Modeling the risk of axillary lymph node metastasis after neoadjuvant chemotherapy in breast cancer: A retrospective study
- Authors:
- Published online on: July 23, 2025 https://doi.org/10.3892/mco.2025.2884
- Article Number: 89
-
Copyright: © Chen et al. This is an open access article distributed under the terms of Creative Commons Attribution License.
Abstract
Introduction
Breast cancer (BC) remains one of the most common and deadly cancers worldwide, with an incidence of >2 million new cases annually (1). Axillary lymph node (ALN) metastasis is a critical factor in BC prognosis and treatment decisions. Its presence is often correlated with advanced disease stages and poor patient outcomes (2). Neoadjuvant chemotherapy (NAC) plays a pivotal role in reducing tumor size and improving surgical outcomes (3), but predicting ALN metastasis after NAC remains a challenge in clinical practice (4). Previous studies have highlighted the significance of molecular subtypes, tumor size changes and biomarker alterations in predicting ALN status after NAC in patients with BC (5,6). A meta-analysis revealed that molecular subtypes are strongly associated with different responses to NAC, emphasizing their predictive value for axillary outcomes (7). Additionally, changes in tumor size and specific biomarkers after NAC have been identified as critical predictors of ALN status, providing a quantitative approach to assess treatment response (8). Moreover, a nomogram integrating clinical and pathological features has been developed to predict axillary pathological complete response (pCR) in patients with clinically node-positive BC, demonstrating excellent predictive performance and potential clinical utility (9). Kim et al (10) examined the consistency of tumor and lymph node responses in patients with BC after NAC and found that certain pathological markers such as tumor size and histological grade were linked to the likelihood of ALN involvement. Larger tumors and higher histological grades have been identified as significant predictors of ALN metastasis (11). Moreover, molecular markers such as Ki-67 proliferation index, estrogen receptor (ER) status, progesterone receptor (PR) status, and human epidermal growth factor receptor 2 (HER2) expression have been implicated in the prediction of metastasis (12,13).
Despite these findings, accurately predicting ALN metastasis post-NAC remains difficult. Although biomarkers and tumor size changes are critical predictors, the lack of standardized thresholds for these variables makes it difficult to generalize findings. Machine learning (ML) and artificial intelligence (AI) offer promising solutions for improving prediction accuracy. Recent work by Wang (14) and Zheng et al (15) demonstrated that ML models can outperform traditional approaches by integrating complex clinical and pathological data. These studies suggest that combining various biomarkers with advanced imaging modalities could significantly improve model accuracy.
The present study aimed to construct a predictive model using clinical and pathological data to assess the likelihood of ALN metastasis after NAC. The goal is to develop a tool that can guide clinical decision-making, such as identifying patients who may benefit from less aggressive treatment approaches, including avoiding axillary surgery.
Materials and methods
This retrospective analysis included 131 female patients with BC treated at the Affiliated Sanming First Hospital of Fujian Medical University over the last five years (Sanming, China). The patients were randomly assigned to a training cohort (97 patients) and a validation cohort (34 patients). A 3:1 ratio was selected to ensure that the training cohort had sufficient data to develop a stable and reliable model, while retaining an independent validation cohort to assess the model's performance. This 3:1 allocation has been commonly adopted in prediction modeling studies, particularly when sample sizes are moderate, to optimize model stability while preserving external validation capability (16). This allocation strategy is commonly used in predictive modeling studies, especially when sample size is limited. Clinical and pathological data collected included age, menopausal status, tumor size, staging, lymph node status, histological grade, molecular subtypes, ER, PR, HER2 status and Ki-67 expression. Inclusion criteria were as follows: i) Female patients with BC with complete medical records; ii) at least four cycles of NAC, including chemotherapy with or without targeted therapy, followed by ALN dissection (ALND) or sentinel lymph node dissection; iii) a minimum of 10 lymph nodes retrieved during surgery; and iv) pre- and post-NAC imaging (ultrasound, mammography, or MRI) of the breast. Exclusion criteria were as follows: i) Inflammatory or bilateral BC; ii) recurrent BC; and iii) patients receiving only endocrine or targeted therapy prior to surgery.
