Predicting axillary lymph node metastasis in breast cancer via quantitative analysis of VueBox ultrasound contrast-enhanced imaging combined with multimodal feature fusion

Introduction

Breast cancer remains the most common malignancy among women worldwide, with incidence rates continuing to rise in recent years (1). The disease typically originates in the epithelial tissue of breast ducts or lobules and can metastasize via lymphatic or hematogenous pathways. Axillary lymph nodes (ALNs) serve as the primary site for breast cancer metastasis (2), and their metastatic status is critical for clinical staging and treatment planning (3).

Traditional preoperative axillary ultrasound has limited sensitivity, with an average accuracy rate of approximately 55% (range, 24–94%) for detecting lymph node metastasis (4). Computed tomography (CT) and magnetic resonance imaging (MRI) facilitate the noninvasive assessment of ALN status in patients with breast cancer. However, CT is less frequently used for preoperative evaluation due to its lower specificity for lymph node assessment and limitations in evaluating primary breast tumors (5). MRI entails certain limitations, including high technical demands, lengthy scan times, and high cost. This underscores the urgent need for the development of more precise predictive tools. Multimodal ultrasound offers an effective solution to this challenge and has been widely adopted in recent years for preoperative tumor prediction. The integration of two-dimensional ultrasound, Doppler imaging, elastography, and contrast-enhanced ultrasound (CEUS) can provide complementary information on tumor morphology, stiffness, vascular density, and perfusion (6,7). For instance, shear wave elastography (SWE) quantifies tumor stiffness (8), while CEUS provides real-time visualization of tumor angiogenesis and perfusion (9). Despite these advances, these techniques heavily rely on operator subjectivity, potentially leading to unnecessary ALN dissection. However, quantitative analysis of CEUS with VueBox software automatically calculates perfusion parameters and generates color perfusion maps. This approach more directly and objectively quantifies subtle enhancement differences, thereby enhancing the accuracy of preoperative prediction of axillary lymph node metastasis (ALNM) in patients with breast cancer (10). This approach is based on time-intensity curve (TIC) quantitative CEUS analysis and has been reported in previous clinical studies of CEUS perfusion (10-12).

In recent years, multiple studies have integrated various imaging modalities or combined pathological features to construct predictive models, achieving precise preoperative prediction of ALNM in breast cancer and providing critical support for clinical decision-making. Xu et al. (5) developed a machine learning model based on contrast-enhanced mammography (CEM) and the Breast Imaging Reporting and Data System (BI-RADS) that accurately predicts preoperative ALNM in patients with breast cancer, providing an important reference for clinical decision-making. Wang et al. (13) integrated ultrasound and pathological features to predict sentinel lymph node metastasis using a nomogram, aiding decision-making regarding ALN dissection and adjuvant therapy. However, studies combining quantitative analysis of CEUS with multimodal ultrasound features to predict ALNM have been relatively scarce in recent years. Building upon this foundation, this study integrated VueBox ultrasound contrast quantitative analysis with multimodal ultrasound features and incorporated key serological and immunohistochemical markers to construct and validate a model for the preoperative prediction ALNM in patients with breast cancer. We present this article in accordance with the TRIPOD reporting checklist (available at https://qims.amegroups.com/article/view/10.21037/qims-2025-aw-2359/rc).

Methods

Study population

This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments and was approved by the Ethics Committee of Zhangzhou Affiliated Hospital of Fujian Medical University (No. 2025LWB358). The requirement for informed consent was waived due to the retrospective nature of the analysis.

A retrospective analysis was conducted on 82 female patients with breast cancer (with 84 breast lesions) diagnosed surgically at Zhangzhou Affiliated Hospital of Fujian Medical University, from January 2023 to July 2025. Patients were categorized into an ALNM-positive group (n=36) and an ALNM-negative group (n=48) based on postoperative pathological findings. The inclusion criteria were as follows: (I) age ≥18 years; (II) completion of breast cancer surgery, with pathological examination confirming breast cancer; and (III) complete preoperative ultrasound data. Meanwhile, the exclusion criteria were as follows: (I) poor ultrasound image quality; (II) lactation or pregnancy; (III) prior adjuvant chemotherapy, endocrine therapy, or radiotherapy; (IV) history of other malignancies; and (V) concomitant psychiatric or cognitive impairment.

