Nomogram models for prediction of axillary lymph node metastasis in breast cancer patients Authors: Zixi Deng, Yuechong Li, Yongchao Luo, Xi Cao, Songjie Shen Published: 2025 ABSTRACT AbstractBackgroundThe purpose of this study was to establish two preoperative nomogram models to evaluate the status of axillary lymph nodes (ALNs) in early breast cancer patients based on ultrasound data, PET/CT data, and clinicopathological features. Accordingly, more appropriate treatment strategies for breast cancer patients could be selected.MethodsFrom April 2017 to April 2023, a total of 254 patients from the Clinical Data Research (CDR) database of Peking Union Medical College Hospital were included in this study and randomly assigned to the training and internal testing groups at a ratio of 8:2, with outcome variables being axillary lymph nodes metastasis (ALNM) and high nodal tumour burden (HNTB). The predictive factors determined by Least Absolute Shrinkage and Selection Operator (LASSO) regression and multivariate logistic regression analyses were used to construct the nomograms. Receiver operating characteristic (ROC) curves and calibration plots were used to assess the prediction models’ discrimination and calibration.ResultsLASSO regression and multivariate logistic regression analyses showed that Plnsuv (PET/CT lymph node SUVmax), unclear cortico-medullary delineation and Progesterone Receptor (PR) positive were independent predictors of ALNM. Moreover, Plnsuv, tumour size and Her-2 positive were independent predictors of HNTB, respectively. Integrating these independent predictors, two nomograms were successfully developed to accurately predict the status of ALN. For nomogram 1 (prediction of ALNM), the area under the ROC curve in the testing group was 0.889. For nomogram 2 (prediction of HNTB), the area under the ROC curve in the testing group was 0.784. The above results showed a satisfactory performance.ConclusionsWe developed two preoperative nomograms that can be used to predict ALN metastasis (LN − vs. LN+) and the number of metastatic ALNs (≤ 2 vs. >2) in early breast cancer patients. They were well verified in internal testing groups. The nomograms can help clinicians predict the status of ALNs and guide discussions regarding axillary management.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12885-025-15260-z.Keywords: Breast cancer, Axillary lymph node metastasis, PET/CT, Nomogram model, Ultrasound Introduction Axillary lymph node metastasis (ALNM) is the most significant predictor of overall recurrence and survival in breast cancer, with lymph nodes status determining the choice of surgical procedure and systemic treatment strategies [1]. Therefore, accurately assessing the involvement of axillary lymph nodes is a crucial part of breast cancer staging and determining the appropriate treatment. For a long time, Axillary Lymph Node Dissection (ALND) has been the standard surgical approach for managing axillary lymph nodes in breast cancer. ALND serves both diagnostic purposes, to determine the extent of lymph node metastasis, and therapeutic purposes, to remove cancer cells [2]. It wasn’t until the National Surgical Adjuvant Breast and Bowel Project (NSABP) trial B-32 demonstrated that in clinically node-negative (cN0) breast cancer, there was no significant difference in regional recurrence or survival outcomes between patients who underwent ALND and those who only had Sentinel Lymph Node Biopsy (SLNB), with ALND being reserved for those with positive SLNs [3]. The American College of Surgeons Oncology Group (ACOSOG) Z0011 trial further established that in women with T1 or T2 invasive breast cancer, cN0, and one or two metastatic sentinel lymph nodes, SLNB could replace ALND as the method for axillary lymph nodes evaluation [4]. SLNB is a less invasive method but still requires intraoperative pathological assessment and carries risks of complications such as wound infection, seroma, lymphedema, and sensory abnormalities [5]. Preoperative axillary ultrasound is a routine diagnostic tool in evaluating breast cancer patients. Data from a previous study confirmed that 43.2% of patients with positive ultrasound findings had low nodal tumour burden (LNTB) (≤ 2 positive lymph nodes), indicating that nearly half of the women meet the criteria for LNTB and can safely avoid ALND [6]. Therefore, the ability of axillary ultrasound to identify nodal tumour burden is limited, necessitating a non-invasive and more effective method. Fluorodeoxyglucose positron emission tomography/computed tomography (FDG-PET/CT) can detect the higher glycolysis rate of cancer cells compared to normal cells, facilitating the detection of clinical metastases. The criteria for detecting ALNM using PET/CT include uptake higher than surrounding tissues and a maximum standardized uptake value (SUVmax). The reported sensitivity and specificity of PET/CT for detecting metastatic axillary lymph nodes (ALNs) are 50–80% and 90–100%, respectively [7], although some studies have suggested that PET/CT may underestimate the number of metastatic ALNs [8]. Thus, achieving accurate preoperative prediction of ALN involvement and identifying patients with node-negative or low nodal tumour burden, as well as selecting further treatment for patients with high nodal tumour burden (HNTB) (>2 positive lymph nodes), would be beneficial for clinical decision-making. A reliable preoperative method to identify ALN status would be highly advantageous. In recent years, several models have been developed to predict LN metastasis [9–14]. However, these models have not been widely adopted in clinical practice. In this study, we used patients’ data from Department of Breast Surgery of Peking Union Medical College Hospital in China to develop two predictive models based on preoperative ultrasound data, PET/CT data, and clinicopathological features, aiming to assess the risk of ALNM and evaluate the metastatic burden in early-stage breast cancer patients, ultimately assisting clinicians in devising appropriate preoperative treatment plans for breast cancer patients. Methods Patient selection Our data obtained from the Clinical Data Research (CDR) database at Peking Union Medical College Hospital, focusing on the clinicopathological characteristics and preoperative PET/CT and ultrasound features of breast cancer patients. We retrospectively collected and screened patients diagnosed with breast cancer from April 2017 to April 2023. Variables included age, laterality, lesion quadrant, Estrogen Receptor (ER) status, Progesterone Receptor (PR) status, Human Epidermal Growth Factor Receptor 2 (Her-2) status, tumour size, histological type, number of postoperative (ALND) positive lymph nodes. PET/CT Imaging features were tumour size, tumour SUVmax, lymph node (LN) size and LN SUVmax. Ultrasound parameters included LN size, shape, cortico-medullary delineation, cortical thickness, and blood flow signal. Inclusion criteria included female patients with invasive breast cancer, cN0, unilateral and single tumour. Exclusion criteria included clinical positive axillary lymph nodes, distant metastasis, multiple breast cancer lesions, neoadjuvant therapy and incomplete data. This study was approved by the Institutional Review Board and Ethics Committee of Peking Union Medical College Hospital. Axillary ultrasound For each patient who underwent ultrasound (US) examination, two or more US images were captured using ACUSON S2000 (Siemens) and EPIQ7 (Philips) machines with linear probes (3–12 MHz) by an experienced radiologist and reviewed by two experienced radiologists to confirm the index lesion. Each participant was placed in a supine or oblique position on the table with the ipsilateral hand placed behind the head. The following parameters of the ipsilateral ALN were evaluated: number of suspicious lymph nodes, long axis, short axis, short-to-long axis ratio, shape, margin, cortical morphologic features, cortical thickness, medulla thickness, hilum status, cortico-medullary delineation and blood flow. US results were considered suspicious if lymph nodes showed asymmetric focal or diffuse cortical thickening > 3 mm, loss of hilum, and nonhilar blood flow. If more than one lymph nodes were detected, the variables were assessed on the most suspicious metastatic lymph node. PET/CT All patients fasted for over 6 h, and blood glucose levels were monitored lower than 200 mg/dL prior to [18F]FDG injection (5.55 MBq/kg). PET/CT (2 min/bed) were performed on dedicated PET/CT scanner (Biograph64 Truepoint True V, Siemens) from the middle thigh to the tip of the skull, with an uptake time of 40–90 min. Unenhanced low-dose CT scans (120 kV, 30–50 mAs) were