ABUS-based glandular tissue component classification for breast cancer risk prediction in Chinese women with dense breasts: a retrospective study.

Introduction
According to the 2022 Global Cancer Statistics released by the International Agency for Research on Cancer, breast cancer has become the second most commonly diagnosed malignancy worldwide and the leading cause of cancer-related death among women [1]. Consequently, reducing breast cancer mortality is an urgent public health priority. Early screening, diagnosis, and intervention have been proven effective in lowering breast cancer mortality rates. The development of breast cancer is associated with a range of factors. Several factors have been reported to be associated with the development of breast cancer, such as genetics, age, dense breasts, reproductive history, and obesity [2–4]. These factors have been widely incorporated into breast cancer screening programs and risk prediction models [5, 6].
Current breast cancer risk prediction models are mainly categorized into two types: those incorporating clinical risk factors and those integrating imaging-derived characteristics. Among them, the Breast Cancer Surveillance Consortium (BCSC) risk prediction model [7] calculates individualized risk estimates by integrating key variables such as chronological age, race/ethnicity, a history of breast cancer in first-degree relatives, a personal history of benign breast biopsy, and Mammographic density (MD). A key advantage of the BCSC model over those that solely incorporate clinical risk factors is its integration of MD—a independent risk determinant—thus significantly enhancing its predictive performance [8].
MD has been established as a well-recognized, independent, and significant risk factor, with studies demonstrating that women with dense breasts have a 4 to 6 fold increased risk of developing breast cancer compared to those with non-dense breast [9–11], its risk level may be second only to age and genetic susceptibility to breast cancer (e.g., BRCA1/2) [12, 13]. Currently, mammography (MG) remains the gold standard for assessing MD, based on this, automated MG-based tools (such as Volpara Density and STRATUS), have become established methods not only for quantifying MD but also for integrating it into validated risk prediction models like the BCSC model [5, 14]. On MG, higher fibroglandular tissue (FGT) content corresponds to greater breast density. Both FGT and lesions appear bright white on X-ray images, resulting in a “masking effect.” This type of dense tissue significantly reduces the sensitivity of MG screening [15]. For dense breasts, ultrasound imaging can clearly distinguish between glandular tissue and fibrous stroma based on differences in their grayscale appearance. The glandular tissue, composed of terminal ducts and lobules, appears hypoechoic, while the fibrous stroma presents as hyperechoic [16]. Nevertheless, criteria for quantifying ultrasonographic FGT composition lack standardization across studies and clinical practice [17]. Lee et al. [18] proposed the concept of the Glandular Tissue Component (GTC) based on a qualitative assessment of the volume percentage of glandular tissue within FGT on ultrasound images, categorizing it into four types: A (Minimal, < 25% of the FGT), B (Mild, 25%-49% of the FGT), C (Moderate, 50%-74% of the FGT), and D (Marked, ≥ 75% of the FGT). They found that a higher GTC, indicating a greater proportion of glandular tissue, can serve as an independent predictor of breast cancer. Bunnell et al. [19] also predicted breast density using ultrasound images. Currently, GTC is receiving extensive attention and has become a focus of research among many scholars. In the latest edition (V2025), GTC has been included as an important reference factor for breast tissue composition [20].
Currently, the assessment of the GTC relies predominantly on two-dimensional images obtained via handheld ultrasound. Lee et al. [18] recommended selecting glandular tissue from the upper outer quadrant of the breast, approximately 2 cm from the nipple, as the region for evaluation. However, some researchers [21] have indicated that a single-frame image of the glandular tissue from this location may not completely and accurately represent the entire breast. They propose that using volumetric ultrasound or video-format ultrasound images may allow for a more accurate assessment of a patient’s GTC.
ABUS is a novel breast imaging technique equipped with a high-frequency, wide-volume probe. It performs automated scanning and three-dimensional reconstruction to obtain standardized breast volume data, and compared to two-dimensional ultrasound images, ABUS provides more comprehensive information on glandular tissue and can reduce operator dependence [22, 23]. This study is the first to select images from the ABUS coronal plane to evaluate the correlation between GTC classification and breast cancer risk among Chinese women with dense breasts. We hypothesize that, after adjusting for multiple confounding variables including BCSC risk factors, GTC classification can provide independent predictive value, thereby offering new imaging evidence to optimize risk stratification for this specific population.

