US).
Imaging and pathologic characteristics of cancers detected only by ABUS and of the other cancers were compared using the independent two-sample t-test and the χ2 test or Fisher’s exact test.
All statistical tests were two-sided, and a P-value <0.05 was considered indicative of a significant difference. All analyses were conducted by biostatisticians (M.Y.L. and S.M.) using R software (version 3.6.3; R Foundation for Statistical Computing, Vienna, Austria).

RESULTS

Baseline Characteristics
Among 9,874 consecutive asymptomatic women, 1,961 had nondense breasts and underwent 3,426 screening examinations during the study period. Of these, 522 examinations were excluded because of 1) less than 1 year of follow-up for BI-RADS category 1 or 2 lesions (n = 436), 2) less than 2 years of follow-up for BI-RADS 3 category lesions (n = 78), 3) BI-RADS category 0 lesions for which final diagnoses were not established (n = 5), and 4) BI-RADS category 4 or 5 lesions without pathological diagnoses (n = 3). Finally, a total of 1,683 asymptomatic women with nondense breasts (mean age, 59 ± 10 years) who underwent 2,904 pairs of DM and ABUS examinations for breast cancer screening were included (Fig. 1).
Table 1 presents the baseline characteristics of the study population by institution. Most participants were postmenopausal (1,493/1,683; 88.7%). A total of 1,569 women (93.2%) had scattered fibroglandular density, whereas 114 women (6.8%) had almost entirely fatty breasts. Seventy-four (4.4%) women had risk factors for breast cancer. Kangbuk Samsung Hospital had a higher proportion of women with risk factors for breast cancer (8.4% vs. 2.9%, P < 0.001) and women with heterogenous echotexture (60.9% vs. 12.2%, P < 0.001) compared with Samsung Medical Center. Of the 1,683 women, 26 (1.5%) were diagnosed with breast cancer, whereas 1,657 (98.5%) did not have breast cancer. Women diagnosed with breast cancer were older (63 ± 9 years vs. 59 ± 10 years, P = 0.04) and tended to have risk factors (19.2% vs. 4.2%, P < 0.001) than those without breast cancer (Supplementary Table 1).

DM vs. ABUS Plus DM: Main Analysis
The outcome measures are presented in Table 2. Overall, ABUS plus DM yielded a higher CDR of 9.0 per 1,000 examinations (95% CI: 7.7, 10.0) than DM alone, which yielded a CDR of 7.9 per 1,000 examinations (95% CI: 6.9, 8.8; P < 0.001). Supplemental ABUS detected an additional 1.1 cancers per 1,000 examinations (95% CI: 0.5, 1.2). ABUS plus DM demonstrated higher sensitivity than DM alone—100% (95% CI: 100, 100) vs. 88.5% (95% CI: 85.7, 93.4; P < 0.001) with a gain of 11.5% (95% CI: 6.5, 14.3). However, ABUS plus DM showed lower specificity (95.0%; 95% CI: 94.8, 95.3) than DM alone (97.9%; 95% CI: 97.7, 98.1; P < 0.001). Additionally, ABUS plus DM demonstrated lower PPV1 (15.4% [95% CI: 13.4, 16.8]) vs. 27.4% [95% CI: 24.3, 30.4]) and PPV3 (49.1% [95% CI: 44.4, 53.5]) vs. 59.0% [95% CI: 52.9, 63.9]) (all P < 0.001). In contrast, ABUS plus DM demonstrated a higher AUC of 0.98 (95% CI: 0.97, 0.98) in comparison with 0.93 (95% CI: 0.92, 0.96; P < 0.001) for DM alone. Meanwhile, other metrics were higher with ABUS plus DM, including an AIR of 5.8% (95% CI: 5.5, 6.1) in comparison with 2.9% (95% CI: 2.7, 3.1); a short-term follow-up rate of 2.9% (95% CI: 2.7, 3.1) in comparison with 0.6% (95% CI: 0.5, 0.7); and a biopsy rate of 1.8% (95% CI: 1.7, 2.0) in comparison with 1.3% (95% CI: 1.2, 1.5) (all P < 0.001). ABUS plus DM also showed a higher false-positive biopsy rate of 0.9% (95% CI: 0.8, 1.0) in comparison with 0.5% (95% CI: 0.4, 0.6; P < 0.001) for DM alone.

