Development and validation of a clinical-bimodal nomogram to predict the malignancy of pathologic nipple discharge

Introduction

Pathologic nipple discharge (PND) is one of the most common complaints in breast clinics worldwide (1). Although the majority of cases are benign in nature, approximately 5–33% of PND are malignant (2). Delayed diagnosis of malignant PND may increase the risk of progression to invasive cancer (3). Therefore, accurate differentiation between benign and malignant is critical for appropriate clinical management.

Currently, mammography (MG) is the standard first-line imaging modality for diagnosing PND (1). However, its diagnostic accuracy varies widely, with sensitivity ranging from 15% to 68%, and specificity ranging from 38% to 94% (4-7). Noticeably, breast ultrasonography (US) is increasingly utilized in the diagnosis of PND, as it can detect the abnormalities in the ducts and associated ductal changes that cannot be identified by MG (8-10).

Several studies have attempted to stratify malignancy risk in patients with PND based on imaging outcomes, core needle biopsy (CNB), or ductoscopy (6,11-14). However, these studies are limited by oversimplified imaging classifications, which affect the objectivity of the models, and restrict clinical applicability due to reliance on invasive procedures. To our knowledge, no prior studies have integrated specific imaging features with clinical variables to develop a practical, non-invasive, and reproducible risk-assessment tool capable of providing more consistent guidance, better reflecting real-world diagnostic practices, and enhancing risk stratification in patients with PND (15).

Thus, the aim of this study was to develop a nomogram incorporating clinical data and bimodal imaging parameters from MG and US to predict the probability of malignancy in patients with PND. This tool is intended to support clinicians in identifying patients at low risk of malignancy who may be suitable for clinical observation. We present this article in accordance with the TRIPOD reporting checklist (available at https://qims.amegroups.com/article/view/10.21037/qims-2025-701/rc).

Methods

Patients

The study was conducted at two different hospital campuses of the Sun Yat-sen Memorial Hospital, Sun Yat-sen University with its health system (hospital 1: the south hospital campus; hospital 2: the north hospital campus). All patients diagnosed with nipple discharge from January 2013 to December 2023 at either of the two hospitals were retrospectively enrolled in this study. The inclusion criteria were as follows: (I) female patients aged 18 years or older diagnosed with PND, defined by having at least one of the following features: spontaneous, unilateral, from a single duct orifice and blood secretion; (II) patients received both preoperative US and MG examinations; (III) all patients had pathological results obtained either through biopsy or surgical excision. Exclusion criteria included: (I) patients who were scheduled to receive biopsy or surgery of the target lesion before examination; (II) unavailable or incomplete image data; (III) those without completed clinical data.

This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Ethics Committee of Sun Yat-sen Memorial Hospital (No. SYSKY-2024-559-01). Due to its retrospective nature, the informed consents from patients were exempted.

Data collection

Demographic and clinical characteristics of the patients were extracted from the electronic medical database, including age (years), palpable or not, characteristics of the discharge (color of discharge, spontaneous or not, discharge from a single/multiple duct orifice).

Imaging acquisition and analysis

MG was conducted by Digital machine (Planmed Nuance Excel; Planmed, Helsinki, Finland). Regular craniocaudal and mediolateral oblique images were consistently acquired. Two experienced radiologists (H.W. and F.L.), specializing in breast imaging diagnosis for over 5 years, thoroughly examined all MG images to achieve the parameters according to the 2013 American College of Radiology Breast Imaging Reporting and Data System (ACR BI-RADS) criteria (16). If any disagreement, a third experienced radiologist (RG) for over 15 years was consulted to re-evaluate and confirm the imaging parameters. The radiologists were blinded to the histopathologic result. The MG parameters were as follows: (I) calcifications: no/benign-looking; suspicious; (II) architectural distortion: no; yes; (III) asymmetry: no; yes; (IV) mass: no/benign-looking; suspicious.