A total of 156 patients with breast cancer who received NAC were initially reviewed. After applying inclusion and exclusion criteria, 131 eligible cases were included in the study. These were randomly divided into a training cohort (n=97) and a validation cohort (n=34). The flow diagram of the study design, including case selection and analysis process, is shown in Fig. 1.
A total of 156 patients with BC treated with NAC were screened. After excluding 25 patients due to ineligibility or missing data, 131 patients were included in the study. They were divided into a training cohort (n=97) and a validation cohort (n=34).
Immunohistochemical (IHC) staining was performed on formalin-fixed, paraffin-embedded tumor tissue samples to assess ER, PR, HER2 and Ki-67 expression. Patients with missing values for pre-chemotherapy histological grade (n=5) were excluded from grade-specific analysis. ER and PR positivity was defined as ≥1% tumor cell nuclear staining. HER2 status was determined using IHC, with fluorescence in situ hybridization performed for equivocal results (IHC score of 2+).
A predictive model for ALN metastasis after NAC was developed using stepwise multivariate logistic regression analysis. Candidate variables were selected based on both univariate analysis (P<0.05) and clinical relevance. The final model included four independent predictors: Age group, pre-chemotherapy lymph node status, Ki-67 reduction level, and pre-chemotherapy molecular subtype. A nomogram was constructed based on the logistic regression coefficients to provide individualized risk estimates. Each predictor contributes a specific point value according to its weight in the model, and the total score corresponds to a predicted probability of lymph node metastasis.
Statistical methods
The data were divided into a training cohort and a validation cohort. The baseline demographic and pathological characteristics of both cohorts were compared using descriptive statistics. Categorical variables were summarized as frequencies and percentages, while continuous variables were expressed as the mean ± standard deviation (SD) or median (interquartile range), depending on the distribution of the data. For comparisons between cohorts, the chi-square test or Fisher's exact test was used for categorical variables, and independent t-tests or Mann-Whitney U tests were applied for continuous variables.
Univariate analysis was performed to identify significant predictors of ALN metastasis in the training cohort. Variables with P<0.05 were included in the subsequent multivariate logistic regression analysis, which aimed to determine the independent factors associated with metastasis. The regression model was constructed using backward stepwise selection.
A nomogram was constructed based on the multivariate analysis to predict the likelihood of ALN metastasis. The performance of the model was evaluated using the area under the receiver operating characteristic (ROC) curve (AUC). Calibration was assessed using a calibration curve. These steps were performed using R software (version 4.3.1; https://www.r-project.org), with the rms package (version 6.7-0; https://cran.r-project.org/package=rms) for nomogram construction, and the pROC package (version 1.18.4; https://cran.r-project.org/package=pROC) for ROC analysis.
The validation cohort was used to externally validate the predictive model. The AUC of the ROC curve for the validation cohort was calculated to assess model performance. Additionally, confusion matrices for both the experimental and validation cohorts were generated, and diagnostic metrics including sensitivity, specificity, positive predictive value, negative predictive value, and false positive rate were calculated. The validation analysis was conducted in R. All statistical analyses were performed using IBM SPSS Statistics 29 (IBM Corp.) and R 4.3.1. P<0.05 was considered to indicate a statistically significant difference.
Results
Comparison of clinicopathological characteristics between the training cohort and the validation cohort
Based on the inclusion criteria, a total of 131 patients with well-documented clinical data were identified for analysis. Of these, 97 patients were assigned to the training cohort and 34 patients to the validation cohort. As shown in Table I, there were no statistically significant differences between the training and validation cohorts in terms of age, tumor size, menopausal status, clinical TNM staging, histological grade, hormone receptor status, HER2 status, Ki-67 index and molecular subtypes (all P>0.05). This indicates that the two groups were comparable at baseline, allowing for valid model training and validation.