DM vs. ABUS plus DM: Subgroup Analysis
The outcome measures according to mammographic density are presented in Table 3. Among women with scattered fibroglandular densities (n = 2,721), ABUS plus DM showed a higher CDR of 8.5 per 1,000 examinations (95% CI: 6.9, 9.4) vs. 7.4 per 1,000 examinations (95% CI: 6.1, 8.2; P < 0.001) for DM alone and a higher sensitivity of 100% (95% CI: 100, 100) vs. 87.0% (95% CI: 83.3, 94.7; P < 0.001). However, ABUS plus DM showed lower specificity (94.9%; 95% CI: 94.6, 95.2) in comparison with DM alone (97.9%; 95% CI: 97.7, 98.1; P < 0.001). In women with almost entirely fatty breasts (n = 183), ABUS plus DM yielded the same CDR of 16.4 per 1,000 examinations (95% CI: 12.2, 18.3) and the same sensitivity of 100% (95% CI: 100, 100) in comparison with DM alone. However, ABUS plus DM showed a slightly lower specificity of 97.2% (95% CI: 96.9, 98.1) in comparison with 98.3% (95% CI: 98.1, 98.7) for DM alone, along with a higher AIR of 4.4% vs. 3.3% (all P < 0.001).
The outcome measures according to the prevalence (n = 780) and incidence screening rounds (n = 2,124) are described in Supplementary Table 2. Twenty-one cancers (80.8%) were detected during the incidence rounds, and five cancers (19.2%) were detected during the prevalence round.

Cancer Characteristics
A total of 26 breast cancers were identified (0.9%, 26/2,904); all of these cancers were detected during screening, and no interval cancers were observed. These included 19 (73.1%) invasive cancers and seven (26.9%) ductal carcinomas in situ. Among these cancers, three (11.5%) had lymph node metastasis. Of the 26 breast cancers, 17 (65.4%) were detected by both DM and ABUS, six (23.1%) were detected only by DM, and three (11.5%) were detected only by ABUS (Table 4, Figs. 2, 3). All three cancers detected only by ABUS were invasive ductal carcinomas at the T1 stage (mean size, 1.0 cm), node-negative, and classified as the luminal A subtype. Most cancers detected only by DM were ductal carcinomas in situ (66.7%, 4/6). No significant differences were observed between cancers detected only by ABUS and those detected by DM. On ABUS, most breast cancers appeared as masses (85.0%, 17/20). No differences were found in the imaging characteristics between cancers detected only by ABUS and those detected by both DM and ABUS (Supplementary Table 3).