A high-frequency US machine (S2000/S1000 and 18L6 transducer with a center frequency of 15 MHz; Siemens Medical Solutions, Erlangen, Germany) was used to evaluate the entire breast. All US examinations were independently reviewed following the 2013 ACR BI-RADS criteria by three expert radiologists specializing in breast imaging diagnosis for over 5 years (J.M., X.J., X.G.), if any disagreement, a third experienced radiologist (Y.H.) for over 15 years was consulted to re-evaluate and confirm the imaging parameters. All radiologists were blinded to the histopathologic result. The image-based indicators for describing US were as follows: (I) mass: no/benign-looking; suspicious; (II) microcalcifications: no; yes; (III) dilated duct: no; yes; (IV) intraductal mass/debris: no; yes.

Pathologic diagnosis

At both hospitals, patients with PND are routinely recommended for biopsy. However, the final decision is generally made by patients and referring physicians. Breast biopsy procedures used either a core instrument with a 14- or 16-gauge needle, or a vacuum-assisted biopsy equipment with an 8- or 11-gauge. Patients with indeterminate or malignant histological results were recommended to undergo open breast surgery, while those with benign results are advised to imaging follow-up every three-six months for the two year. Patients with negative imaging findings underwent a selective duct excision. Patients were classified as malignant if their histopathologic results confirmed the presence of breast cancer. Atypical lesions, papilloma and phyllodes tumor results were defined as benign.

Statistical analysis

Baseline characteristics were compared between the two cohorts using the independent t-test, Pearson’s Chi-squared test or Fisher’s exact test. In training cohort, a comparison was made between the characteristics of patients in the benign and malignant groups. Mann-Whitney U test was employed for continuous variables, while Pearson’s Chi-squared test and Fisher’s exact test were utilized for categorical variables. Risk factors were identified by univariate analysis. Then, risk factors with P<0.05 selected by univariate analysis were put into a multivariate logistic regression analysis to identify the independent predictors of breast cancer.

The predictive models with independently MG and MG-US parameters were established respectively. The model underwent 1,000 times of the bootstrap resample approach for internal validation. This process was carried out to reduce the model’s over-fitting bias and improve the accuracy of its predictions. The discriminative capabilities of both models were measured using the area under the receiver operating characteristic (ROC) curve (AUC). The model’s calibration capacity was assessed using the calibration curve and the Hosmer-Leme show goodness-of-fit test. Decision curve analysis (DCA) was performed to evaluate the clinical usefulness of both models by measuring the net benefits across a variety of threshold probabilities. The optimal model was selected by comparing the AUCs of two models with De-long test, and the better model was visualized with nomograms. Additionally, the cohort was divided into small and large lesion subgroups to evaluate the diagnostic performance of the nomogram in these groups. Lesions <10 mm or with negative imaging findings were included in the small lesion subgroup, while lesions ≥10 mm were included in the large lesion subgroup.

SPSS 25 (IBM Corp., Armonk, NY, USA) and R version 3.1.0 (The R Foundation for Statistical Computing, Vienna, Austria) were used for the statistical analyses. P<0.05 was considered statistically significant.

Results

Patients characteristics

A total of 1,137 patients with PND who underwent imaging examinations and had pathological results were considered for inclusion. Ultimately, 280 patients (217 benign and 63 malignant lesions; median age 44 years, range, 23–84 years) from hospital 1 formed the training cohort. Ninety-seven patients (72 benign and 25 malignant lesions; median age 43 years, range, 23–80 years) from hospital 2 formed the validation cohort. The workflow is illustrated in Figure 1. Characteristics were compared between the training and validation cohorts. The cohorts were comparable for clinical characteristics (age, color of discharge, palpable mass, spontaneous or not), MG findings (suspicious calcifications, architectural distortion, asymmetry, suspicious mass, breast composition) and US findings (presence of microcalcifications, simple dilated duct, intraductal mass/debris, suspicious mass). Figure 2 demonstrates the selected imaging features of a case presenting with PND. No evidence of differences between the training and validation samples was found (P>0.05). The comparison results are given in Table 1.

Figure 1 The study workflow. Two hospitals: the Sun Yat-sen Memorial Hospital, Sun Yat-sen University with its health system (hospital 1: the south hospital campus; hospital 2: the north hospital campus).