Materials and methods

Patients
Employing a retrospective case-control design, this study consecutively screened and enrolled patients with pathologically confirmed diagnoses who were admitted to Hangzhou First People’s Hospital between January and December 2020. After applying the exclusion criteria, 203 patients who met all eligibility criteria were ultimately included as the case group. During the same period, 211 patients with pathologically confirmed benign breast lesions were consecutively enrolled as the control group. All control patients had completed a 5-year follow-up without evidence of malignancy. The histological subtype distributions of benign and malignant lesions are presented in Table 1. Data were collected via retrospective review of electronic medical records, including the following variables for comparison: Demographic and clinical characteristics: age, age at menarche, menopausal status, body mass index (BMI), history of benign breast biopsy, and family history of breast or ovarian cancer.Lesion characteristics: maximum lesion diameter (measured on imaging).These seven parameters are well-established breast cancer risk factors, as referenced in previous literature [24–26]. Additionally, the BCSC Risk Prediction Model (Version 2.0) was used to calculate 5-year breast cancer risk probabilities. GTC classification was conducted using the standardized qualitative assessment method on ultrasound images. The inclusion criteria were: (1) Age ≥ 40 years; (2) Completed both ABUS and MG examinations with complete imaging datasets; (3) Lesions confirmed by histopathology (via surgical excision or core needle biopsy); (4) MD classified as ACR Category C/D (heterogeneously dense or extremely dense). (5) Patients with pathologically confirmed benign lesions were required to complete a 5-year follow-up period. The exclusion criteria were: (1) The interval between ABUS and MG examinations exceeding 6 months; (2) Indeterminate pathological diagnosis; (3) Presence of extensive pathological involvement in the region of interest (ROI); (4) A prior history of breast cancer treatment or any related treatment, including but not limited to breast-conserving surgery, mastectomy, radiotherapy, endocrine therapy, or chemotherapy; (5) Pregnant or lactating status; (6) History of augmentation mammoplasty. Figure 1 describes the flowchart of study population screening. Figure 2 described the complete research process.

The protocol for the present study was approved by the Institutional Review Board (IRB) of Hangzhou First People’s Hospital (Approval No.: IIT 20221202-0194-01) and was conducted in strict accordance with the principles of the Declaration of Helsinki.

Mammography examinations
All MG examinations were performed using a Hologic Selenia Dimensions digital breast tomosynthesis system (Hologic Inc., Marlborough, MA, USA). For each breast, standard craniocaudal (CC) and mediolateral oblique (MLO) views were acquired, with adequate compression applied in the superior-inferior and mediolateral oblique directions, respectively. In cases of asymmetric breast density between bilateral breasts, the highest MD category (per ACR criteria) was recorded for subsequent analysis.

Ultrasound examinations
ABUS images were acquired using a Siemens ACUSON S2000 ABUS system (Siemens Medical Solutions, Mountain View, USA). This system is equipped with a wide-aperture linear transducer (5–14 MHz). A robotic arm was used to automatically acquire volumetric data along pre-defined scanning paths in three standard orientations: anteroposterior (AP), lateromedial (LAT), and mediolateral (MED). Original image data were retrieved from the Picture Archiving and Communication System (PACS) and transferred to a dedicated workstation for multi-planar review. All datasets were systematically re-evaluated in the axial plane, as well as in reconstructed sagittal and coronal planes.