Univariate logistic regression analysis of ALNM in the training cohort
Using data from the experimental cohort (97 cases), univariate analysis was performed on factors such as age, tumor stage before chemotherapy, lymph node stage, histological grade, molecular subtyping, tumor stage, estrogen and PR expression, HER-2 status, Ki67 expression, post-chemotherapy tumor stage, and the proportion of tumor and Ki67 regression (≤30% and >30%) before and after chemotherapy, in relation to post-chemotherapy ALN status (Table II). Age, pre-chemotherapy lymph node stage, histological grade, molecular subtyping, estrogen and PR expression, HER-2 status, post-chemotherapy tumor stage, and the proportion of tumor and Ki67 regression before and after chemotherapy were significantly associated with post-chemotherapy ALN metastasis (P<0.05).
 | Table IIUnivariate analysis of clinical and pathological characteristics in the training cohort and post-chemotherapy axillary lymph node status. |
Multivariate logistic regression analysis of ALNM in the training cohort
The age, pre-chemotherapy lymph node status, histological grade, molecular subtype, post-chemotherapy clinical tumor stage, pre- and post-chemotherapy tumor size and Ki67 regression rate were included in the multivariate analysis (Table III). In the multivariate logistic regression analysis, several variables were independently associated with post-chemotherapy ALN metastasis. Compared with patients with N0 status, those with N1 (OR=49.91; 95% CI: 3.061-813.86; P=0.006), N2 (OR=51.895; 95% CI: 1.959-1374.982; P=0.018) and N3 (OR=60.435; 95% CI: 1.520-2402.544; P=0.029) had significantly higher odds of residual ALN involvement. Regarding molecular subtype, the HER2+/hormone receptor (HR)- group had a significantly reduced risk compared with the HER2-/HR- group (OR=0.05; 95% CI: 0.003-0.828; P=0.037). Post-chemotherapy clinical tumor stage was also a strong predictor: T1 (OR=19.858; 95% CI: 2.464-160.057; P=0.005) and T2 (OR=233.234; 95% CI: 8.020-6782.535; P=0.002) were associated with significantly increased ALN metastasis risk compared with T0. Although Ki67 reduction >30% was not statistically significant (P=0.071), it showed a trend toward reduced ALN metastasis risk (OR=0.063; 95% CI: 0.118-0.685). Other variables, including age, histological grade, and tumor regression level, did not reach statistical significance in the multivariate model (Table III).
| Table IIIMultivariate analysis of clinical pathological features and axillary lymph node metastasis after chemotherapy in the training cohort. |
Nomogram for predicting ALN metastasis
A nomogram was built to visualize the predictive model based on the multivariate logistic regression results. The model integrates four independent predictors: age group, pre-chemotherapy lymph node status, Ki-67 reduction level and molecular subtype. Each variable is assigned a specific point value proportional to its predictive weight. The total score, calculated by summing these values, corresponds to a predicted probability of ALN metastasis after NAC on a scale from 0 to 1 (Fig. 2). This graphical tool allows for individualized risk estimation and supports clinical decision-making, particularly in evaluating whether axillary surgery is necessary.
The nomogram integrates age group, pre-chemotherapy lymph node status (N stage), Ki67 reduction level and molecular subtype. Each predictor contributes a specific score, and the sum of these scores corresponds to a probability of ALN metastasis. The model enables individualized risk assessment and can assist clinicians in selecting appropriate axillary management strategies.
Discrimination and calibration in the training cohort
The model's performance in the training cohort is demonstrated in Fig. 3. The ROC curve (Fig. 3A) shows an area under the curve (AUC) of 0.877 (95% CI: 0.856-0.899), indicating excellent discrimination between patients with and without ALN metastasis. The calibration curve (Fig. 3B) demonstrates favorable agreement between predicted and observed probabilities, indicating that the model is well calibrated and reliable for clinical use.