Instruments and methods

Instrumentation and ultrasound examination methods

All ultrasound examinations were performed preoperatively as part of routine clinical care at our institution, and the imaging protocol was not influenced by this present retrospective study. Ultrasound examinations were performed with the ACUSON Sequoia ultrasound diagnostic system (Siemens Healthineers, Erlangen, Germany) equipped with an L10-4 linear array probe (frequency range 4–10 MHz). Patients were positioned supine with arms raised to fully expose the breast and axillary regions. For conventional two-dimensional ultrasound, multi-angle, multiplane scanning was employed to localize lesions, with two orthogonally oriented images of significant characteristics stored. For color Doppler flow imaging (CDFI), optimized sampling frame and velocity scale were used to save the image with the richest blood flow signal. For SWE, the sampling frame was appropriately positioned, and a sufficient coupling agent was applied to avoid probe compression. The patient was instructed to briefly hold their breath, and the image was frozen and saved after stabilization. For CEUS, the slice with the most prominent perfusion was selected for contrast enhancement. A total of 4.8 mL of SonoVue (Bracco Imaging S.p.A., Milan, Italy) microbubble suspension was administered via the cubital vein, which was followed by 5.0 mL of saline for flushing. Dynamic contrast-enhanced images were observed for 3 minutes and recorded. All CEUS examinations were completed before surgical intervention, and no postoperative imaging data were included in this study.

Ultrasound image analysis and parameter extraction

All preoperative ultrasound images were retrospectively analyzed by two senior physicians with over 10 years of breast ultrasound diagnostic experience under a double-blind method. Discrepancies were resolved through discussion. The evaluation included several aspects. For conventional two-dimensional ultrasound, lesions were evaluated based on the BI-RADS of the American College of Radiology (ACR) (14), including lesion morphology, orientation, margins, posterior echo characteristics, microcalcifications, and internal echoes. For CDFI, lesions were categorized into grade 0–I and grade II–III groups according to the Alder semiquantitative blood flow grading method (15). For SWE, the maximum Young modulus (Emax), mean Young modulus (Emean), and minimum Young modulus (Emin) were recorded, and the lesion-to-surrounding-normal-tissue hardness ratio (Eratio) was calculated. For qualitative CEUS analysis, feature assessment included enhancement margin, enhancement morphology, enhancement distribution, enhanced intensity, enhancement direction, enhancement area, perfusion defects, and perforating vessels. For the quantitative CEUS analysis method, dynamic contrast-enhanced images stored in Digital Imaging and Communications in Medicine (DICOM) were imported into VueBox software (Bracco Suisse SA, Geneva, Switzerland), a commercially available postprocessing tool for the quantitative analysis of CEUS perfusion. The image frame in which the lesion reached peak contrast enhancement was selected, and a large reference region encompassing the target lesion and adjacent normal breast tissue was included. Two regions of interest (ROIs) were manually delineated: ROI 1 was positioned within the lesion at the site of maximal enhancement, and ROI 2 was positioned in adjacent normal glandular tissue at the same imaging depth (Figure 1). After multiple delineation verifications, the optimal data set with a goodness of fit exceeding 75% was selected. The software automatically generated TICs and extracted quantitative perfusion parameters, including area under the curve (AUC), mean time to intensity (mTTI), time to peak intensity (TTP), and wash-in rate (WiR).