Figure 2 A 30-year-old woman presented with persistent single-duct bloody nipple discharge from the right breast for 6 months. Core needle biopsy revealed ductal carcinoma in situ with upgrade to invasive ductal carcinoma at excision. (A) Right MLO and CC mammogram demonstrated segmentally distributed coarse heterogeneous calcifications in the central region; (B) two orthogonal ultrasound images demonstrated irregular hypoechoic mass with indistinct margins, which contains calcifications. CC, craniocauda; MLO, mediolateral oblique.

Analysis of clinical-imaging characteristics in the training cohorts

In the training cohorts, predictive indicators were determined by univariate analysis. Among these variables, age, color of discharge, palpable mass, suspicious calcifications on MG, mass on MG, calcifications on US, mass on US and intraductal mass/debris on US were associated with malignancy (Table 2). Based on a multivariate logistic regression analysis, several factors were identified as independent predictors of malignancy in patients with PND, including age [odds ratio (OR), 4.7; 95% confidence interval (CI): 2.3–9.7; P<0.001], color of discharge (OR: 3.0; 95% CI: 1.4–6.4; P=0.005), palpable mass (OR: 2.8; 95% CI: 1.3–6.4; P=0.011), suspicious calcifications on MG (OR: 3.4; 95% CI: 1.3–10.5; P=0.014), mass on US (OR: 6.3; 95% CI: 2.5–15.6; P<0.001), and intraductal mass/debris on US (OR: 8.6; 95% CI: 3.3–22.7; P<0.001). The ORs of the indicators for malignancy are shown in Table 3.

Comparison of MG model versus MG-US model

The predictive models were developed based on the multivariate analysis results. For MG model, four independent predictors were introduced to predict the likelihood of malignancy (age, color of discharge, palpable mass, suspicious calcifications on MG). For MG-US model, six independent predictors were included to predict the likelihood of malignancy (age, color of discharge, palpable mass, suspicious calcifications on MG, mass on US and intraductal mass/debris on US). The AUC for the MG-US model (0.851, 95% CI: 0.802–0.901) was significantly higher than that of the MG model (0.789, 95% CI: 0.729–0.850) (P=0.014) (Figure 3A). Both models were well calibrated, the predicted probabilities were close to the observed probabilities (Figure 3B,3C). The DCAs of the two predictive models are shown in Figure 3D. With respect to their clinical utility, DCA suggested that the MG-US model was superior to the MG model across a wider range of threshold probabilities.

Figure 3 ROC, calibration, and decision curves for the MG and MG-US models. (A) ROC curves for MG and MG-US models; (B) calibration curve for the MG model; (C) calibration curve for the MG-US model; (D) decision curve analysis was conducted for both the MG model and the MGUS model. AUC, area under the receiver operating characteristic curve; CI, confidence interval; MG, mammography; ROC, receiver operating characteristic; US, ultrasonography.

Development of the clinical-bimodal nomogram

The clinical-bimodal nomogram, which integrates clinical, MG, and US findings, was developed because its discriminative ability is higher than that of the MG model alone (Figure 4). The nomogram identified intraductal mass/debris on US and suspicious mass on US as the most significant factors contributing to prediction, followed by age and suspicious calcifications on MG. Other clinical features (i.e., palpable mass, color of discharge) had a moderate effect on prediction.

Figure 4 The clinical-bimodal nomogram constructed based on the MG-US model for evaluating the risk of malignancy in patients with pathological nipple discharge using age, palpable mass, color of discharge, suspicious calcifications on mammography, suspicious mass on ultrasound and intraductal mass/debris on ultrasound. The sum of the scores is located on the Total point axis and corresponds to the malignancy risk. MG, mammography; US, ultrasonography.