External validation in the validation cohort
The performance of the predictive model was further evaluated in the validation cohort. As demonstrated in Fig. 4, the ROC curve yielded an AUC of 0.842 (95% CI: 0.818-0.867), demonstrating favorable discriminatory ability in distinguishing patients with and without ALN metastasis after NAC. Although the model's false-negative rate in this cohort was relatively high at 24%, the use of only routinely available clinical and pathological variables makes the model suitable for real-world application, particularly in settings where advanced imaging or molecular testing is limited.
The curve demonstrates an AUC of 0.842 (95% CI: 0.818-0.867), indicating favorable discrimination. The dashed diagonal line represents the performance of a random classifier. The model shows acceptable predictive performance when applied to an independent dataset.
Predictive ability of the model
The predictive ability of the model was calculated for both the training cohort and the validation cohort. Training Cohort: Positive predictive value (PPV), 0.8793; negative predictive value (NPV), 0.8718; true positive rate (TPR), 0.9107; true negative rate (TNR), 0.8293; false positive rate (FPR), 0.1707; and false negative rate (FNR), 0.0893. Validation Cohort: PPV, 0.8125; NPV, 0.778; TPR, 0.7647; TNR, 0.8235; FPR, 0.1765; and FNR, 0.2353 (Table IV).
Model performance illustrated by confusion matrices
The predictive performance of the model is further illustrated in Fig. 5, which displays the confusion matrices for the training and validation cohorts.
In the training cohort, the confusion matrix in Fig. 5 shows that the model successfully identified most patients with and without ALN metastasis. Specifically, the majority of true positive and true negative cases were correctly classified, with a limited number of false positives and false negatives. In the validation cohort, while overall performance remained consistent, there was a modest increase in false negatives, suggesting a slightly reduced sensitivity compared with the training group. The visual layout of the confusion matrices offers an intuitive summary of the model's predictive behavior in both datasets, complementing the numeric metrics in Table IV and reinforcing the model's clinical applicability as well as its limitations in real-world validation.
The confusion matrices showing the classification results of the predictive model in the training (A) and validation (B) cohorts. Each matrix displays the number of true positives, true negatives, false positives and false negatives. The visual representation provides an intuitive assessment of the model's diagnostic performance and highlights differences in predictive accuracy between the two datasets.
Discussion
In the present study, a predictive model for ALN metastasis after NAC was developed in patients with BC, using clinical and pathological variables. The model showed promising results with favorable discriminatory ability (AUC=0.87) in the training cohort and moderate performance (AUC=0.84) in the validation cohort.
To contextualize these findings, our model's performance was compared with several established studies. Wang et al (17) developed a nomogram to predict ALN pCR after NAC in HER2-positive patients with biopsy-confirmed node-positive BC. The model showed favorable performance, with AUCs of 0.719, 0.753 and 0.720 in the training, test, and external validation sets, respectively. However, the present study focused on a specific patient subgroup, which may limit its generalizability. By contrast, our model incorporates a broader range of patients with different molecular subtypes and nodal statuses, enhancing its applicability to the general BC population receiving NAC. Compared with the model developed by Kim et al (18), which incorporated MRI, ultrasound findings, and molecular biomarkers to predict residual ALN metastasis after NAC, our model demonstrated comparable or superior predictive performance. Kim's model achieved AUCs of 0.84 in the test set and 0.78 in the validation set, with a false-negative rate ranging from 5-10% in low-risk patients. In the present study, the nomogram yielded AUCs of 0.877 and 0.842 in the training and validation groups, respectively. Although the false-negative rate in our validation group was higher (24%), our model uses only routinely available clinical and pathological data, making it easier to implement in real-world settings, particularly in resource-limited hospitals. This trade-off between accessibility and accuracy highlights the potential of our model as a practical screening tool, especially when imaging resources are limited or patient selection is needed for further axillary evaluation. Other models have utilized ML and radiomics to improve prediction. For example, Lin et al (6) developed a radiomics-based model with an AUC of 0.833, and Chen et al (19) showed high AUCs (up to 0.899) and reduced the false positive rate from 77.9-32.9%. However, it relies on advanced imaging and AI tools. Our model, relying solely on routinely available clinical and pathological features, offers a more accessible and cost-effective alternative with comparable predictive performance.