Figure 1 Quantitative contrast-enhanced ultrasound analysis of a breast cancer lesion. (A) Image from VueBox quantitative analysis. The yellow circle indicates the ROI of maximal enhancement in the lesion (ROI 1), and the red circle indicates the reference ROI in normal tissue (ROI 2). (B) Quantitative perfusion curves derived from contrast-enhanced ultrasound imaging. The lesion ROI curve markedly exceeds the reference ROI curve, demonstrating a rapid rise-and-fall pattern with high enhancement, with the peak intensity being reached at 5.2 s. ROI, region of interest.

General data collection

Basic clinical information, including patient age and menopausal status, was collected. Preoperative serum and complete blood count parameters were obtained from peripheral blood samples, and the neutrophil-to-lymphocyte ratio (NLR), platelet-to-lymphocyte ratio (PLR), and lymphocyte-to-monocyte ratio (LMR) were calculated. Pathological type and immunohistochemical markers [estrogen receptor (ER), progesterone receptor (PR), human epidermal growth factor receptor 2 (HER2), tumor protein p53 (P53), epithelial cadherin (E-cadherin)] were evaluated on tissue specimens. According to the National Comprehensive Cancer Network (NCCN) guidelines (16), ER and PR positivity were defined as ≥1% nuclear staining; HER2 status was classified as negative (score 0/1+), positive (score 3+), or equivocal (score 2+, as confirmed by fluorescence in situ hybridization). P53 and E-cadherin expression were categorized as positive or negative based on standard staining patterns.

Statistical analysis

All statistical analyses were performed with R 4.5.0 (The R Foundation for Statistical Computing, Vienna, Austria). The significance level was set at α=0.05, with P<0.05 indicating a statistically significant difference. Count data are expressed as counts and percentages. Differences in rates between groups were compared with the chi-squared test (χ² test). Normally distributed continuous data are presented as the mean ± standard deviation (x¯±s) and were analyzed with the t-test. Nonnormally distributed continuous data are expressed as the median and interquartile range, and intergroup comparisons were performed via the Mann-Whitney U test. Binary logistic regression analysis was employed to determine the associations between variables and ALNM in patients with breast cancer. Potentially relevant variables were first screened via univariate logistic regression. Variables with P<0.05 in univariate analysis were then included in multivariate logistic regression. Based on the selected independent predictors, a risk chart prediction model for ALNM was constructed with the rms package in R (17). Model performance evaluation was conducted across multiple dimensions. Specifically, discriminative ability was assessed through the plotting of receiver operating characteristic (ROC) curves and calculation of the AUC. Calibration and goodness of fit were evaluated by the plotting of calibration curves and the application of the Hosmer-Lemeshow test (18). Clinical utility was analyzed via decision curve analysis (DCA) plots and the calculation of the net benefit at different threshold probabilities. Internal model validation was performed with the bootstrap method (1,000 repeated samples), with the validated AUC and its 95% confidence interval (CI) being calculated to assess model stability and generalizability.

Results

Baseline clinicopathologic characteristics

The baseline clinicopathologic characteristics of the 84 patients are summarized in Table 1, including 48 ALNM-negative and 36 ALNM-positive cases. The mean age of the overall cohort was 53.4±10.8 years, with comparable age distributions between the two groups. More than half of the patients were postmenopausal (55.95%).

Immunohistochemical analysis demonstrated high expression rates of ER (79.76%), PR (70.24%), P53 (82.14%), and E-cadherin (96.43%) in the overall population, while HER2 positivity was observed in 52.38% of cases. The majority of tumors shared the same pathological type across both groups. Detailed distributions of these variables stratified by ALN status are shown in Table 1.

Comparison of multimodal features between the ALNM and non-ALNM groups

Univariate analysis revealed no statistically significant differences between the metastatic and nonmetastatic groups in terms of age, menopausal status NLR, PLR, and LMR. Compared with the nonmetastatic group, the metastatic group showed statistically significant differences in Emax, postenhancement extension, microcalcifications, maximum diameter >2 cm, and TTP (all P values <0.05; Table 2).

Logistic regression analysis for identifying independent predictors of ALNM in patients with breast cancer

Incorporating the above variables into a multivariate logistic regression model revealed that Emax, enlarged enhancement scope, microcalcifications, and TTP were all independent risk factors for ALNM in patients with breast cancer (all P values <0.05; Table 3).

