Predictive value of multiparametric magnetic resonance imaging combined with pathological biomarkers for axillary lymph node metastasis of breast cancer

Introduction

Accurate detection of axillary lymph node (ALN) metastasis is crucial for determining optimal therapeutic strategies in breast cancer management (1). Currently, sentinel lymph node biopsy (SLNB) and ALN dissection (ALND) are standard approaches for evaluating ALN involvement; however, these invasive procedures carry significant risks and potential complications (2). Hypoxia-inducible factor-1 alpha (HIF-1α), a nuclear transcription factor expressed by tumor cells under hypoxic conditions, plays an essential role in tumor progression, angiogenesis, invasion, and metastatic dissemination (3,4). Programmed death ligand-1 (PD-L1), an immune checkpoint molecule during tumor immune escape, is upregulated in several solid tumors, and serves as a prognostic marker and predictive factor for the efficacy of anti-programmed cell death 1 (PD-1)/PD-L1 therapy (5).

Magnetic resonance imaging (MRI) provides excellent resolution for soft tissue visualization and is a non-invasive technique. However, MRI has limitations in directly visualizing lymph node status, primarily due to the peripheral anatomical location of ALNs relative to the scanning coil, often resulting in inadequate imaging. Furthermore, early metastatic involvement may not present distinct radiological characteristics, leading to false-negative results. To address these limitations, functional MRI techniques, such as intravoxel incoherent motion (IVIM), have been introduced. IVIM differentiates between perfusion-related pseudodiffusion and true molecular diffusion through bi-exponential fitting, providing valuable insights into tumor microenvironmental conditions (6). Additionally, quantitative dynamic contrast-enhanced (DCE)-MRI employs pharmacokinetic models for quantitative evaluation of tumor microvascular perfusion and permeability (7).

Previous studies have employed modalities including mammography, diffusion-weighted imaging (DWI), ultrasound, conventional MRI, and semi-quantitative DCE-MRI parameters to predict ALN metastasis in breast cancer patients (8,9). However, limited research has explored the integrated predictive value of combining quantitative DCE-MRI, IVIM parameters, conventional MRI features, and pathological biomarkers. Addressing this gap, the current study investigates the efficacy of multimodal MRI combined with pathological biomarkers in predicting ALN metastasis. Findings from this study aim to inform clinical decisions, potentially minimizing the need for invasive lymph node biopsy procedures. We present this article in accordance with the TRIPOD reporting checklist (available at https://gs.amegroups.com/article/view/10.21037/gs-2025-215/rc).

Methods

Patients

This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Ethics Committee of the Affiliated Hospital of Jining Medical University (No. 2021C019), and informed consent was taken from all the patients. The study included a cohort of 188 patients diagnosed with breast cancer based on postoperative pathology at the Affiliated Hospital of Jining Medical University from October 2019 to September 2024. All participants completed a 3.0T MRI examination prior to surgery and SLNB or ALND. The screening process and exclusion criteria are illustrated in the flow chart in Figure 1. Ultimately, 149 patients were included in the analysis. Patients were randomly allocated to the training cohort (70%, 42 and 62 presented with ALN and non-ALN metastasis) or validation cohort (30%, 18 and 27 presented with ALN and non-ALN metastasis), respectively (7:3 ratio). Regarding histological type, invasive ductal carcinoma, ductal carcinoma in situ, cribriform carcinoma, invasive lobular carcinoma, medullary carcinoma, and mucinous carcinoma were detected in 138 (92.6%), 4 (2.7%), 2 (1.3%), 1 (0.8%), 2 (1.3%), and 2 (1.3%) cases, respectively.

Figure 1 Flowchart of enrolled patients. MR, magnetic resonance.