performed for attenuation correction and anatomical positioning. The image data were reconstructed using the ordered-subset expectation maximization method (2 iterations, 8 subsets, Gaussian filter, image matrix size 168 × 168). Two experienced nuclear medicine physicians retrospectively reviewed the PET/CT images, and were in consensus for the image interpretation. The maximum standardized uptake value (SUVmax) was calculated by the measured activity, decay-corrected injected dose, and patient body weight. Since there is no reference standard value for SUVmax, we regard the PET/CT scan as positive based on the qualitative visual assessment. For example, PET/CT scan was defined as positive if the FDG activity in the lesion was moderately or markedly increased (moderate or strong uptake) relative to surrounding soft tissues. Conversely, lesions that showed no or weak FDG uptake (less than or equal to the surrounding soft tissue) were considered negative. SUVmax of the lesion with the highest metabolic activity in the patients with positive PET/CT scan was measured for statistical analysis. Pathological assessment Tumor tissues were obtained from paraffin-embedded specimens for hematoxylin and eosin, immunohistochemical, and fluorescence in situ hybridization staining. ER and PR status were considered positive if immunostaining was positive in more than 1% of tumor cells. Her-2 positivity was defined as a score of 3 + on IHC or amplification on FISH. Surgical procedures All participants underwent axillary surgery. Lymphatic mapping of biopsied SLNs was performed using dual tracer, methylene blue, and indocyanine green (ICG). Surface localization was done by ICG, and all bluestained nodes, nodes with linked blue-stained lymphatic vessels, and palpable suspicious lymph nodes were dissected and submitted for frozen sectioning. ICG was used to reconfirm the SLNs. ALND was avoided in most of the patients who met the ACOSOG Z0011 criteria; otherwise, ALND was performed. Statistical analysis Patients were randomly divided into a training group and a testing group in an 8:2 ratio for the development and validation of the nomograms. Statistical analyses and nomogram development were performed using R software (version 4.3.0; https://www.r-project.org). The chi-square test was used for the comparison of categorical variables, and the t-test was used for the comparison of continuous variables. The glmnet package was used for Least Absolute Shrinkage and Selection Operator (LASSO) regression to screen predictive variables, and the lrm function was used for multivariate logistic regression analysis to identify independent predictors. Based on significant independent predictors, the rms package was used to create the nomogram models. We developed Nomogram 1 to identify ALNM (LN- vs. LN+) and Nomogram 2 to distinguish the number of metastatic ALNs (≤ 2 vs. >2). The receiver operating characteristic (ROC) curves of the predictive model were plotted, and the area under the curve (AUC) was calculated to evaluate the discrimination of the predictive model. The Hosmer-Lemeshow test was conducted to assess the calibration of the nomogram. P < 0.05 was considered statistically significant. Results Patients’ characteristics A total of 254 patients from the CDR database of Peking Union Medical College Hospital were included in this study, spanning from April 2017 to April 2023. The data were divided into two groups: the training group (n = 204) and the internal testing group (n = 50), with outcome variables being axillary lymph nodes metastasis (ALNM) and high nodal tumour burden (HNTB). The specific clinicopathological and imaging features of the patients in the training and testing groups are summarized in Table 1. In the training group, ALNM, LNTB and HNTB were 62.3%, 57.8%, and 42.2%. In the internal testing group, these rates were 62.0%, 54.0%, and 46.0%, respectively. Table 1Comparison of the descriptive characteristics between the testing group and the training groupVariablesWhole population(N = 254)Training Group(N = 204)Testing Group(N = 50)p valueAge (years)50.89 (12.25)50.7 (12.6)51.7 (10.8)0.549Tumour size (mm)28.2 (18.0)27.9 (18.9)29.7 (14.1)0.446Ptsize (cm)2.53 (1.50)2.49 (1.53)2.71 (1.39)0.329Ptsuv8.34 (5.73)7.96 (5.50)9.93 (6.39)0.049Plnsize (cm)0.92 (0.87)0.94 (0.89)0.82 (0.75)0.309Plnsuv3.17 (3.80)3.25 (3.93)2.82 (3.22)0.416Ulnsize (cm)1.54 (1.22)1.53 (1.23)1.54 (1.18)0.959LN shape0.392 Round222 (87.4%)176 (86.3%)46 (92.0%) Irregular32 (12.6%)28 (13.7%)4 (8.00%)LN cortico-medullarydelineation0.548 Clear179 (70.5%)146 (71.6%)33 (66.0%) Unclear75 (29.5%)58 (28.4%)17 (34.0%)LN cortical thickness (cm)0.911 ≤ 0.3111 (43.7%)90 (44.1%)21 (42.0%) > 0.3143 (56.3%)114 (55.9%)29 (58.0%)LN blood flow signal0.743 NO99 (39.0%)78 (38.2%)21 (42.0%) YES155 (61.0%)126 (61.8%)29 (58.0%)Laterality0.969 Left134 (52.8%)107 (52.5%)27 (54.0%) Right120 (47.2%)97 (47.5%)23 (46.0%)Lesion quadrant0.342 UOQ131 (51.6%)104 (51.0%)27 (54.0%) central16 (6.3%)11 (5.39%)5 (10.0%) Other107 (42.1%)89 (43.6%)18 (36.0%)Histological type0.561 IDC241 (94.9%)193 (94.6%)48 (96.0%) ILC6 (2.4%)5 (2.45%)1 (2.00%) Other7 (2.8%)6 (2.94%)1 (2.00%)ER status0.496 Neg69 (27.2%)53 (26.0%)16 (32.0%) Pos185 (72.8%)151 (74.0%)34 (68.0%)PR status0.287 Neg83 (32.7%)63 (30.9%)20 (40.0%) Pos171 (67.3%)141 (69.1%)30 (60.0%)Her-2 status0.362 Neg177 (69.7%)139 (68.1%)38 (76.0%) Pos77 (30.3%)65 (31.9%)12 (24.0%)Number of positive ALNs0.739 ≤ 2145 (57.1%)118 (57.8%)27 (54.0%) > 2109 (42.9%)86 (42.2%)23 (46.0%)ALN status0.772 Neg96 (37.8%)77 (37.7%)19 (38.0%) Pos158 (62.2%)127 (62.3%)31 (62.0%)Ptsize PET/CT tumour size, Ptsuv PET/CT tumour SUVmax, Plnsize PET/CT lymph node size, Plnsuv PET/CT lymph node SUVmax, Ulnsize Ultrasound lymph node size, UOQ Upper outer quadrant, IDC Invasive ductal carcinoma, ILC Invasive lobular carcinoma, Neg Negative, Pos Positive Comparison of the descriptive characteristics between the testing group and the training group LN cortico-medullary delineation Ptsize PET/CT tumour size, Ptsuv PET/CT tumour SUVmax, Plnsize PET/CT lymph node size, Plnsuv PET/CT lymph node SUVmax, Ulnsize Ultrasound lymph node size, UOQ Upper outer quadrant, IDC Invasive ductal carcinoma, ILC Invasive lobular carcinoma, Neg Negative, Pos Positive LASSO regression This study included a total of 17 clinicopathological and imaging characteristics. To achieve multivariate selection and reduce irrelevant and redundant information, LASSO regression was used in the training model, setting the coefficients of non-contributory features to zero by adjusting the λ parameter (Fig. 1). When the dependent variable was ALNM, the significant variables identified were age, Ptsize (PET/CT tumour size), Ptsuv (PET/CT tumour SUVmax), Plnsize (PET/CT lymph node size), Plnsuv (PET/CT lymph node SUVmax), shape, cortico-medullary delineation, cortical thickness, Ulnsize (Ultrasound lymph node size), laterality, lesion quadrant, tumour size, PR status and Her-2 status (Fig. 1 a, b). When the dependent variable was HNTB, the significant variables identified were age, Ptsize, Ptsuv, Plnsuv, shape, cortico-medullary delineation, cortical thickness, blood flow signal, Ulnsize, tumour size, histological type, PR status and Her-2 status (Fig. 1 c, d). a, b LASSO regression in the training model, the dependent variable was ALNM; (c), (d) LASSO regression in the training model, the dependent variable was HNTB Multivariate logistic regression analysis The above statistically significant factors were included in the multivariate logistic regression analysis. The results showed that Plnsuv (OR = 1.12, 95% CI: 1.08–1.17, p < 0.001), unclear cortico-medullary delineation (OR = 25.31, 95% CI: 3.17–202.40, p = 0.002), and PR positive (OR = 4.53, 95% CI: 1.48–13.91, p = 0.008) were independent predictors of ALNM (Table 2). Plnsuv (OR = 1.09, 95% CI: 1.06–1.12, p < 0.001), tumour size (OR = 1.05, 95% CI: 1.01–1.09, p < 0.001), and Her-2 positive (OR = 1.39, 95% CI: 0.62–3.10, p < 0.001) were independent predictors of HNTB (Table 3). Table 2Multivariate logistic regression analysis of ALNM in the training groupVariablesMultivariable logistic regression analysisOR95% CIp valuePlnsuv1.12(1.08, 1.17)< 0.001LN cortico-medullary delineation Unclear25.31(3.17, 202.40)0.002PR status Pos4.53(1.48, 13.91)0.008 Multivariate logistic regression analysis of ALNM in the training group Table 3Multivariate logistic regression analysis of HNTB in the training groupVariablesMultivariable logistic regression analysisOR95% CIp valuePlnsuv1.09(1.06, 1.12)< 0.001tumour size (mm)1.05(1.01, 1.09)< 0.001Her-2 status Pos1.39(0.62, 3.10)< 0.001 Multivariate logistic regression analysis of HNTB in the training