This predictive model stands out for its integration of clinical, molecular and dynamic treatment response factors, such as Ki67 reduction, age, pre-chemotherapy lymph node status and detailed molecular subtypings. Unlike traditional models that rely solely on imaging or static biomarkers, this model incorporates chemotherapy-induced changes, providing a more accurate and personalized prediction of ALN status. Its practical nomogram format enables easy clinical application, advancing precision oncology by aligning with the latest trends in BC management. This approach improves upon existing models by offering dynamic, individualized, and treatment-specific predictions.
To support the findings of the present study regarding the significance of pCR and biomarkers such as ER, PR, HER2 and Ki67, relevant literature was incorporated (20,21). Fasching et al (22) demonstrated that Ki67, as well as hormone receptor status, significantly influenced response to NAC and patient prognosis. Similarly, Yoshioka et al (23) showed that higher Ki67 expression levels were associated with increased pCR rates, reinforcing Ki67's utility as a predictive biomarker. Furthermore, a comprehensive meta-analysis by Spring et al (24) highlighted pCR as a valuable surrogate endpoint for long-term survival outcomes across different molecular subtypes. These studies lend further credibility to the biomarkers used in our predictive model.
While the model showed strong performance, it did not fully meet the ideal criteria for clinical utility, particularly in terms of false negative rates. The relatively high false-negative rate (24%) observed in the validation cohort is a critical limitation of the current model. This may be attributed to factors such as tumor heterogeneity, variability in tumor biology across molecular subtypes, and inter-observer differences in pathological assessment. Such tumor heterogeneity reflects the underlying biological complexity of BC, including variability in the tumor microenvironment (TME), molecular signaling pathways and treatment sensitivity. These biological factors can lead to different chemotherapy responses and complicate accurate prediction of residual disease. Therefore, predictive models must account for this complexity to improve generalizability and reliability. These issues can obscure the true extent of residual disease, leading to misclassification. To address this limitation, future models may benefit from integrating radiomic features derived from imaging data and multi-omics biomarkers, including genomic, transcriptomic and proteomic profiles. Such comprehensive data integration could capture tumor complexity more effectively and improve predictive accuracy.
Additionally, the integration of ML techniques could further enhance model performance (25-27). Recent advances in AI have shown promise in improving predictive accuracy in oncology and combining traditional clinical and pathological features with ML algorithms may yield even more precise results. Additionally, the integration of ML techniques and radiomics approaches could further enhance model performance.
Limitations of the study include: First, since the data were derived from a single institution, the sample size is relatively small, which may limit the generalizability of the results. The modest number of cases, especially in the validation cohort, may affect the statistical power and stability of the predictive model. Although the nomogram showed favorable accuracy, its performance needs to be further confirmed in larger, more diverse populations. Future multicenter studies with larger sample sizes are needed to validate the findings and enhance the model's applicability to broader clinical settings. Second, although we attempted to incorporate multiple clinical and pathological factors, some novel molecular biomarkers and imaging techniques that may influence metastasis were not considered. Third, long-term outcome data such as recurrence and overall survival were not available in this retrospective study. Therefore, it was not possible to evaluate the prognostic implications of our model beyond its predictive accuracy for lymph node status. Future prospective studies with extended follow-up are needed to assess the model's ability to predict long-term outcomes and to further validate its clinical utility. Furthermore, our model did not include emerging biomarkers such as TME characteristics or circulating tumor cells, which have shown potential in predicting treatment response and metastasis. The exclusion of these factors was due to limited data availability in our retrospective cohort. Future studies incorporating such biomarkers may improve predictive accuracy and provide deeper insights into tumor biology.