Image analysis
MG Image Interpretation: A double-blinded study design was implemented wherein two radiologists, each with over 10 years of experience in breast imaging, independently assessed the images. They were blinded to the patients’ group assignments (case vs. control) and to each other’s assessments throughout the interpretation process. Evaluations were performed in accordance with the 5th Edition of the ACR BI-RADS, using the following standardized categories: Category A (Almost entirely fatty), Category B (Scattered fibroglandular densities), Category C (Heterogeneously dense), and Category D (Extremely dense). All interpretations were conducted on standardized diagnostic workstations. In cases of initial disagreement between the two readers, a consensus review was conducted to determine the final MD classification.
ABUS Image Interpretation: For each participant, coronal plane images from the LAT view of both breasts were selected for analysis. The ROI for GTC assessment was defined within the upper outer quadrant—corresponding to the 12 to 3 o’clock region in the left breast or the 12 to 9 o’clock region in the right breast—with the retroareolar area explicitly excluded. In cases of asymmetric GTC distribution between the bilateral breasts, the breast with the higher GTC was selected for final analysis. Image quality was ensured by requiring clear visualization of glandular structures. If one breast contained a large cyst, a dominant mass, or significantly dilated ducts (which could obscure parenchymal evaluation), the contra lateral breast was used for analysis instead. Three radiologists, each with over a decade of experience in breast ultrasound interpretation, independently reviewed the ABUS images. To ensure standardized and consistent evaluation, all three readers first completed a structured training module: this required them to independently classify GTC on a set of more than 50 practice ABUS cases. This training ensured that all readers demonstrated proficiency and established interpretive consistency in GTC assessment using ABUS before the formal study initiation. Two of the three readers manually annotated all ABUS images (Fig. 2) according to the GTC classification from the BI-RADS V2025 and were blinded to patient group assignment and each other’s assessments. Any discrepancies in initial assessments were adjudicated by a third radiologist specializing in breast imaging. To evaluate inter-observer agreement in GTC classification between the two radiologists, a four-category agreement assessment was performed on all 414 cases. The consensus interpretation reached by the radiologists was thereby established as the reference standard for all subsequent experimental analyses. Figure 3 shows the GTC classification results and associated annotations based on ABUS images.

Data and statistical analysis
Data were analyzed using SPSS 27.0. Inter-rater agreement for GTC classification was assessed using Fleiss’ weighted kappa and the intraclass correlation coefficient (ICC) with a two-way mixed-effects model [27].Continuous variables (e.g., age, BMI, BCSC risk) are reported as mean ± SD or median (interquartile range [IQR]), depending on the distribution (assessed by the Shapiro-Wilk test and histograms); nonparametric tests (Mann–Whitney U) were used for group comparisons. Categorical variables are presented as n (%), and group differences were assessed with the chi-square test or Fisher’s exact test. To assess the independent predictive value of GTC classification, univariate and multivariate logistic regression analyses were performed. The multivariate model was constructed using a backward stepwise selection procedure. Variables were included based on predefined clinical relevance and previous literature, rather than statistical significance in univariate analysis. The predetermined potential confounders were: age, menopausal status, BMI, age at menarche, family history of breast or ovarian cancer, personal history of benign biopsy, and maximum lesion diameter (the BCSC 5-year risk score was excluded). GTC classification, with type A as the reference, was included as the primary exposure variable. Subgroup analyses stratified by menopausal status were conducted using the same multivariate modeling approach. All tests were two-sided, with a P < 0.05 considered statistically significant.

Results

Patient characteristics
A total of 414 female patients were included in this study. The median age was 50 years (IQR: 45–58; range: 40–73), with the largest proportion (43.0%) in the 45-54-year age group. Among the participants, 211 (51.0%) had benign lesions and 203 (49.0%) had malignant lesions. The median BMI was 22.7 (IQR: 20.90–24.80). The median age at menarche was 15 years (IQR: 14–16), with most patients (60.6%) reporting menarche between 13 and 15 years of age. The majority of patients were premenopausal (52.7%) and had no history of benign breast biopsy (87.0%). A positive family history of breast or ovarian cancer was reported in 4.8% of participants. The median lesion size was 1.6 cm (IQR: 1.00-2.50). The median BCSC 5-year breast cancer risk for all patients was 1.00% (IQR: 0.76–1.27). For breast composition, most patients (75.4%) were classified as ACR BI-RADS Category C (heterogeneously dense). The distribution of ultrasound-based GTC classifications was as follows: A (Minimal), 11.6% (48/414); B (Mild), 44.2% (183/414); C (Moderate), 25.8% (107/414); and D (Marked), 18.4% (76/414) (Table 2).