MRI techniques

A Discovery MR750w 3.0T magnetic resonance scanner (GE Medical Systems, Chicago, USA) equipped with a specialized 16-channel phased-array coil for breast imaging was utilized for the MRI procedure. The scanning protocol was executed as follows: an initial transverse axial fast spin-echo T1-weighted imaging (FSE-T1WI) sequence [slice thickness/gap =5/1 mm, field of view (FOV) =320 mm × 288 mm, and repetition time (TR)/echo time (TE) =420/10 ms], followed by an axial fat-suppressed fast-recovery T2-weighted imaging (FRFSE-T2WI) sequence with slice thickness/gap =5/1 mm, FOV =320 mm × 288 mm, and TR/TE =6,000/88 ms. Subsequently, IVIM imaging was conducted prior to administration of the contrast agent, employing a spin-echo echo-planar imaging (SE-EPI) sequence. The IVIM parameters were set as TR/TE of 2,500/90 ms and matrix size of 128 mm × 128 mm. A series of 13 b-values in total (0, 20, 30, 50, 70, 100, 150, 200, 500, 700, 1,000, 1,500, 2,000 s/mm²) were utilized across three orthogonal directions. Number of excitations (NEX) was increased from 1 to 6 with escalation of the b-value, with the objective of ensuring the optimal signal-to-noise ratio (SNR) in the acquired images. Finally, three-dimensional volume imaging for breast assessment (3D-VIBRANT) with gradient echo was implemented to conduct DCE-MRI. The contrast material was concurrently administered via a high-pressure syringe with MRI scanning following acquisition of the pre-contrast mask image. After acquisition of one precontrast image, gadodiamide (0.1 mmol/kg BW, Omniscan, GE Healthcare) was injected as a bolus of 3 mL/s, followed by rinsing in isotonic saline (20 mL) at an equal infusion rate. In total, 46 phases were acquired (including one phase for masking and 45 phases for enhancement).

Data analysis

The post-processing workstation was selected for analysis of MRI images. Two breast radiologists with 15 and 7 years of experience conducted a double-blind evaluation. In case of a disagreement, the assessors reached a consistent conclusion through discussion. Each sample of breast cancer was evaluated based on five MRI characteristics, namely, time intensity curve (TIC) classification (I, II, or III), tumor size (longest diameter), tumor margin (circumscribed or not circumscribed), tumor shape (oval/round or irregular), and tumor enhancement (homogeneous/heterogeneous or rim).

The Advantage Workstation (version AW 4.6, GE Medical System) was utilized for IVIM image transfer, whereby the processing of all images was completed through vendor-provided software, resulting in the generation of parametric maps. The biexponential model was applied to measure the IVIM parameters, as outlined below:

Sb/S0=(1−f)exp(−bD)+fexp[−b(D∗+D)][1]

The slow apparent diffusion coefficient (D), perfusion fraction (f), and fast apparent diffusion coefficient (D*) were determined using the biexponential model (10).

DCE-MRI data were analyzed using non-commercial software (Omni Kinetics, GE Healthcare, Beijing, China). An automated population-based selection of the arterial input function was subsequently applied to create perfusion maps. Quantitative parameters were derived from the DCE-MRI model through applying a two-compartment pharmacokinetic Tofts model, as follows:

dCt/dt=KtransCp−KepCt[2]

Ct(t)=Ktrans[Cp(t)⊗e−(Kept)][3]

Kep=Ktrans/Ve[4]

The K^trans parameter measures the transfer of the contrast agent from the plasma into the extravascular extracellular space (EES) by traversing the capillary wall, Kep evaluates the diffusion of the contrast agent returning to the plasma from the EES, and Ve represents the EES volume (11).

Regions of interest (ROIs) were manually outlined on the axial IVIM maps at a b value of 1,000 s/mm² as well as the DCE images in the final phase. The largest cross-sectional areas of lesions were identified across three adjacent slices. The ROIs for the IVIM and DCE images were created on distinct grayscale maps, ensuring that their locations and shapes were as consistent as possible. Subsequently, the ROIs were accurately transferred onto the colour-coded parametric maps derived from the IVIM or DCE images using 3D-synchro-view technology (GE Healthcare, Chicago, USA; Figure 2). Considerable efforts were made to incorporate the solid region of the tumour in the ROIs to the greatest extent, with reference to the final-phase DCE-MRI and T2WI images to prevent the impact of cystic changes, large blood vessels, necrotic components. To minimize measurement errors resulting from ROI selection bias, two radiologists with experience of 15 and 5 years independently performed ROI delineation. All lesions were assessed in triplicate to derive the necessary parameters, followed by calculation of the corresponding average values.