Post-chemotherapy clinical tumor stage was included in the multivariate analysis but was ultimately excluded from the final nomogram model. This decision was based on its strong collinearity with pre-treatment tumor burden and its inconsistent predictive contribution across modeling iterations. To maintain model simplicity and robustness, only the most stable and clinically relevant predictors were retained.
Finally, although the model's false-negative rate remains a limitation, it still holds potential clinical value. Specifically, it may serve as a preliminary screening tool to identify patients at low risk of ALN metastasis after NAC. In carefully selected cases, this could inform decisions to omit ALND and, potentially, even sentinel lymph node biopsy, thereby minimizing surgical morbidity. This highlights the model's role in promoting personalized, less invasive surgical strategies while maintaining oncologic safety.
In conclusion, a predictive model for ALN metastasis after NAC in patients with BC was developed and validated, based on age, pre-chemotherapy lymph node status, Ki67 reduction level and molecular subtyping. This model provides a valuable tool for clinicians in predicting ALN metastasis and making personalized treatment decisions. Although the model demonstrates favorable accuracy, further refinement and validation in larger cohorts are necessary to improve its clinical applicability and reduce false negative rates.
Acknowledgements
Not applicable.
Funding
Funding: The present study was supported by the Natural Science Foundation of Fujian of China (grant no. 2021J011387).
Availability of data and materials
The data generated in the present study may be requested from the corresponding author.
Authors' contributions
WC was responsible for the conception and design of the study, clinical oversight, data collection and critical revision of the manuscript. RH performed the statistical analysis, participated in data interpretation, and drafted the initial manuscript. CC was in charge of pathological data analysis and contributed to result interpretation. MZ contributed to clinical data acquisition and patient coordination. XF participated in clinical data validation and quality control. YW managed patient records and assisted in manuscript revision. All authors read and approved the final version of the manuscript, agree to be accountable for all aspects of the work, and meet the ICMJE authorship criteria. WC and RH confirm the authenticity of all the raw data.
Ethics approval and consent to participate
The present study was approved [approval no. Ming Yi Lun (2021) no. 26] by the Ethics Committee of the Affiliated Sanming First Hospital of Fujian Medical University (Sanming, China). Written informed consent was obtained from all participants prior to their inclusion in the study.
Patient consent for publication
Not applicable.
Competing interests
The authors declare that they have no competing interests.
References
|
Singh D, Vignat J, Lorenzoni V, Eslahi M, Ginsburg O, Lauby-Secretan B, Arbyn M, Basu P, Bray F and Vaccarella S: Global estimates of incidence and mortality of cervical cancer in 2020: A baseline analysis of the WHO global cervical cancer elimination initiative. Lancet Glob Health. 11:e197–e206. 2023.PubMed/NCBI View Article : Google Scholar | |
|
Lyman GH, et al: Axillary node dissection in breast cancer: Review of the literature and clinical guidelines. J Clin Oncol. 23:6313–6320. 2005. | |
|