Validation of the nomograms

The nomogram showed good discrimination ability in the internal validation, which was conducted using bootstrap method with 1,000 resamples [AUC, 0.859 (95% CI: 0.803–0.907)] (Figure S1). Similar AUC were achieved at the independent validation cohort made up of 97 women [AUC, 0.879 (95% CI: 0.807–0.951)] (Figure 5A). In the subgroup analysis, small and large lesion subgroups were evaluated for the nomogram’s diagnostic performance. The small lesion subgroup, consisting of 184 patients, exhibited an AUC of 0.858 (95% CI: 0.793–0.923), while the large lesion subgroup, consisting of 96 patients, had an AUC of 0.845 (95% CI: 0.766–0.924). The calibration plots showed excellent agreement, and the Hosmer-Lemeshow test statistic was 0.988, indicating a good statistical fit (Figure 5B).

Figure 5 ROC curve and calibration plot of the nomogram in the validation cohort. (A) ROC curve of the nomogram in the validation cohort; (B) calibration plot of the nomogram in the validation cohort. AUC, area under the receiver operating characteristic curve; CI, confidence Interval; ROC, receiver operating characteristic.

Discussion

In this study, we established a clinical-bimodal nomogram to predict the risk of malignancy in patients with PND. As the radiologic workup remains controversial (15), we evaluated the additional significance of MG-US model compared with MG model, aiming to ascertain the supplementary value of ultrasound for breast cancer detection in patients with PND. The AUC of the MG-US model was significantly higher than that of the MG model (0.851 vs. 0.789, P=0.014). Obviously, the inclusion of ultrasound predictors in the MG-US model significantly enhanced its discriminating power. The calibration plots presented excellent agreement and the Hosmer-Leme test showed a good statistical fit for MG-US nomogram (P=0.889). In addition, DCA revealed that this model exhibited favorable clinical utility.

Previous studies have explored various strategies to stratify malignancy risk in patients with PND (6,11,13,14). Lian et al. (13) developed a nomogram based on mammary ductoscopic findings, although its predictive accuracy (AUC =0.812) was lower than that of our model. Another study involving 311 women with PND suggested that CNB was useful in identifying malignancies (14). However, both CNB and ductoscopy are invasive procedures. Given that most patients with PND ultimately diagnosed with benign conditions, less invasive diagnostic approaches are preferable for both patients and clinicians.

In this study, six variables were found to be independent predictors for breast cancer, including age, color of discharge, presentation with a palpable mass, suspicious calcifications on MG, mass and intraductal mass/debris on ultrasound. Consistent with previous studies (17), patients with older age were more likely to have malignancy. Additionally, we have noticed a correlation between the presence of bloody nipple discharge and a heightened susceptibility to breast cancer (OR 3.0, 95% CI: 1.4–6.4). Chen et al. performed a meta-analysis which demonstrated a markedly higher breast cancer risk (OR 2.27, 95% CI: 1.32–3.89) in patients with bloody nipple discharge compared with non-bloody nipple discharge (404/1,632, 24.7%; 179/1,478, 12,1%, respectively) (18). The presence of suspicious lumps also showed distinct predictive value in this study (OR 2.8, 95%CI: 1.3–6.4), as was found in previous research (17,19).

Recently, Chung et al. (20) reported that the combination of MG and US is highly effective in detecting breast malignancy in the setting of PND. However, no statistical analysis focused on the relationships between the objective image findings and the likelihood of breast malignancy. Previous research (19) showed that patients diagnosed with malignancy were more likely to have calcifications (P<0.001) and mass/asymmetry (P=0.020) on mammogram, which was similar to our study (suspicious calcifications on mammogram, P<0.001; mass on mammogram, P=0.004). However, only suspicious calcification showed predictive value (OR 3.4, 95% CI: 1.3–10.5). The most influential factor in determining the success or failure of a MG in detecting a mass is the density of the breast tissue. The tumor mass in the thick breast may be concealed by nearby benign tissues (21). In the training cohort, 86.4% (242/280) of patients had heterogeneously or extremely dense breasts. US plays an indispensable role in dense breast examinations. Besides, the diagnostic performance of US is comparable to fiberoptic ductoscopy for PND, the diagnostic accuracy rates were reported as 83.87% (208/248) and 85.08% (211/248) respectively (22). Our results showed that intraductal mass/debris (OR 8.6, 95% CI: 3.3–22.7) and mass (OR 6.3, 95% CI: 2.5–15.6) on US were significantly associated with predictive value for malignancy. Consistent with previous research (23,24), microcalcification on US were not observed to be an independent predictor for breast cancer.