group Construction and validation of the prediction nomogram Two nomograms were constructed based on the independent predictors identified by LASSO regression and multivariate logistic regression (Fig. 2). By delineating the scores of corresponding predictors in the nomogram, the probability of a specific patient developing ALNM or HNTB can be predicted. Using the training group model training, ROC curves were applied to the testing group to further evaluate the effectiveness of the nomogram in predicting ALN status (Fig. 3). In the testing group, the AUC for the ALNM prediction model was 0.889 (95% CI: 0.800–0.978.800.978) (Fig. 3 a), the optimal cutoff value based on Youden index was 0.481, with a mean absolute error of 0.024, a mean squared error of 0.00076, and a 0.9 quantile of absolute error of 0.043 (Fig. 3 b). For the HNTB prediction model, the AUC was 0.784 (95% CI: 0.651–0.917) (Fig. 3 c), the optimal cutoff value was 0.267, with a mean absolute error of 0.023, a mean squared error of 0.00073, and a 0.9 quantile of absolute error of 0.042 (Fig. 3 d). The nomograms of prediction model in breast cancer patients: (a) the prediction model 1 of ALNM (LN− vs. LN+); (b) the prediction model 2 of HNTB [number of metastatic ALNs (≤ 2 vs. >2)]. Plnsuv, PET/CT lymph node SUVmax; PR status: progesterone Receptor status; Her-2 status: Human Epidermal Growth Factor Receptor 2 status; tsize, tumour size a ROC curve of predicting ALNM (LN − vs. LN+) in the testing group; (b) The calibration curve for prediction model 1 of ALNM (LN − vs. LN+); (c) ROC curve of predicting HNTB [number of metastatic ALNs (≤ 2 vs. >2)] in the testing group; (d) The calibration curve for prediction model 2 of HNTB [number of metastatic ALNs (≤ 2 vs. >2)] Discussion Currently, breast cancer remains one of the leading diseases threatening women’s health [15]. Accurate lymph node staging is crucial in choosing the appropriate surgical method and subsequent individualized comprehensive treatment. The ACOSOG Z0011 clinical trial indicated that axillary surgical treatment strategies vary based on the number of positive ALNs. Therefore, not only the ALNM but also the number of metastatic lymph nodes is a new focus of axillary status evaluation in current research. For patients with suspected breast cancer, ultrasound is the preferred non-invasive method for preoperative evaluation, predicting malignancy by evaluating lymph node morphological characteristics [16]. Previous studies have reported that morphological features such as a low longitudinal-transverse ratio, central and peripheral blood flow signals, and a cortex thickness greater than 2.5–3 mm are significant independent predictors of malignant lymph nodes [17, 18]. The sensitivity of axillary ultrasound in determining metastasis is 70.6%, but the sensitivity for assessing metastatic burden is 47.8%. This reduction in sensitivity might be due to the exclusion of lymph nodes in levels II and III from routine scans unless there are enlarged level I nodes [12]. PET can identify the typical glucose metabolism increase in malignant tumors. PET combined with CT, i.e. PET/CT scanners, addresses PET’s limitations of low spatial resolution and lack of anatomical detail. PET/CT aids in identifying advanced axillary metastasis and metastatic lymph nodes spread beyond the axilla, particularly in the internal mammary chain. In the study by Fuster D et al. [19], PET/CT identified all extra-axillary lymph nodes, such as infraclavicular, supraclavicular, and contralateral axillary nodes, with high sensitivity (80–94%) and specificity (86–90%). However, its limitation lies in its lower accuracy for small metastases, with sensitivity dropping to 33%, and inflammatory lymph nodes also showing FDG uptake [16, 19]. Thus, PET/CT may overcome the limitations of ultrasound scanning while ultrasound enriches local morphological characteristics. In the study by Y.