According to the 2020 global cancer statistics, breast cancer has become the most commonly diagnosed malignancy worldwide. In China, both the incidence and mortality of breast cancer are increasing rapidly, with a growth rate approximately twice the global average, ranking among the highest globally (1, 2). Breast cancer cells exhibit a strong propensity for hematogenous and lymphatic dissemination, with axillary lymph node metastasis (ALNM) being the earliest and most common route of spread. ALNM status is closely associated with clinical staging, treatment strategies, and patient prognosis (3).Currently, ALNM is primarily assessed through sentinel lymph node biopsy (SLNB) or axillary lymph node dissection (ALND). While both approaches provide histopathological confirmation, they are invasive and carry a risk of complications such as lymphedema, nerve injury, and sensory deficits, which significantly affect patients’ quality of life (4, 5). Moreover, the false-negative rate of SLNB is highly dependent on the operator’s level of experience (6). These limitations underscore the urgent need for a noninvasive, accurate preoperative method to predict ALNM, facilitating individualized and precision-based surgical management of breast cancer.

Current imaging modalities for assessing ALNM include ultrasound (US), magnetic resonance imaging (MR), mammography (MG), computed tomography (CT), and positron emission tomography-computed tomography (PET-CT). PET-CT is limited in clinical use due to its high radiation exposure and cost. MG is highly sensitive to calcifications but lacks sufficient spatial resolution to accurately evaluate axillary lymph node morphology and vascularity, resulting in low sensitivity. US enables real-time assessment of lymph node size, cortical thickness, vascularity, and elasticity, but metastatic nodes often lack specific features in the early stage, increasing the risk of underdiagnosis. MR offers excellent soft-tissue contrast and provides a detailed view of the tumor and surrounding structures; however, its diagnostic utility is constrained by coil size, motion artifacts, and overlap in imaging features between benign and malignant nodes. Thus, reliance solely on imaging features for ALNM detection remains insufficient.

ALNM in breast cancer is a complex, multistep process influenced by tumor growth and cellular proliferation. The morphology of the primary tumor correlates with its biological characteristics and undergoes changes during lymph node metastasis (7, 8). Previous studies have demonstrated a significant association between the imaging features of primary breast tumors and ALNM (9, 10). Furthermore, immunohistochemical (IHC) markers such as estrogen receptor (ER), human epidermal growth factor receptor 2 (HER-2), and progesterone receptor (PR) contribute to pathological changes and are linked to ALNM occurrence (11–13). However, most prior research has been limited to clinical-pathological data or single imaging modalities, with few studies integrating multi-modality imaging of both primary tumors and axillary lymph nodes to predict ALNM.

This study integrated US and MR imaging features of primary breast tumors and axillary lymph nodes with IHC markers to identify independent risk factors for ALNM using univariate and multivariate analyses. A nomogram prediction model was developed and validated to estimate the probability of ALNM. This intuitive and practical tool enables preoperative risk assessment of ALNM, supporting clinical decision-making and individualized treatment planning.

Breast cancer has surpassed lung cancer as the most prevalent malignancy worldwide, with both incidence and mortality rates in China continuing to rise (19). Despite advances in early detection, distant metastasis and tumor recurrence remain the primary causes of death. ALNM is the most common route of dissemination and plays a crucial role in staging and prognostication (20, 21).

This study explored the predictive value of multimodal imaging features combined with IHC markers for ALNM in breast cancer and constructed a nomogram model to assist clinical decision-making. While imaging techniques such as ultrasound and MRI, along with IHC markers, play significant roles in the preoperative assessment of ALNM, their accuracy has yet to reach the level of histopathological evaluation. SLNB and ALND are widely regarded as the gold standard for assessing ALNM. SLNB, which involves pathological examination of the sentinel node, is effective in identifying metastatic lymph nodes and offers lower complication risks compared to ALND, such as lymphedema and nerve injury (22, 23). Moreover, evidence suggests that ALND may not provide additional survival benefit for patients with only sentinel node metastases but contributes to overtreatment and unnecessary morbidity (22, 24, 25). Therefore, a reliable, noninvasive method for preoperative ALNM assessment is urgently needed.

Tumor size showed a significant association with ALNM in univariate analysis but was not an independent predictor in multivariate regression, possibly due to inter-operator measurement variability.

Notably, microcalcification emerged as an independent predictor (OR = 10.431, P < 0.001) and was included in the final model. Microcalcifications are considered a surrogate for necrosis and aggressive tumor behavior. SWE also showed promise, as tumors in the ALNM group exhibited significantly lower SWEmin values—likely reflecting altered stromal architecture or desmoplastic response—while other SWE parameters (SWEmean, SWEmax), though previously suggested, were not predictive in our cohort, possibly due to equipment differences or sample size limitations (26).