DISCUSSION
The efficacy of supplemental ABUS in women with mammographically nondense breasts remains uncertain. This study aimed to evaluate the screening outcomes of supplemental ABUS combined with DM in this population. ABUS plus DM achieved higher CDR and sensitivity, but lower specificity than DM alone. In summary, supplemental ABUS improved CDR and sensitivity, but reduced specificity in women with nondense breasts.
While supplemental HHUS or ABUS is well-established in women with dense breasts [5691011], evidence for their use in nondense breasts is limited. In this study, adding ABUS to DM achieved 100% sensitivity and a CDR of 9.0 per 1,000 examinations, which is consistent with prior findings in a smaller nondense cohort [13]. The observed sensitivity in women with nondense breasts was higher than the previously reported rate for supplemental HHUS (77.5%–93.2%) [613141620], and comparable to or higher than that for ABUS (81.0%–100%) [101121] in women with dense breasts. The consistently high sensitivity of ABUS across both dense and nondense breasts may be attributed to its standardized image acquisition, whole-breast coverage, and multiplanar reconstruction, particularly in the coronal view [22].
Despite the high absolute sensitivity, the incremental gain in our study was 11.5%, which was lower than the values reported in populations with dense breasts (14.1%–41.0%) [678910111820]. This smaller gain likely reflects the inherently higher performance of DM in nondense breasts, where better tissue contrast facilitates lesion detection [2321]. Previous studies in nondense breast populations reported lower and variable sensitivity gains (8.4%–32.2%), which were largely influenced by baseline DM sensitivity. For example, the J-START trial reported an increase from 60.9% to 93.1% [14], whereas Buchberger et al. [16] reported an increase from 86.6% to 95.0%. In our study, DM alone achieved 88.5% sensitivity, exceeding the Breast Cancer Surveillance Consortium benchmark of 87.6% [23], which may have limited the additional benefits of ABUS. Similarly, the CDR gain was modest. Although the observed CDR of 9.0 per 1,000 examinations fell within the range reported for women with dense breasts (2.4–12.3 per 1,000) [13141516], the incremental gain of 1.1 per 1,000 examinations was lower than the range reported for women with dense breasts (1.9–7.7 per 1,000) [6910111314202124]. Thus, while ABUS enhances detection in nondense breasts, its incremental value is smaller when the baseline DM performance is strong. Although sensitivity and CDR improved, specificity and PPV declined, and AIR increased. Nevertheless, overall performance remained favorable, with the specificity, PPV1, and AIR of 95.0%, 15.4%, and 5.8%, respectively, outperforming HHUS in dense breasts (specificity: 72%–89.4% PPV1: 6.6%–10.5%; AIR: 11.0%–15.2%) [6131416]. However, the observed sensitivity of 100% should be interpreted with caution since it was likely influenced by the small number of cancer cases. Only three cancers were additionally detected by ABUS, thereby limiting the robustness of the findings. Still, these cancers were all small, invasive, and node-negative, suggesting that supplemental US may have clinical relevance by enabling the detection of earlier-stage cancers despite the modest absolute gain and the accompanying decline in specificity and PPV.
The clinical implications of our findings warrant careful consideration. The observed benefit of supplemental ABUS was primarily evident in women with scattered fibroglandular densities, who constituted the vast majority of our nondense cohort. Given the intermediate composition of breasts with scattered fibroglandular densities—with mammographic sensitivity lower than that in almost entirely fatty breasts but higher than that in dense breasts—further optimization of US-based screening strategies may still be possible in this population. In contrast, DM alone achieved excellent sensitivity and specificity in women with almost entirely fatty breasts, and ABUS provided no additional benefit in this subgroup. As the number of women with almost entirely fatty breasts was small, our study was underpowered to evaluate this subgroup, and the findings should be interpreted with caution. Taken together, these results suggest that ABUS should not be routinely applied to all women with nondense breasts but may be considered selectively for those with scattered fibroglandular density.
Beyond diagnostic performance, ABUS has been shown to improve reproducibility, reduce operator dependence, and increase efficiency in comparison with HHUS [22], making it suitable for integration into population-based screening workflows [2526]. However, its implementation requires dedicated equipment and a longer acquisition time, which may affect resource allocation and cost-effectiveness. Nevertheless, acquisition times decrease as technologists gain experience, reflecting a learning-curve effect [27]. Because image acquisition can be delegated to technologists, and interpretation is performed separately, ABUS may improve workflow efficiency in high-volume screening settings [28]. In addition, recent teleradiology screening programs in China have demonstrated that ABUS can achieve favorable screening performance while mitigating local resource constraints [29]. These practical considerations underscore the potential of ABUS to enhance screening efficiency and accessibility, even in women with nondense breasts, while highlighting the need for further evaluation of its cost-effectiveness in population-based settings.
Women diagnosed with breast cancer in the present study were more likely to have risk factors. Given that 80.8% (21/26) of the cancers were detected during incidence screening, these findings emphasize the importance of regular screening, particularly in women at a higher-than-average risk. Supplemental US screening can help identify early-stage invasive cancers that are not visible on mammography, thereby facilitating diagnosis at a more treatable stage and allowing for less aggressive intervention. This is particularly relevant for women with higher-than-average risk, including those with nondense breast tissue, who may benefit from supplemental screening methods—supporting the broader incorporation of this approach into risk-based screening strategies.
Our study had several limitations. First, its retrospective design and reliance on radiologic reports may have introduced selection bias. Breast density was classified according to BI-RADS assessments from radiologic reports, which were subject to inter-observer variability. However, previous studies have demonstrated high inter-observer agreement when classifying dense and nondense breasts [3031], and the use of experienced breast radiologists in this study likely minimized misclassification. Second, our study cohort may not fully represent the general population of women with nondense breasts, because the study cohort only included women who underwent both DM and ABUS. Because screening mammography generally performs well in nondense breasts and supplemental US is not typically recommended in this group, our findings should be interpreted as reflecting ABUS performance in a selected subgroup who received supplemental imaging rather than in the overall population of women with nondense breasts. Nevertheless, in countries such as Korea and Japan, supplemental US screening is frequently performed even in women with nondense breasts, owing to the high accessibility and low cost of US as well as the smaller breast sizes in these populations in comparison with those in Western women [81432]. In addition, in China, US-based screening programs have been adopted to improve screening access for the rural population [833]. Therefore, our findings may still have relevance in Asian screening settings, although further population-based studies are required for validation. Third, although our study met the pre-specified sample size requirement, the number of cancer cases, particularly those detected only by ABUS, was limited. This limited the generalizability of our findings and underscored the need for validation in larger prospective cohorts. However, to the best of our knowledge, this study is one of the largest dedicated analyses of supplemental ABUS in Asian women with nondense breasts, addressing an important evidence gap that has not been addressed in previous subgroup or cohort studies. Fourth, the number of women with almost entirely fatty breasts was small, limiting the statistical power to evaluate the performance of ABUS in this subgroup. Further studies involving a larger cohort of women with almost entirely fatty breasts are warranted. Fifth, institutional differences in practice patterns may have affected the results. Kangbuk Samsung Hospital performed ABUS only during incidence screening and had more women with risk factors for breast cancer, whereas Samsung Medical Center included prevalence screening. Sixth, inter-observer variability in image interpretation was not assessed because of the retrospective nature of the study. Lastly, only three cancers were detected only by ABUS, limiting the comparative analysis to DM-detected cancers.
In conclusion, supplemental ABUS improved cancer detection and sensitivity in women with nondense breasts, with the benefit primarily observed in women with scattered fibroglandular density. These findings indicate that supplemental screening may be valuable in women with scattered fibroglandular density, rather than in all women with nondense breasts.