Breast magnetic resonance imaging (MRI) is a valuable adjunct for evaluating lesions that are difficult to detect with conventional imaging due to its high sensitivity and excellent soft tissue contrast (25-27). While MG and US are typically the first-line imaging methods, MRI may be preferred in cases with dense breasts or inconclusive conventional imaging but high clinical suspicion of malignancy (28). Given that smaller lesions are more challenging to diagnose (29), we performed a subgroup analysis of lesions smaller than 10 mm or with negative imaging findings in the training cohort. The results showed that our clinical-bimodal nomogram exhibited strong diagnostic performance in this subgroup, with an AUC of 0.858 (95% CI: 0.793–0.923). This demonstrates that, while MRI offers advantages for small lesions, our model provides complementary diagnostic accuracy with excellent discrimination, calibration, and clinical value for patients with PND.

In this study, we developed a noninvasive and practical alternative by incorporating routinely available clinical and imaging parameters into a predictive nomogram. By quantitatively integrating bimodal imaging findings with clinical features, the model overcomes the limitations of simplified imaging classification and provides a more applicable tool for individualized risk stratification. There are still several limitations in this study. First, this is a retrospective study, and the inclusion criteria requiring biopsy or surgery may result in a selection bias. Second, MRI data were not included in the nomogram due to its high cost, limited availability, and the lack of universal recommendation as a first-line imaging modality for evaluating PND. This exclusion may modestly limit the comprehensiveness of the model, and future studies incorporating MRI features are warranted to further improve diagnostic accuracy. Third, a larger sample size, further prospective studies, and machine learning-based radiomics or texture analysis are needed to improve the ability of the nomogram. Finally, this study did not identify novel diagnostic paradigms or imaging biomarkers, which may limit its innovation. However, our primary aim was to develop a practical tool by integrating routinely available and objective parameters, thereby facilitating real-world decision-making in the evaluation of PND.

Conclusions

A clinical-bimodal nomogram incorporating MG-US parameters was developed to predict the malignancy risk in patients with PND. Our nomogram may enable clinicians to identify the patients with a low risk of malignancy and formulate the individualized management strategies. Further prospective studies are required to validate the suitability of this nomogram for clinical applications.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://qims.amegroups.com/article/view/10.21037/qims-2025-701/rc

Data Sharing Statement: Available at https://qims.amegroups.com/article/view/10.21037/qims-2025-701/dss