Figure 2 A 30-year-old female patient with ALN metastasis of breast cancer. (A) This mass showed hyper-intensity on DWI (b=1,000). (B-D) Pseudo-colored maps of the D, D*, f values (D =0.56×10⁻³ mm²/s, D* =90.7×10⁻³ mm²/s, f=19.3%). (E) Contrast-enhancement scan showed homogeneous enhancement. (F-H) Pseudo- colored maps of the K^trans, Kep, and Ve values (K^trans =0.788/min, Kep =0.879/min, Ve =0.880). (I) H&E staining of invasive lobular carcinoma (400× magnification). Immunohistochemical staining (400× magnification) for Ki-67 (+, 20–30%) (J), HIF-1α (+++) (K), and PD-L1(+) (L). ALN, axillary lymph node; D, apparent diffusion coefficient; D*, fast apparent diffusion coefficient; DWI, diffusion-weighted imaging; f, perfusion fraction; H&E, hematoxylin-eosin; HIF-1α, hypoxia-inducible factor-1 alpha; PD-L1, programmed death ligand-1.

Pathological examination

Hematoxylin-eosin (H&E) staining was performed to evaluate lymph node metastasis status. The pathological status of ALNs was classified as positive if macrometastases (where the largest metastasis exceeded 2 mm in size) or micrometastases (ranging from 0.2 to 2 mm in size) were detected in one or more ALNs. Hormone receptor status, including estrogen receptor (ER) and progesterone receptor (PR), was determined based on the percentage of positively stained nuclei in at least 10 high-power fields, with ≥10% considered positive. Ki-67 expression was categorized as positive (≥14%) or negative (<14%). Human epidermal growth factor receptor 2 (HER-2) expression intensity was assessed semi-quantitatively, scored from 0 to 3+, with scores of 0 and 1+ considered negative, 3+ positive, and 2+ requiring further confirmation via fluorescence in situ hybridization (FISH). Molecular subtypes were classified as luminal A (ER or PR positive and HER-2 negative), luminal B (either ER or PR positive and HER-2 positive), HER-2 positive (ER and PR negative and HER-2 positive), or triple negative (ER, PR, and HER-2 negative). PD-L1 expression, predominantly localized in the cytoplasm and membrane of activated T-cells, was classified as positive when ≥1% of cells were stained (12). HIF-1α staining was graded into four categories based on the proportion of positively stained cells: negative (−; <10%), weak positive (+; ≥10% to <50%), moderate positive (++; ≥50% to <80%), and strong positive (+++; ≥80%). Grades ‘−’ and ‘+’ indicated low expression, while ‘++’ and ‘+++’ indicated high expression levels (13).

Statistical analysis

Statistical analyses were conducted using SPSS (version 25.0; IBM Corporation, Armonk, NY, USA), R (version 4.0.0; http://www.r-project.org/), and MedCalc (version 19.5.1; Ostende, Belgium). Interclass correlation coefficients (ICCs) were utilized for the evaluation of inter-observer consistency, with ICC values of 0.00–0.20, 0.21–0.40, 0.41–0.60, 0.61–0.80, and 0.81–1.00 corresponding to poor, fair, moderate, good, and excellent agreement, respectively. Clinicopathologic features and conventional MRI characteristics were analyzed using the Mann-Whitney U test, independent samples t-test, and χ² test to assess differences between the ALN metastasis and non-ALN metastasis groups. Additionally, IVIM and DCE-MRI parameters between groups were compared using both the Mann-Whitney U test and independent samples t-test. Independent risk factors for ALN metastasis prediction were identified through univariable and multivariate logistic regression analysis. Receiver operating characteristic (ROC) curve analysis was conducted on every parameter or model for evaluation of its diagnostic performance, with differences in area under the ROC curve (AUC) compared using the Delong test. The best model was visualized by plotting a nomogram of the model. Calibration curves were constructed using bootstrapping, with internal validation via 1,000 resamples. The calibration of the model was additionally evaluated through the Hosmer-Lemeshow goodness-of-fit test. The clinical utility of the nomogram was evaluated via decision curve analysis (DCA). The validation cohort was used to validated its generality. Statistically significant differences were indicated by P values <0.05.