-S. An et al. [20], the sensitivity and specificity of ultrasound alone in evaluating axillary lymph nodes involvement were 72.3% and 77.3%, respectively. For PET/CT alone, these values were 62.7% and 88.6%. When combined, the sensitivity and specificity improved to 77.1% and 92.4%, with the AUC of the ROC curve reaching 0.848 (P < 0.05). Therefore, the combined diagnostic performance of PET/CT and morphological assessment is superior to either imaging modality alone. In the study by Kang Y.J, a nomogram was developed to predict the number of metastatic ALNs greater than two, incorporating variables such as cortical thickness on ultrasound and lymph nodes SUV uptake on PET/CT [10]. In this study, more objective imaging indicators, such as blood flow signals and cortico-medullary delineation on ultrasound, as well as lymph node size on PET/CT, were included. Based on clinicopathological and imaging features, we developed two predictive models that more accurately assess the axillary lymph nodes involvement and metastatic burden in breast cancer patients, allowing for more precise clinical intervention and surgical decision-making. Our study included 17 relevant variables as potential predictors of ALN status in breast cancer. Following LASSO regression to screen out some variables, multivariate logistic regression analysis revealed that LN SUVmax, tumor size, unclear cortico-medullary delineation, PR positive, and Her-2 positive were independent predictors of ALNM and HNTB. The current studies confirm that tumor size is one of the significant predictors of ALNM. It is widely agreed that the larger the tumor, the higher the likelihood of ALNM, and this factor was included in our model [21–24]. Regarding the included ultrasound features, the findings align with previous studies, indicating that an unclear cortico-medullary structure of axillary lymph nodes is a predictor of lymph node metastasis [9–12, 18]. Normal lymph nodes exhibit clear boundaries, with linear, sheet-like, or clustered hyperechoic medullary-hilum structures connected to one side, known as the lymph node hilum. When infiltrated by tumors, proliferating tumor cells can break through the capsule and compress the medullary region, causing the hyperechoic medullary-hilum to narrow or disappear, leading to an unclear cortico-medullary delineation and irregular cortical thickening, which is also a suspicious sign of lymph node metastasis and was included in our final model. Regarding PET/CT-related variables, it has been confirmed that the LN SUVmax can predict HNTB with high accuracy and specificity [10]. In the study by Heusner TA et al., the combined use of PET/CT for staging breast cancer axillary lymph nodes was superior to morphological imaging methods alone, despite lower sensitivity, possibly because PET/CT is more accurate in detecting a small number of metastatic foci compared to multiple metastatic sites in the axilla [25]. As for molecular subtypes of breast cancer, there is some controversy. In the study by L. Dihge et al., multivariate analysis of breast cancer patients (N0 vs. N+) indicated that ER status and Her-2 status were independent predictors [26]. PR-positive status was ultimately retained in the model, which may be attributed to biases arising from retrospective data collection and internal validation. This finding is inconsistent with conclusions from previous studies. Ugras et al. showed that positive Her-2 status was an independent predictor of HNTB, with the study cohort including cN1 disease patients and a larger proportion of lymph node-positive disease patients [27]. In our study, Her-2 status were independent predictors of higher metastasis risk and burden, consistent with the above findings. However, some studies suggest that hormone receptor-positive and Her-2 negative subtypes posing greater risks for ALNM [28]. Finally, based on the above independent predictors, we used nomograms for model visualization. ROC curve analysis of the validation set showed AUCs of 0.889 and 0.784, respectively, indicating good discrimination in predicting ALN status. The nomogram can provide a basis for appropriate personalized ALN treatment strategies, reducing unnecessary preoperative invasiveness. According to our findings, omission of axillary surgery is a potential option for patients with a low risk of ALNM, while direct ALND could be considered for those with a high risk of HNTB. Despite SLNB being a minimally invasive procedure for axillary evaluation, as complications such as lymphedema and sensory abnormalities can still occur [5]. More clinical trials are investigating the safety of omitting SLNB in breast cancer patients (cN0). The latest conclusions from the SOUND trial suggest that omitting axillary surgery is not inferior