Funding: This work is supported by the Noncommunicable Chronic Diseases-National Science and Technology Major Project (2023ZD0501100); Natural Science Foundation of China (82371739, 81872139, 82072907, 82304120, U1911204 and 51861125203); Science and Technology Projects in Guangzhou (2025A04J7152); Project of The Beijing Xisike Clinical Oncology Research Foundation (Y-Roche2019/2-0078, Y-pierrefabre202102-0107); the Science and Technology Planning Project of Guangdong Province (2023B1212060013); the Fundamental Research Funds for the Central Universities, Sun Yat-sen University (2022005); High-tech, Major and Characteristic Technology Projects in Guangzhou Area (2023-2025) (2023P-ZD14); National Key R&D Program of China (2021YFC3001000); Guangdong Provincial Clinical Research Center for Breast Diseases (2023B110005); Breast Cancer Clinical Research and Treatment Optimization Public Welfare Project (KYJJ20240509); Sun Yat-sen Pilot Scientific Research Fund (No. SYSQH-II-2024-01).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://qims.amegroups.com/article/view/10.21037/qims-2025-701/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. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Ethics Committee of Sun Yat-sen Memorial Hospital (No. SYSKY-2024-559-01) and the need to obtain written informed consent from all patients was waived owing to the retrospective nature of the 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. Sanford MF, Slanetz PJ, Lewin AA, Baskies AM, Bozzuto L, Branton SA, Hayward JH, Le-Petross HT, Newell MS, Scheel JR, Sharpe RE Jr, Ulaner GA, Weinstein SP, Moy L. ACR Appropriateness Criteria® Evaluation of Nipple Discharge: 2022 Update. J Am Coll Radiol 2022;19:S304-18. [Crossref] [PubMed]
2. Panzironi G, Pediconi F, Sardanelli F. Nipple discharge: The state of the art. BJR Open 2019;1:20180016. [Crossref] [PubMed]
3. Adachi K, Kimura F, Takahashi H, Kaise H, Yamada K, Ueno E, Kawate T, Miyahara K, Ueda A, Sato S, Asaoka M, Okazaki M, Uenaka N, Orimoto K, Wu R, Koyama Y, Ishikawa T. Delayed Diagnosis and Prognostic Impact of Breast Cancer During the COVID-19 Pandemic. Clin Breast Cancer 2023;23:265-71. [Crossref] [PubMed]
4. Gray RJ, Pockaj BA, Karstaedt PJ. Navigating murky waters: a modern treatment algorithm for nipple discharge. Am J Surg 2007;194:850-4; discussion 854-5. [Crossref] [PubMed]
5. Lorenzon M, Zuiani C, Linda A, Londero V, Girometti R, Bazzocchi M. Magnetic resonance imaging in patients with nipple discharge: should we recommend it? Eur Radiol 2011;21:899-907. [Crossref] [PubMed]
6. Ashfaq A, Senior D, Pockaj BA, Wasif N, Pizzitola VJ, Giurescu ME, Gray RJ. Validation study of a modern treatment algorithm for nipple discharge. Am J Surg 2014;208:222-7. [Crossref] [PubMed]
7. Bahl M, Baker JA, Greenup RA, Ghate SV. Diagnostic Value of Ultrasound in Female Patients With Nipple Discharge. AJR Am J Roentgenol 2015;205:203-8. [Crossref] [PubMed]
8. Yoon JH, Yoon H, Kim EK, Moon HJ, Park YV, Kim MJ. Ultrasonographic evaluation of women with pathologic nipple discharge. Ultrasonography 2017;36:310-20. [Crossref] [PubMed]
9. Gupta D, Mendelson EB, Karst I. Nipple Discharge: Current Clinical and Imaging Evaluation. AJR Am J Roentgenol 2021;216:330-9. [Crossref] [PubMed]
10. Choi Y, Kim SM, Jang M, Yun B, Kang E, Kim EK, Park SY, Kim B, Cho N, Moon WK. The Value of Adding Ductography to Ultrasonography for the Evaluation of Pathologic Nipple Discharge in Women with Negative Mammography. Korean J Radiol 2022;23:866-77. [Crossref] [PubMed]
11. Sabel MS, Helvie MA, Breslin T, Curry A, Diehl KM, Cimmino VM, Chang AE, Newman LA. Is duct excision still necessary for all cases of suspicious nipple discharge? Breast J 2012;18:157-62. [Crossref] [PubMed]