Construction and validation of the nomogram model

Based on the independent factors identified by logistic regression analysis, a nomogram was developed to predict ALNM in patients with breast cancer. The clinical application of this nomogram is illustrated in Figure 2A-2E.

Figure 2 Multimodal ultrasound features of a breast cancer lesion and construction of the nomogram. (A) The two-dimensional ultrasound image showed no microcalcifications in the lesion. (B) The SWE image had an Emax of 300 kPa and Emin of 161.9 kPa. (C) Postcontrast enhancement showed lesion enlargement. (D) Contrast-enhanced quantitative perfusion curve demonstrated a rapid rise in high enhancement, with the peak intensity being reached at 8.7 seconds. (E) Nomogram developed for predicting axillary lymph node metastasis in breast cancer. *, P<0.05; **, P<0.01. Emax, maximum Young modulus; Emin, minimum Young modulus; SWE, shear wave elastography; TTP, time to peak intensity.

In the nomogram, the total score ranged from 210 to 375 points, corresponding to a predicted probability of ALNM ranging from 0.2% to 99%. As shown in Figure 2E, for one patient, the total score nomogram-calculated score was 331 points (61 points + 100 points + 73 points + 97 points), which corresponded to an estimated ALNM probability of approximately 81.9%.

ROC curve analysis indicated that the nomogram achieved an AUC of 0.926 (95% CI: 0.872–0.980), with a sensitivity of 97% and a specificity of 75% for predicting ALNM. Internal validation via bootstrap resampling yielded a concordance index (C-index) of 0.926 (95% CI: 0.865–0.974).

The calibration curve demonstrated good agreement between the predicted probabilities and the observed outcomes. The Hosmer–Lemeshow test further confirmed good model fit (χ²=4.19; P=0.84). Meanwhile, DCA indicated that the model provided a higher net benefit compared with the treat-all or treat-none strategies across a range of reasonable threshold probabilities (Figures 3-5).

Figure 3 Receiver operating characteristic curve for predicting axillary lymph node metastasis. AUC, area under the curve; CI, confidence interval.

Figure 4 Calibration curve of the predictive model for axillary lymph node metastasis.

Figure 5 Decision curve analysis for the predictive model in predicting axillary lymph node metastasis.

Discussion

An increasing number of studies have begun employing radiomics to assess ALNM. Guo et al. (19) reported that a multimodal model combining radiomics and deep learning outperformed either approach alone, while Li et al. (20) developed a multicenter multimodal ultrasound-based radiomics model with promising performance across training and validation cohorts. Although the clinical translation of radiomics has historically faced challenges regarding reproducibility and generalizability (21-23), the introduction of standardization initiatives such as the Image Biomarker Standardisation Initiative (IBSI) (22) and reporting guidelines such as Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis (TRIPOD) (24,25) has substantially improved methodological consistency and reporting transparency. However, despite these methodological advancements, challenges related to model interpretability remain. Despite conventional radiomics being generally considered relatively interpretable, the increasing use of complex deep learning-based frameworks has prompted ongoing concerns regarding the “black box” effect, which may hinder clinical trust and adoption (26). Therefore, distinct from these complex black-box approaches, there is a need for robust yet interpretable tools. In this study, we developed and validated a multimodal nomogram integrating quantitative CEUS parameters, conventional ultrasound features, and serological and pathological biomarkers for preoperative ALNM prediction, with an AUC of 0.926 and favorable calibration.