Devane LA, Baban CK, O'Doherty A, Quinn C, McDermott EW and Prichard RS: The impact of neoadjuvant chemotherapy on margin re-excision in breast-conserving surgery. World J Surg. 44:1547–1551. 2020.PubMed/NCBI View Article : Google Scholar | |
|
Cortazar P, Zhang L, Untch M, Mehta K, Costantino JP, Wolmark N, Bonnefoi H, Cameron D, Gianni L, Valagussa P, et al: Pathological complete response and long-term clinical benefit in breast cancer: The CTNeoBC pooled analysis. Lancet. 384:164–172. 2014.PubMed/NCBI View Article : Google Scholar | |
|
Min SK, Lee SK, Woo J, Jung SM, Ryu JM, Yu J, Lee JE, Kim SW, Chae BJ and Nam SJ: Relation between tumor size and lymph node metastasis according to subtypes of breast cancer. J Breast Cancer. 24:75–84. 2021.PubMed/NCBI View Article : Google Scholar | |
|
Lin G, Chen W, Fan Y, Zhou Y, Li X, Hu X, Cheng X, Chen M, Kong C, Chen M, et al: Machine learning radiomics-based prediction of non-sentinel lymph node metastasis in Chinese breast cancer patients with 1-2 positive sentinel lymph nodes: A multicenter study. Acad Radiol. 31:3081–3095. 2024.PubMed/NCBI View Article : Google Scholar | |
|
Houssami N, Macaskill P, von Minckwitz G, Marinovich ML and Mamounas E: Meta-analysis of the association of breast cancer subtype and pathologic complete response to neoadjuvant chemotherapy. Eur J Cancer. 48:3342–3354. 2012.PubMed/NCBI View Article : Google Scholar | |
|
Yang YF, Liao YY, Li LQ, Xie SR, Xie YF and Peng NF: Changes in ER, PR and HER2 receptors status after neoadjuvant chemotherapy in breast cancer. Pathol Res Pract. 209:797–802. 2013.PubMed/NCBI View Article : Google Scholar | |
|
Vila J, Mittendorf EA, Farante G, Bassett RL, Veronesi P, Galimberti V, Peradze N, Stauder MC, Chavez-MacGregor M, Litton JF, et al: Nomograms for predicting axillary response to neoadjuvant chemotherapy in clinically node-positive patients with breast cancer. Ann Surg Oncol. 23:3501–3509. 2016.PubMed/NCBI View Article : Google Scholar | |
|
Kim TH, Kang DK, Kim JY, Han S and Jung Y: Histologic grade and decrease in tumor dimensions affect axillary lymph node status after neoadjuvant chemotherapy in breast cancer patients. J Breast Cancer. 18:394–399. 2015.PubMed/NCBI View Article : Google Scholar | |
|
Rivadeneira DE, Simmons RM, Christos PJ, Hanna K, Daly JM and Osborne MP: Predictive factors associated with axillary lymph node metastases in T1a and T1b breast carcinomas. J Am Coll Surg. 191:1–6. 2000.PubMed/NCBI View Article : Google Scholar | |
|
Tawfik K, Kimler BF, Davis MK, Fan F and Tawfik O: Ki-67 expression in axillary lymph node metastases in breast cancer is prognostically significant. Hum Pathol. 44:39–46. 2013.PubMed/NCBI View Article : Google Scholar | |
|
Li FY, Wu SG, Zhou J, Sun JY, Lin Q, Lin HX, Guan XX and He ZY: Prognostic value of Ki-67 in breast cancer patients with positive axillary lymph nodes: a retrospective cohort study. PLoS One. 9(e87264)2014.PubMed/NCBI View Article : Google Scholar | |
|
Wang SR, Cao CL, Du TT, Wang JL, Li J, Li WX and Chen M: Machine learning model for predicting axillary lymph node metastasis in clinically node positive breast cancer based on peritumoral ultrasound radiomics and SHAP feature analysis. J Ultrasound Med. 43:1611–1625. 2024.PubMed/NCBI View Article : Google Scholar | |
|
Zheng X, Yao Z, Huang Y, Yu Y, Wang Y, Liu Y, Mao R, Li F, Xiao Y, Wang Y, et al: Deep learning radiomics can predict axillary lymph node status in early-stage breast cancer. Nat Commun. 11(1236)2020.PubMed/NCBI View Article : Google Scholar | |
|