12. Lustig DB, Warburton R, Dingee CK, Kuusk U, Pao JS, McKevitt EC. Is microductectomy still necessary to diagnose breast cancer: a 10-year study on the effectiveness of duct excision and galactography. Breast Cancer Res Treat 2019;174:703-9. [Crossref] [PubMed]
13. Lian ZQ, Wang Q, Zhang AQ, Zhang JY, Han XR, Yu HY, Xie SM. A nomogram based on mammary ductoscopic indicators for evaluating the risk of breast cancer in intraductal neoplasms with nipple discharge. Breast Cancer Res Treat 2015;150:373-80. [Crossref] [PubMed]
14. Dupont SC, Boughey JC, Jimenez RE, Hoskin TL, Hieken TJ. Frequency of diagnosis of cancer or high-risk lesion at operation for pathologic nipple discharge. Surgery 2015;158:988-94; discussion 994-5. [Crossref] [PubMed]
15. Patel BK, Ferraro C, Kosiorek HE, Loving VA, D'Orsi C, Newell M, Gray RJ. Nipple Discharge: Imaging Variability Among U.S. Radiologists. AJR Am J Roentgenol 2018;211:920-5. [Crossref] [PubMed]
16. D’Orsi CJ, Sickles EA, Mendelson EB, Morris EA. ACR BI-RADS atlas, breast imaging reporting and data system. Reston, VA: American College of Radiology; 2013.
17. Jin L, Zhu L, Li S, Zeng Y, Haixiong L, Su F, Chen K. Predictors of Malignancy for Female Patients with Suspicious Nipple Discharge: A Retrospective Study. Anticancer Res 2017;37:4655-8. [Crossref] [PubMed]
18. Chen L, Zhou WB, Zhao Y, Liu XA, Ding Q, Zha XM, Wang S. Bloody nipple discharge is a predictor of breast cancer risk: a meta-analysis. Breast Cancer Res Treat 2012;132:9-14. [Crossref] [PubMed]
19. Li GZ, Wong SM, Lester S, Nakhlis F. Evaluating the risk of underlying malignancy in patients with pathologic nipple discharge. Breast J 2018;24:624-7. [Crossref] [PubMed]
20. Chung HL, Bevers TB, Legha RS, Speer ME, Tso HH, Sun J, Leung JWT. Nipple Discharge Imaging Evaluation with Mammography, Ultrasound, Galactography, and MRI. Acad Radiol 2023;30:783-97. [Crossref] [PubMed]
21. Sickles EA. Mammographic features of "early" breast cancer. AJR Am J Roentgenol 1984;143:461-4. [Crossref] [PubMed]
22. Yuan H, Tang X, Mou X, Fan Y, Yan X, Li J, Hou L, Ren M. A comparative analysis of diagnostic values of high-frequency ultrasound and fiberoptic ductoscopy for pathologic nipple discharge. BMC Med Imaging 2022;22:155. [Crossref] [PubMed]
23. Wang H, Lai J, Li J, Gu R, Liu F, Hu Y, Mei J, Jiang X, Shen S, Yu F, Su F. Does establishing a preoperative nomogram including ultrasonographic findings help predict the likelihood of malignancy in patients with microcalcifications? Cancer Imaging 2019;19:46. [Crossref] [PubMed]
24. Park CJ, Kim EK, Moon HJ, Yoon JH, Kim MJ. Reliability of Breast Ultrasound BI-RADS Final Assessment in Mammographically Negative Patients with Nipple Discharge and Radiologic Predictors of Malignancy. J Breast Cancer 2016;19:308-15. [Crossref] [PubMed]
25. Berger N, Luparia A, Di Leo G, Carbonaro LA, Trimboli RM, Ambrogi F, Sardanelli F. Diagnostic Performance of MRI Versus Galactography in Women With Pathologic Nipple Discharge: A Systematic Review and Meta-Analysis. AJR Am J Roentgenol 2017;209:465-71. [Crossref] [PubMed]
26. Lubina N, Schedelbeck U, Roth A, Weng AM, Geissinger E, Hönig A, Hahn D, Bley TA. 3.0 Tesla breast magnetic resonance imaging in patients with nipple discharge when mammography and ultrasound fail. Eur Radiol 2015;25:1285-93. [Crossref] [PubMed]
27. Bahl M, Baker JA, Greenup RA, Ghate SV. Evaluation of Pathologic Nipple Discharge: What is the Added Diagnostic Value of MRI? Ann Surg Oncol 2015;22:S435-41. [Crossref] [PubMed]
28. Breast Cancer Screening and Diagnosis. NCCN Clinical Practice Guidelines in Oncology (NCCN Guidelines®) 2020; Version 1.2020 — September 17, 2020. Available online: https://www.nccn.org
29. Xiao X, Jiang Q, Wu H, Guan X, Qin W, Luo B. Diagnosis of sub-centimetre breast lesions: combining BI-RADS-US with strain elastography and contrast-enhanced ultrasound-a preliminary study in China. Eur Radiol 2017;27:2443-50. [Crossref] [PubMed]