Predictors of breast cancer risk in the overall population (univariate and multivariate analysis)
Univariate logistic regression analysis revealed that patients with GTC classifications C (50–74%) and D (≥ 75%) had a significantly higher risk of breast cancer compared to those with A (< 25%) (C: Odds ratio [OR] = 2.05, P = 0.044; D: OR = 3.21, P = 0.002; P for trend < 0.001). Larger lesion size was also positively correlated with risk (Category 2: OR = 2.03; Category 3: OR = 4.59; Category 4: OR = 5.41; P for trend < 0.001). Furthermore, advanced age (55–64 years: OR = 2.71; ≥65 years: OR = 6.57), a higher BCSC 5-year risk (OR = 3.29), postmenopausal status (OR = 2.68), and a positive family history of breast or ovarian cancer (OR = 3.29) were all significantly associated with increased risk (all P < 0.05, unless specified above in the GTC analysis). In contrast, no significant associations were found for BMI, history of benign biopsy, or age at menarche categories (all P > 0.05). Detailed information on reference group settings is provided in Table 3.

In the multivariate logistic regression analysis, after adjusting for all relevant confounding factors (regardless of their P-values), menopausal status, family history of breast or ovarian cancer, age, lesion size, and GTC classification were identified as factors significantly associated with breast cancer risk. Postmenopausal women had significantly higher odds of breast cancer compared with premenopausal women (OR = 2.17, P = 0.015). A positive family history was associated with significantly higher odds compared with a negative history (OR = 3.63, P = 0.022).In the age-stratified analysis, the ≥ 65 years group demonstrated significantly elevated odds compared with the 40–44 years reference group (OR = 3.79, P = 0.013). No statistically significant associations were found for the 45–54 years and 55–64 years groups (all P > 0.05). Larger lesion size was significantly associated with increased risk in a graded manner (Category 2: OR = 2.05, P = 0.019; Category 3: OR = 4.20, P < 0.001; Category 4: OR = 6.93, P = 0.002). In terms of GTC classification, both C (50–74%) and D (≥ 75%) categories showed significantly elevated risk compared with the A (< 25%) reference category (C: OR = 2.62, P = 0.017; D: OR = 3.21, P = 0.007). No significant association with breast cancer risk was observed for the B (25–49%) (P = 0.321).All reported analyses were based on the multivariate model, and potential interactions between covariates were tested and adjusted for (Table 3).

Interobserver agreement analysis
Interobserver agreement analysis demonstrated statistically significant concordance between the two readers in the four-category GTC classification (P < 0.001). Cross-tabulation (n = 414) showed an overall raw agreement rate of 80.68% (334/414). The weighted kappa coefficient was 0.810 (asymptotic SE = 0.020, P < 0.001). According to the benchmark scale proposed by Landis and Koch, both the raw agreement and kappa value indicated “almost perfect agreement.” The single-measure intraclass correlation coefficient (ICC) (A,1) was 0.888 (95%CI: 0.866–0.907), and the average-measure ICC(A, k) was 0.941 (95%CI: 0.928–0.951). The results were statistically significant (F[1, 413] = 16.847, P < 0.001) (Table 4).