Steyerberg EW: Clinical prediction models: A practical approach to development, validation, and updating, 2nd edition. New York: Springer, 2019. | |
|
Wang W, Wang X, Liu J, Zhu Q, Wang X and Wang P: Nomogram for predicting axillary lymph node pathological response in node-positive breast cancer patients after neoadjuvant chemotherapy. Chin Med J (Engl). 135:333–340. 2021.PubMed/NCBI View Article : Google Scholar | |
|
Kim R, Chang JM, Lee HB, Lee SH, Kim SY, Kim ES, Cho N and Moon WK: Predicting axillary response to neoadjuvant chemotherapy: Breast MRI and US in patients with node-positive breast cancer. Radiology. 293:49–57. 2019.PubMed/NCBI View Article : Google Scholar | |
|
Chen M, Kong C, Lin G, Chen W, Guo X, Chen Y, Cheng X, Chen M, Shi C, Xu M, et al: Development and validation of convolutional neural network-based model to predict the risk of sentinel or non-sentinel lymph node metastasis in patients with breast cancer: A machine learning study. EClinicalMedicine. 63(102176)2023.PubMed/NCBI View Article : Google Scholar | |
|
Fang Y, Zhang Q, Wu Y and Wu J: HER2-positive is an independent indicator for predicting pathological complete response to neoadjuvant therapy and Ki67-changed after neoadjuvant chemotherapy predicts favorable prognosis in Chinese women with locally advanced breast cancer. Medicine (Baltimore). 103(e37170)2024.PubMed/NCBI View Article : Google Scholar | |
|
Rais G, Mokfi R, Boutaggount F, Maskrout M, Bennour S, Senoussi C and Rais F: Assessment of the predictive role of Ki-67 in breast cancer patients' responses to neoadjuvant chemotherapy. Eur J Breast Health. 20:199–206. 2024.PubMed/NCBI View Article : Google Scholar | |
|
Fasching PA, Heusinger K, Haeberle L, Niklos M, Hein A, Bayer CM, Rauh C, Schulz-Wendtland R, Bani MR, Schrauder M, et al: Ki67, chemotherapy response, and prognosis in breast cancer patients receiving neoadjuvant treatment. BMC Cancer. 11(486)2011.PubMed/NCBI View Article : Google Scholar | |
|
Yoshioka T, Hosoda M, Yamamoto M, Taguchi K, Hatanaka KC, Takakuwa E, Hatanaka Y, Matsuno Y and Yamashita H: Prognostic significance of pathologic complete response and Ki67 expression after neoadjuvant chemotherapy in breast cancer. Breast Cancer. 22:185–191. 2015.PubMed/NCBI View Article : Google Scholar | |
|
Spring LM, Fell G, Arfe A, Sharma C, Greenup R, Reynolds KL, Smith BL, Alexander B, Moy B, Isakoff SJ, et al: Pathologic complete response after neoadjuvant chemotherapy and impact on breast cancer recurrence and survival: A comprehensive meta-analysis. Clin Cancer Res. 26:2838–2848. 2020.PubMed/NCBI View Article : Google Scholar | |
|
Wei W, Ma Q, Feng H, Wei T, Jiang F, Fan L, Zhang W, Xu J and Zhang X: Deep learning radiomics for prediction of axillary lymph node metastasis in patients with clinical stage T1-2 breast cancer. Quant Imaging Med Surg. 13:4995–5011. 2023.PubMed/NCBI View Article : Google Scholar | |
|
Shi W, Su Y, Zhang R, Xia W, Lian Z, Mao N, Wang Y, Zhang A, Gao X and Zhang Y: Prediction of axillary lymph node metastasis using a magnetic resonance imaging radiomics model of invasive breast cancer primary tumor. Cancer Imaging. 24(122)2024.PubMed/NCBI View Article : Google Scholar | |
|
Zhang D, Zhou W, Lu WW, Qin XC, Zhang XY, Luo YH, Wu J, Wang JL, Zhao JJ and Zhang CX: Ultrasound-based deep learning radiomics for enhanced axillary lymph node metastasis assessment: a multicenter study. Oncologist. 30(oyaf090)2025.PubMed/NCBI View Article : Google Scholar : https://pubmed.ncbi.nlm.nih.gov/40349137/. |