In addition, US-detected suspicious lymph nodes with cortical thickening, round shape, and loss of hilar architecture were significantly associated with ALNM, confirming the diagnostic value of nodal morphology on US (27).

Although tumor morphology, size, margin, and TIC types were statistically significant in univariate analysis, none proved to be independent predictors in multivariate models. This may be attributed to selection bias or limited sample size. Importantly, ADC values were inversely associated with ALNM, with higher ADC values serving as protective factors (OR = 0.93). The best cutoff identified was 0.955 × 10⁻³ mm²/s. These findings are consistent with reports linking lower ADC values to high tumor cellularity and restricted diffusion (28, 29).

MRI-detected suspicious lymph nodes, characterized by enlarged size, cortical thickening, irregular borders, or absence of fatty hilum, also independently predicted ALNM, reaffirming their clinical utility in preoperative evaluation (27).

Among molecular biomarkers, HER-2 overexpression was identified as an independent risk factor for ALNM (OR = 2.916, P < 0.05), consistent with its role in promoting cell proliferation and invasion (11). Although Ki-67 and P63 showed significant differences in univariate analysis, neither entered the final model. The inconsistency in Ki-67 may stem from heterogeneous cutoff definitions. P63’s dual role in different subtypes and the existence of multiple isoforms might explain its variable association with metastasis (30).

A nomogram incorporating six independent variables (SWEmin, microcalcification, suspicious lymph nodes on US and MR, ADC value, and HER-2 status) was developed. The model demonstrated excellent discriminatory ability with AUCs of 0.854 and 0.826 in the training and validation cohorts, respectively. Calibration curves showed strong agreement between predicted and actual probabilities, and decision curve analysis indicated substantial clinical utility. Five-fold cross-validation further confirmed the model’s robustness.

In this study, we developed a non-invasive predictive model for preoperative ALNM by integrating multimodal imaging features with IHC markers. This approach, combining imaging and molecular biomarkers, offers a more comprehensive and individualized method for assessing ALNM risk, which has not been widely explored in previous literature.

Several studies have explored the role of imaging techniques alone in predicting ALNM. Zhao et al. demonstrated the utility of MRI in identifying ALNM, showing that MRI features such as tumor size, shape, and margin can effectively predict axillary lymph node involvement (31). However, their model focused on MRI alone, without integrating IHC markers like HER-2 or Ki-67, which could improve predictive accuracy. Similarly, Yilmaz et al. found that ultrasound elastography could serve as a valuable tool for ALNM detection (32). While ultrasound elastography is a useful technique for assessing tissue stiffness and detecting abnormalities, it also lacks the ability to provide detailed molecular information that might refine risk stratification.

Our study builds upon these findings by integrating both ultrasound and MRI features with IHC markers, such as HER-2 expression. The inclusion of molecular biomarkers allows for a more nuanced prediction of ALNM risk, adding a layer of biological insight to imaging-based assessments. This combination of multimodal imaging and molecular markers is an important innovation in the current literature, as it offers a more comprehensive, non-invasive alternative to the traditional SLNB and ALND, which are invasive and carry potential risks such as lymphedema and nerve injury.

Furthermore, the incorporation of HER-2 expression in our model is particularly significant, as it has been shown to correlate with aggressive tumor behavior and higher rates of metastasis. Our model offers a more precise and individualized approach to preoperative ALNM prediction, potentially reducing unnecessary invasive procedures and improving clinical decision-making. While previous studies have explored the individual use of ultrasound or MRI in ALNM prediction, few have combined these imaging modalities with molecular biomarkers such as HER-2 expression in a single predictive framework (33). Compared to the HRCT-based approach, our model combining ultrasound, MRI, and IHC markers demonstrated higher predictive accuracy and performed more precisely in preoperative axillary lymph node metastasis prediction, particularly in early screening and surgical decision-making for low-risk patients (34).

This was a retrospective single-center study with a relatively small sample size and no external validation. Variability in image acquisition and interpretation across operators and equipment may have influenced the results. Inter-observer agreement was not evaluated. Future multicenter, prospective studies with standardized imaging protocols and quality control measures are warranted to enhance the generalizability of the model.

Our study confirms the feasibility and predictive value of a nomogram integrating US, MR, and IHC features for noninvasive preoperative assessment of ALNM in breast cancer. The model may help reduce unnecessary axillary interventions and optimize individualized treatment strategies. Further validation in larger, multicenter cohorts is needed to support its clinical application.