Subgroup analysis: postmenopausal women
This subgroup analysis included 196 postmenopausal women (75 in the benign group and 121 in the malignant group). Compared with the benign group, the malignant group had a significantly higher median age (P = 0.003), older age at menarche (P = 0.017), a greater proportion of individuals with a history of benign breast biopsy (P = 0.009), larger lesion size (P < 0.001), and higher GTC (P < 0.001).A significant difference in GTC distribution was observed between the two groups (χ²=17.180, P < 0.001). The proportion of combined GTC C and D classifications was significantly higher in the malignant group than in the benign group (malignant: 71.1% vs. benign: 28.9%; malignant: 82.9% vs. benign: 17.1%; P < 0.001). Furthermore, the proportion of the D classification was markedly higher in the malignant group, indicating a positive association between higher GTC category and malignancy.No statistically significant differences were found between the groups regarding family history of breast or ovarian cancer, BMI, or BCSC 5-year risk (all P > 0.05) (Table 5).

Univariate and multivariate logistic regression analyses performed on the 196 postmenopausal women identified larger lesion size and higher GTC category as factors significantly associated with disease risk. Specifically, compared with the reference category (Category 1), the risk was significantly elevated for Category 2 (OR = 2.50, P = 0.036) and Category 3 (OR = 5.98, P < 0.001). In the GTC classification, both the C (50–74%) and D (≥ 75%) categories demonstrated significantly higher odds compared with the A (< 25%) reference category (C: OR = 4.17, P = 0.017; D: OR = 7.38, P = 0.003). No significant association was observed for the B category (25–49%) (P = 0.219). Additionally, a history of benign breast biopsy remained independently associated with a significantly reduced risk of breast cancer (OR = 0.28, P = 0.043) (Table 6).

Subgroup analysis: premenopausal women
This analysis included 218 premenopausal women (136 in the benign group and 82 in the malignant group). Compared with the benign group, the malignant group exhibited a significantly higher proportion of participants with a positive family history of breast or ovarian cancer (P = 0.033) and larger lesion size (P < 0.001). No significant differences were observed between the benign and malignant groups with respect to age, age at menarche, BMI, history of benign breast biopsy, BCSC 5-year risk, or GTC classification (all P > 0.05) (Table 7).

Multivariate logistic regression analysis, following adjustment for potential confounders, revealed a positive family history of breast or ovarian cancer and larger lesion size as factors significantly associated with an increased risk of breast cancer. Specifically, a positive family history was associated with significantly higher odds compared with a negative family history (OR = 3.77, P = 0.038). For lesion size, compared with the reference category (Category 1), the risk was significantly higher for Category 3 (OR = 2.54, P = 0.038) and Category 4 (OR = 6.25, P = 0.010). No significant association was observed for Category 2 (P = 0.392) (Table 8).