Previous studies have applied TIC-based quantitative CEUS analysis using VueBox in breast imaging. For example, Jung et al. (12) applied VueBox to investigate noncystic breast lesions. They found that the peak enhancement and AUC of CEUS perfusion parameters could effectively assess the risk of malignant breast lesions, thereby downgrading BI-RADS category 4 lesions. Li et al. (10) applied this software for the preoperative prediction of the histopathological features of breast cancer. Their results included differences in CEUS qualitative characteristics and quantitative parameters among breast cancers with varying histological grades and molecular subtypes. CEUS can noninvasively predict histological features of breast cancer. However, evidence regarding the predictive value of TIC-derived quantitative parameters for ALNM remains limited. Our study demonstrated that TTP obtained through VueBox is an independent risk factor for ALNM. Mechanistic analysis suggests that tumor cells with high metabolic activity promote disorganized neovascularization, accelerating blood flow and creating a hyperperfusion state. This microenvironment facilitates nutrient supply and tumor cell migration, thereby promoting lymphatic metastasis. The shortened TTP observed in this study reflects the rapid contrast agent filling and peak accumulation in the tumor region, supporting the presence of the aforementioned highly invasive pathophysiological mechanism.

In addition, SWE technology provides objective and reproducible diagnostic information by quantitatively measuring the Young modulus of tissue, which plays a crucial role in enhancing the accuracy of breast cancer diagnosis (27). Our study found that elevated Emax values positively correlated with the risk of ALNM, with the underlying mechanism involving increased tissue stiffness at tumor margins due to necrosis, calcification, and proliferative fibrous connective tissue reactions (28). In vitro studies indicate that elevated matrix stiffness correlates with the pathogenesis of breast cancer invasion and metastasis (29). Therefore, elastographic parameters such as Emax can serve as quantitative indicators reflecting tumor invasive potential. This finding aligns with conclusions from previous studies (30,31).

Beyond imaging ultrasound features, tumor biology remains critical. Active tumor cell proliferation creates relative oxygen and nutrient deprivation, fostering a hostile tumor microenvironment. In this setting, suppressed metabolic activity stimulates abnormal calcium secretion and deposition, forming microcalcifications. Concurrently, this hypoxic and nutritional stress itself drives tumor cells to acquire enhanced invasive and metastatic capabilities for survival (32). The findings of our study indicate that microcalcifications in primary breast tumors are associated with the risk of ALNM, which is consistent with the aforementioned theory. Furthermore, during rapid tumor proliferation, necrosis often occurs due to relative blood supply insufficiency. The increased number of compensatory neovascularization around necrotic areas leads to expanded contrast agent penetration, resulting in enlarged enhancement areas. However, these necrotic areas are not static lesions. They release growth factors and inflammatory mediators, initiating a series of pathophysiological processes that promote the formation of new lymphatic vessels, thereby providing a “channel” for ALNM. This also makes the enlarged contrast enhancement area an indirect potential reference indicator for assessing metastasis risk. Xiong et al. (33) found that patients with tumors exceeding 2 cm in diameter were more likely to develop metastatic lesions, yet our findings did not support this conclusion. The reason for this discrepancy may lie in the complex relationship between tumor size and ALNM. This association varies depending on the molecular subtype of the tumor, its genetic background, and the expression levels of molecules regulating tumor growth and lymphatic metastasis (34).

We constructed an ALNM prediction model based on VueBox ultrasound contrast quantitative analysis and multimodal feature fusion, integrating imaging with clinical pathology to enhance clinical applicability. Ultrasound contrast quantitative analysis reduces subjective error, while constructing a nomogram improves the interpretability.

This study involved certain limitations that should be acknowledged. First, the single-center retrospective design might have introduced selection bias, thereby limiting the generalizability of the findings. External validation in a multicenter prospective cohort is required to further assess the robustness and potential clinical applicability of the proposed model. Second, although the sample size was sufficient for hypothesis generation and initial model development, the relatively small cohort limited further improvement in model stability and the identification of additional predictive features. Larger-scale studies are therefore warranted to address these limitations. Third, the model did not incorporate advanced imaging modalities such as MRI, the inclusion of which may provide incremental predictive value and further enhance model performance. Fourth, although prediction of pathological nodal burden—particularly the presence of ≥3 positive lymph nodes—is clinically important for guiding decisions regarding ALN dissection and regional nodal irradiation, the limited number of ALNM-positive patients in our cohort, combined with the inherent limitations of preoperative ultrasound in quantifying the total extent of axillary disease, precluded reliable nodal burden stratification. Future studies with larger, dedicated cohorts are needed to address this question.