to SLNB in T1-2 breast cancer patients with negative axillary ultrasound results, if the lack of pathological information does not affect postoperative treatment plans, patients with these characteristics can safely avoid axillary surgery [29]. The SOUND trial baseline characteristics are largely consistent with those of our cohort. This homogeneity in population demographics lends further credibility to the reliability of our predictive model, particularly in identifying patients with a low risk of ALNM for whom SLNB can be safely omitted. The latest data from INSEMA trial reported a 5-year invasive disease-free survival (IDFS) rate of 91.7% in the SLNB group and 91.9% in the non-SLNB group [30]. This high-quality evidence supports that omitting axillary surgery does not compromise survival outcomes for patients with early-stage, cN0 breast cancer planned for breast-conserving surgery. Therefore, we propose that SLNB omission may be considered for patients assessed as low risk for ALNM. For those at high risk for ALNM and LNTB, SLNB remains necessary for axillary staging. In cases assessed as HNTB, proceeding directly to ALND may be a viable option. However, this study has some limitations: as a retrospective study, selection or information bias is unavoidable. The internal validation cohort was from a single center with a small sample size, and external validation could further support the conclusions. Moreover, ultrasound examination has a certain degree of subjectivity, and the extraction of imaging variables based on textual data rather than image data might impact the model’s precision. Conclusions In summary, we developed and visualized two predictive models based on common clinicopathological and imaging characteristics of breast cancer patients to accurately predict preoperative ALNM and HNTB risks. The quantitative risk assessment and prediction provided by the nomograms can help patients avoid additional invasive treatments and offer personalized axillary management strategies. Supplementary Information Supplementary Material 1 Abbreviations Axillary lymph node Clinical Data Research Axillary lymph nodes metastasis High nodal tumour burden Least Absolute Shrinkage and Selection Operator Receiver operating characteristic PET/CT lymph node SUVmax Progesterone Receptor Human Epidermal Growth Factor Receptor 2 PET/CT tumour SUVmax Axillary Lymph Node Dissection National Surgical Adjuvant Breast and Bowel Project Sentinel Lymph Node Biopsy American College of Surgeons Oncology Group Low nodal tumour burden Fluorodeoxyglucose positron emission tomography/computed tomography Maximum standardized uptake value High nodal tumour burden Clinical Data Research Estrogen Receptor lymph node ultrasound American Society of Clinical Oncology Indocyanine green Receiver operating characteristic Area under the curve PET/CT tumour size PET/CT lymph node size Ultrasound lymph node size Upper outer quadrant Invasive ductal carcinoma Invasive lobular carcinoma Negative Positive Tumour size Author contributions All authors have made substantial contributions to the conception and design of the study, acquisition of data, drafting and revision of the article. All authors have approved the final article. Funding The author(s) declare financial support was received for the research, authorship, and/or publication of this article. This study was supported by National High Level Hospital Clinical Research Funding (2022-PUMCH-B-038). Data availability All data generated or analysed during this study are included in supplementary information files. Declarations Ethics approval and consent to participate This study was approved by the Ethics Committee of Peking Union Medical College Hospital (approval number: K22C2507). Due to the retrospective nature of the study the informed consent is waived by the Ethics Committee of PUMCH. The study was performed in accordance with the national legislation and the institutional requirements. Competing interests The authors declare no competing interests. Footnotes Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Zixi Deng and Yuechong Li contributed equally to this work. References Associated Data Supplementary Materials Supplementary Material 1 Data Availability Statement All data generated or analysed during this study are included in supplementary information files.