Discussion
MD is an established, independent risk factor for breast cancer, which is routinely assessed via MG—a key tool for early detection. Automated MG-based tools (e.g., Volpara) represent the current standard for quantifying MD and integrating it into risk models, underscoring its predictive role. However, dense breasts present a dual challenge: they confer a substantially higher cancer risk, while also creating a “masking effect” on MG that reduces screening sensitivity in this population. The 2025 Edition of the ACR BI-RADS Atlas [20] classifies breast density into four categories based on the quantity and distribution of FGT observed on MG: almost entirely fatty, scattered areas of fibroglandular density, heterogeneously dense, and extremely dense. A critical limitation of MG is that its sensitivity for lesion detection in extremely dense breasts may be as low as 30% [16, 28]. This inherent limitation of MG in dense breast tissue underscores the need for supplemental screening with ultrasound.
Ultrasound classifies breast parenchyma differently. The 5th Edition of the ACR BI-RADS® Atlas classifies ultrasound tissue composition into three distinct background tissue patterns: homogeneous background echotexture-fat, homogeneous background echotexture-fibroglandular, and heterogeneous background echotexture. Heterogeneous patterns can present as either focal or diffuse. A critical limitation of a heterogeneous background echo pattern is its tendency to obscure subtle malignant features (e.g., microcalcifications or spiculations). This masking effect may compromise the sensitivity of ultrasound examinations, potentially leading to a higher false-negative rate. In response to this challenge, the 2025 Edition of the ACR BI-RADS® Atlas [20] introduced the pivotal concept of GTC for ultrasound assessment. This GTC classification system enables breast cancer risk stratification, thereby providing a more effective approach to identifying high-risk women with dense breast parenchyma. Meanwhile, in healthcare settings where ultrasound is widely used for screening, such as in China, GTC can serve as a valid alternative or complementary risk marker. By analyzing ABUS images, this study aims to evaluate the potential of GTC classification for risk prediction among Chinese women with dense breasts.
Moon et al. [29] were the first to put forward the concept of GTC and conducted relevant investigations in a Korean cohort. Their study revealed a positive correlation between GTC and MD, indicating that women with higher MD typically have higher GTC. Furthermore, they found that higher GTC was associated with the following characteristics: relatively younger age (< 50 years), non-obesity (BMI < 25 kg/m²), premenopausal status, nulliparity, and a first-degree family history of breast cancer. A prospective cohort study evaluating GTC revealed that women with high baseline GTC exhibited a higher cumulative incidence of breast cancer. In a secondary analysis, this significant association persisted even in women with extremely dense breasts on MG. In a multivariate analysis, after adjusting for other risk factors—including age, menopausal status, family history of breast cancer, history of benign breast biopsy, and MD—baseline GTC remained the only independent risk factor significantly associated with breast cancer (OR = 1.49, P = 0.03) [30]. Notably, GTC maintained its predictive value for breast cancer risk even when repeated GTC measurements from each participant were incorporated into the analysis.
This study explored the clinical applicability of ultrasound-based GTC classification for breast cancer risk assessment. Building on the foundational work by Moon et al., our study focused on a Chinese female cohort and incorporated additional predictive factors, including lesion size, 5-year risk calculated by the BCSC risk calculator, and age at menarche. In the analysis of the overall cohort, both univariate and multivariate regression analyses confirmed that menopausal status, family history of breast cancer, age, lesion size, and GTC category are independent risk factors for breast cancer. Further analysis of GTC showed that women in the higher GTC (C/D) had a significantly higher risk of breast cancer compared to those in the low GTC (A). A significant dose-response relationship was observed, where the risk increased progressively with rising GTC (P < 0.001). The sonographic features specific to high GTC (C/D) may reflect underlying structural changes in breast parenchyma, such as ductal architectural distortion, stromal fibrosis, and collagen deposition. These pathological changes may be associated with an elevated risk of breast carcinogenesis.
Kim et al. [31], based on prior experimental studies in BRCA1-mutant mouse models and pathological assessments of 233 women who underwent GTC assessment, concluded that ultrasound-based GTC shows an inverse correlation with the degree of lobular involution in normal background parenchyma. Our study further confirms this finding through subgroup analysis.The results indicate that menopausal status significantly influences the contribution of risk factors. Specifically, in the postmenopausal subgroup, a higher GTC (D vs. A: OR = 7.38) emerged as a predominant risk factor, whereas the independent effect of age was weakened in the regression analysis for this subgroup. This shift suggests a potential close association with the altered endocrine environment following menopause [32], which further reinforces the correlation between GTC and lobular involution. Notably, in postmenopausal women, less extensive lobular involution is associated with a higher GTC, which in turn correlates with elevated breast cancer risk.In the premenopausal subgroup of our cohort, a positive family history of breast or ovarian cancer and larger lesion size (Category 4: >5 cm) were found to be the predominant risk factors. In contrast, GTC classification lacked statistically significant predictive value in this specific population.The predictive efficacy of GTC classification was significantly higher in postmenopausal women than in premenopausal women in our study.This discrepancy may be attributed to the enhanced stability of imaging features resulting from postmenopausal degenerative changes in breast tissue [33]. Thus, additional research into the application and significance of GTC classification in premenopausal women is needed.