Conclusions

This study established a predictive model for ALNM from breast cancer based on TIC analysis via VueBox, integrating imaging indicators such as Emax, TTP, enhanced lesion spread, and microcalcifications. This work contributes to the ongoing efforts in developing noninvasive preoperative models for assessing the presence of ALNM in patients with breast cancer. The aim is to support future research on treatment planning while potentially reducing the morbidity associated with surgical axillary evaluation. Future studies with larger and more balanced cohorts will focus on extending this framework toward the estimation of pathological nodal burden, which will address clinically relevant decision-making needs.

Acknowledgments

We would like to thank all the patients and their families who participated in this study.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://qims.amegroups.com/article/view/10.21037/qims-2025-aw-2359/rc

Data Sharing Statement: Available at https://qims.amegroups.com/article/view/10.21037/qims-2025-aw-2359/dss

Funding: This study was funded by the Medical Innovation Project of Fujian Provincial Health Commission (No. 2022CXA055), and the Natural Science Foundation of Fujian Province (No. 2023J011813).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://qims.amegroups.com/article/view/10.21037/qims-2025-aw-2359/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Ethics Committee of Zhangzhou Affiliated Hospital of Fujian Medical University (No. 2025LWB358). Informed consent was waived in this retrospective study.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Siegel RL, Giaquinto AN, Jemal A. Cancer statistics, 2024. CA Cancer J Clin 2024;74:12-49. [Crossref] [PubMed]
2. Zha HL, Zong M, Liu XP, Pan JZ, Wang H, Gong HY, Xia TS, Liu XA, Li CY. Preoperative ultrasound-based radiomics score can improve the accuracy of the Memorial Sloan Kettering Cancer Center nomogram for predicting sentinel lymph node metastasis in breast cancer. Eur J Radiol 2021;135:109512. [Crossref] [PubMed]
3. Sun Q, Lin X, Zhao Y, Li L, Yan K, Liang D, Sun D, Li ZC. Deep learning vs. radiomics for predicting axillary lymph node metastasis of breast cancer using ultrasound images: don’t forget the peritumoral region. Front Oncol 2020;10:53. [Crossref] [PubMed]
4. Le Boulc'h M, Gilhodes J, Steinmeyer Z, Molière S, Mathelin C. Pretherapeutic Imaging for Axillary Staging in Breast Cancer: A Systematic Review and Meta-Analysis of Ultrasound, MRI and FDG PET. J Clin Med 2021;10:1543. [Crossref] [PubMed]
5. Xu W, Zheng B, Wen C, Zeng H, Wang S, He Z, Liao X, Chen W, Li Y, Qin G. Enhancing Specificity in Predicting Axillary Lymph Node Metastasis in Breast Cancer through an Interpretable Machine Learning Model with CEM and Ultrasound Integration. Technol Cancer Res Treat 2025;24:15330338251334735. [Crossref] [PubMed]
6. Lugano R, Ramachandran M, Dimberg A. Tumor angiogenesis: causes, consequences, challenges and opportunities. Cell Mol Life Sci 2020;77:1745-70. [Crossref] [PubMed]
7. Lim HF, Sharma A, Gallagher C, Hall P. Value of ultrasound in assessing response to neoadjuvant chemotherapy in breast cancer. Clin Radiol 2023;78:912-8. [Crossref] [PubMed]
8. Zhang X, Wang L, Feng N, Ni T, Tang W. Reassessing the Value of Contrast-Enhanced Ultrasonography in Differential Diagnosis of Cervical Tuberculous Lymphadenitis and Lymph Node Metastasis of Papillary Thyroid Carcinoma. Front Oncol 2021;11:694449. [Crossref] [PubMed]
9. Huang JX, Liu FT, Tan YT, Wang XY, Huang JH, Lin SY, Huang GL, Zhang YT, Pei XQ. Enhancing detection of high-level axillary lymph node metastasis after neoadjuvant therapy in breast cancer patients with nodal involvement: a combined approach of axilla ultrasound and breast elastography. Radiol Med 2025;130:121-31. [Crossref] [PubMed]
10. Li LL, Su QL, Deng YX, Guo WW, Lun HM, Hu Q. Contrast-enhanced ultrasound for the preoperative prediction of pathological characteristics in breast cancer. Front Oncol 2024;14:1320714. [Crossref] [PubMed]
11. Huang Y, Wang Y, Liu L, Zhu L, Qiu Y, Zuo D, Lu X, Dong Y, Jung EM, Wang W. VueBox® perfusion analysis of dynamic contrast enhanced ultrasound provides added value in the diagnosis of small thyroid nodules. Clin Hemorheol Microcirc 2023;83:409-20. [Crossref] [PubMed]
12. Jung EM, Jung F, Stroszczynski C, Wiesinger I. Quantification of dynamic contrast-enhanced ultrasound (CEUS) in non-cystic breast lesions using external perfusion software. Sci Rep 2021;11:17677. [Crossref] [PubMed]
13. Wang XF, Zhang GC, Zuo ZC, Zhu QL, Liu ZZ, Wu SF, Li JX, Du JH, Yan CL, Ma XY, Shi Y, Shi H, Zhou YD, Mao F, Lin Y, Shen SJ, Zhang XH, Sun Q. A novel nomogram for the preoperative prediction of sentinel lymph node metastasis in breast cancer. Cancer Med 2023;12:7039-50. [Crossref] [PubMed]
14. Irshad A, Leddy R, Ackerman S, Cluver A, Pavic D, Abid A, Lewis MC. Effects of Changes in BI-RADS Density Assessment Guidelines (Fourth Versus Fifth Edition) on Breast Density Assessment: Intra- and Interreader Agreements and Density Distribution. AJR Am J Roentgenol 2016;207:1366-71.
15. Adler DD, Carson PL, Rubin JM, Quinn-Reid D. Doppler ultrasound color flow imaging in the study of breast cancer: preliminary findings. Ultrasound Med Biol 1990;16:553-9. [Crossref] [PubMed]
16. Gradishar WJ, Moran MS, Abraham J, Abramson V, Aft R, Agnese D, et al. Breast Cancer, Version 3.2024, NCCN Clinical Practice Guidelines in Oncology. J Natl Compr Canc Netw 2024;22:331-57. [Crossref] [PubMed]
17. Harrell FE Jr. Regression Modeling Strategies: With Applications to Linear Models, Logistic Regression, and Survival Analysis. 2nd ed. Cham: Springer; 2015.
18. Hosmer DW, Lemeshow S, Sturdivant RX. Applied Logistic Regression. 3rd ed. New York: Wiley; 2013.
19. Guo F, Sun S, Deng X, Wang Y, Yao W, Yue P, Wu S, Yan J, Zhang X, Zhang Y. Predicting axillary lymph node metastasis in breast cancer using a multimodal radiomics and deep learning model. Front Immunol 2024;15:1482020. [Crossref] [PubMed]
20. Li SY, Li YM, Fang YQ, Jin ZY, Li JK, Zou XM, Huang SS, Niu RL, Fu NQ, Shao YH, Gong XT, Li MR, Wang W, Wang ZL. Contrast-enhanced ultrasound radiomics model for predicting axillary lymph node metastasis and prognosis in breast cancer: a multicenter study. BMC Cancer 2025;25:1315. [Crossref] [PubMed]
21. Teng X, Wang Y, Nicol AJ, Ching JCF, Wong EKY, Lam KTC, Zhang J, Lee SW, Cai J. Enhancing the Clinical Utility of Radiomics: Addressing the Challenges of Repeatability and Reproducibility in CT and MRI. Diagnostics (Basel) 2024;14:1835. [Crossref] [PubMed]