Further analysis that compares GTC with other risk factors underscores its distinct clinical value—particularly its independent association with lesion size. Within the postmenopausal cohort of our study, both a high GTC and larger lesion size exhibited strong, independent predictive value for breast cancer. Based on these findings, key clinical implications can be drawn: GTC classification can serve as an imaging-based auxiliary tool for risk stratification in postmenopausal women. Specifically, for individuals presenting with larger lesions (> 2 cm) and GTC C or D, intensified clinical surveillance and monitoring are strongly recommended. Additionally, we observed that a history of benign breast biopsy was associated with a significantly reduced risk of breast cancer, indicating a protective effect. This finding also provides a novel perspective for optimizing breast cancer screening strategies.
Our study incorporated 5-year breast cancer risk estimates from the BCSC risk model into the analysis. The results showed that the median BCSC 5-year risk was significantly higher in the malignant group than in the benign group, and the malignant group exhibited a wider distribution range of this risk estimate. Consistent with prior studies, the BCSC model showed considerable value in predicting breast cancer risk. Notably, while the BCSC 5-year risk model showed significance in univariate analysis of the overall cohort, its predictive performance was limited in subgroup analyses stratified by menopausal status (all P > 0.05). Sprague et al. [34], utilizing the BCSC 5-year breast cancer risk prediction model in a study of women with dense breasts, found that women classified as high-risk had significantly higher cancer detection rates and positive predictive values for biopsy after ultrasound screening. This critical finding provides a rationale for future studies to integrate GTC classification as a novel variable into the BCSC risk calculator, which should aim to evaluate its potential to improve predictive accuracy for breast cancer risk in women with dense breast.
Furthermore, this study used ABUS for image acquisition and GTC assessment. This choice was predicated on the limitations of HHUS, which only yields single, static images dependent on the operator. Variability in scanning techniques and anatomical coverage across operators can result in substantial inconsistencies in image quality—and thus in GTC assessment. By contrast, ABUS offers three major advantages: a standardized scanning protocol, coronal plane imaging, and complete breast coverage. These features provide a comprehensive, reproducible dataset of the entire breast volume, supporting retrospective review and generating more quantitative information. ABUS effectively addresses the operator dependency and limited field of view inherent to HHUS, thereby enabling more standardized and reliable GTC assessment. In a prior prospective study by Moon et al. [29], which included 38 women with dense breasts, 11 physicians conducted handheld breast ultrasound examinations on all participants and completed GTC assessment. The results showed moderate inter-observer agreement (Kappa = 0.41, 0.52). In our study, we conducted inter-observer agreement analysis in 414 women with dense breasts (ACR BI-RADS C/D). For the GTC assessment performed using ABUS, “almost perfect agreement” was achieved (weighted k = 0.810; ICC = 0.888)—further confirming that ABUS significantly improves the standardization of GTC assessment. Additionally, our research team has previously developed a ResNet101 artificial intelligence (AI) model to assist with GTC classification [21], suggesting that AI may improve the accuracy of physician-led GTC assessments. Future work could leverage AI for the quantitative measurement of GTC, enabling more precise and reproducible breast cancer risk stratification.
Simultaneously, this study has several limitations that warrant acknowledgment. First, its retrospective single-center design inherently introduces a potential risk of selection bias. Second, this study exclusively enrolled women with breast MD classified as ACR BI-RADS Categories C or D. Consequently, the findings cannot be generalized to women with predominantly fatty or scattered fibroglandular breast densities (ACR BI-RADS Categories A and B). Future studies should include participants across the full spectrum of breast density categories. Third, our study did not directly compare the consistency of GTC classification between ABUS and HHUS. Thus, future research should prioritize large-scale, multi-center prospective cohort studies. Leveraging deep learning-based AI models may further enhance the objectivity and reliability of GTC assessment. Concurrently, it is essential to actively explore the correlations between GTC classification and other risk factors and to compare its performance with established mammography-based tools for quantifying MD, thereby establishing a more comprehensive, precise, and reliable basis for breast cancer risk assessment.

Conclusions
This study indicates that GTC is a significant and independent risk factor for breast cancer among women with heterogeneously or extremely dense breasts, particularly in postmenopausal populations. Moving forward, research should prioritize two key objectives: first, to clarify the determinants of GTC assessment and examine its dynamic evolution during malignant progression; and second, integrate GTC into established breast cancer risk prediction models.