22. van Timmeren JE, Cester D, Tanadini-Lang S, Alkadhi H, Baessler B. Radiomics in medical imaging-"how-to" guide and critical reflection. Insights Imaging 2020;11:91. [Crossref] [PubMed]
23. Zwanenburg A. Radiomics in nuclear medicine: robustness, reproducibility, standardization, and how to avoid data analysis traps and replication crisis. Eur J Nucl Med Mol Imaging 2019;46:2638-55. [Crossref] [PubMed]
24. Collins GS, Moons KGM, Dhiman P, Riley RD, Beam AL, Van Calster B, et al. TRIPOD+AI statement: updated guidance for reporting clinical prediction models that use regression or machine learning methods. BMJ 2024;385:e078378. [Crossref] [PubMed]
25. Moons KG, Altman DG, Reitsma JB, Ioannidis JP, Macaskill P, Steyerberg EW, Vickers AJ, Ransohoff DF, Collins GS. Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis (TRIPOD): explanation and elaboration. Ann Intern Med 2015;162:W1-73. [Crossref] [PubMed]
26. Muhammad D, Bendechache M. Unveiling the black box: A systematic review of Explainable Artificial Intelligence in medical image analysis. Comput Struct Biotechnol J 2024;24:542-60. [Crossref] [PubMed]
27. Chen X, Yu H, Wei N, Ozcan BB, An G, Wu Q, Wang N. Diagnostic performance of contrast-enhanced ultrasound combined with shear wave elastography in differentiating benign from malignant breast lesions: a systematic review and meta-analysis. Gland Surg 2023;12:1610-23. [Crossref] [PubMed]
28. Zong Q, Deng J, Ge W, Chen J, Xu D. Establishment of Simple Nomograms for Predicting Axillary Lymph Node Involvement in Early Breast Cancer. Cancer Manag Res 2020;12:2025-35. [Crossref] [PubMed]
29. Luo C, Lu L, Zhang W, Li X, Zhou P, Ran Z. The Value of Shear Wave Elastography in the Diagnosis of Breast Cancer Axillary Lymph Node Metastasis and Its Correlation With Molecular Classification of Breast Masses. Front Oncol 2022;12:846568. [Crossref] [PubMed]
30. Evans A, Rauchhaus P, Whelehan P, Thomson K, Purdie CA, Jordan LB, Michie CO, Thompson A, Vinnicombe S. Does shear wave ultrasound independently predict axillary lymph node metastasis in women with invasive breast cancer? Breast Cancer Res Treat 2014;143:153-7. [Crossref] [PubMed]
31. Zhao Q, Sun JW, Zhou H, Du LY, Wang XL, Tao L, Jiang ZP, Zhou XL. Pre-operative Conventional Ultrasound and Sonoelastography Evaluation for Predicting Axillary Lymph Node Metastasis in Patients with Malignant Breast Lesions. Ultrasound Med Biol 2018;44:2587-95. [Crossref] [PubMed]
32. Piddubnyi A, Kolomiiets O, Danilchenko S, Stepanenko A, Moskalenko Y, Moskalenko R. The Prospects of Using Structural Phase Analysis of Microcalcifications in Breast Cancer Diagnostics. Diagnostics (Basel) 2023;13:737. [Crossref] [PubMed]
33. Xiong J, Zuo W, Wu Y, Wang X, Li W, Wang Q, Zhou H, Xie M, Qin X. Ultrasonography and clinicopathological features of breast cancer in predicting axillary lymph node metastases. BMC Cancer 2022;22:1155. [Crossref] [PubMed]
34. Min SK, Lee SK, Woo J, Jung SM, Ryu JM, Yu J, Lee JE, Kim SW, Chae BJ, Nam SJ. Relation Between Tumor Size and Lymph Node Metastasis According to Subtypes of Breast Cancer. J Breast Cancer 2021;24:75-84. [Crossref] [PubMed]