Results

Clinicopathologic biomarkers

In the training and validation groups, patients with ALN metastasis exhibited significantly higher Ki-67 (P=0.01, P=0.03) and HIF-1α (P<0.001, P=0.04) expression levels compared to those without metastasis. Additionally, lymphovascular invasion (LVI) and PD-L1 expression were more frequently observed in the ALN metastasis group. No significant differences were found between groups regarding age, PR, histologic grade, HER-2, molecular subtype, or ER expression (all P>0.05) (Table 1). In the training group, the combined predictive model, incorporating Ki-67, HIF-1α, LVI, and PD-L1, demonstrated significantly higher predictive accuracy (AUC =0.985) compared to individual parameters alone (Ki-67: Z=3.292, P=0.001; HIF-1α: Z=2.840, P=0.005; PD-L1: Z=2.852, P=0.004; LVI: Z=2.821, P=0.005).

Conventional MRI features

In the training and validation groups, the lesion margins in patients with ALN metastasis were less distinct compared to those without ALN metastasis (P<0.001, P=0.03). Additionally, a higher proportion of individuals in the ALN metastasis group exhibited heterogeneous enhancement and rim enhancement relative to the non-ALN metastasis group (P=0.02, P=0.04). However, the size, shape, and TIC type presented no significant intergroup variations (P>0.05) (Table 2). In the training group, ROC curve analyses showed that the lesion margin, enhancement, and the conventional MRI model (margin + enhancement) yielded AUCs of 0.603, 0.622, and 0.673, respectively, which were not significantly different (P>0.05).

DCE-MRI and IVIM

The ICCs for D, D*, f, K^trans, Kep, and Ve parameters, assessed independently by two radiologists, ranged from good to excellent (0.79–0.93). In the training and test cohorts, compared to the non-metastatic group, patients with ALN metastasis exhibited significantly lower D values (all P<0.001) and significantly higher values of D* (P=0.02, P=0.04), K^trans (all P<0.001), and Kep (all P<0.001). There were no statistically significant differences between the two groups for f (P=0.41, P=0.19) and Ve (P=0.16, P=0.35) (Table 3).

Univariable and multivariable logistic regression analysis of risk factors in the training validation

Based on univariable and multivariate logistic regression analysis, PD-L1 expression [odds ratio (OR) =82.55, P=0.045], lesion margin (OR =21.08, P=0.048), D (OR <0.001, P=0.01) and K^trans (OR >1,000, P=0.01) were identified as independent predictors for ALN metastasis. When PD-L1 was positive, lesion margin was circumscribed, D≤0.86×10⁻³ mm²/s, and K^trans ≥0.49 (min⁻¹), the ALN metastasis tended to be high risk. Subsequently, a combined model was developed that incorporated PD-L1, margin, D, and K^trans. A nomogram model was further created using these four independent predictors. The calibration curve indicated strong agreement of the nomogram-derived predicted values with the actual observed outcomes with no significant differences, as determined from the Hosmer-Lemeshow test (P=0.99). DCA further showed that the nomogram provided higher net clinical benefits across a wide threshold probability range (0.1–1.0), compared to the strategies of treating all or treating none (Table 4, Figure 3).

Figure 3 Nomogram model establishment and predictive performance evaluation. Nomogram for predicting ALN metastasis of breast cancer (A). Calibration plot for internal validation of ALN metastasis (B). The decision curves analysis of the nomogram (C). ALN, axillary lymph node; D, apparent diffusion coefficient; PD-L1, programmed death ligand-1.

ROC curve analysis

In the training cohort, the AUC value of the DCE-MRI model in predicting ALN metastasis was significantly higher than those obtained with the pathologic, conventional MRI, and IVIM models (Z=3.027, P=0.003; Z=4.400, P<0.001; Z=2.680, P=0.007). Furthermore, in both training and test cohorts, the combined model achieved significantly enhanced AUCs compared with the pathologic, conventional MRI, IVIM, and DCE-MRI models (Z=2.083–4.402, P<0.05) (Table 5, Figure 4).

Figure 4 The ROC curves and AUC values of the models for predicting ALN metastasis of breast cancer in the training cohort (A) and validation cohort (B). ALN, axillary lymph node; AUC, area under the curve; DCE-MRI, dynamic contrast-enhanced magnetic resonance imaging; IVIM, intravoxel incoherent motion; MRI, magnetic resonance imaging; ROC, receiver operating characteristic.

Discussion

The accurate detection of ALN metastasis in breast cancer is critical for guiding clinical treatment decisions. This study identified PD-L1 expression, lesion margin characteristics, D, and K^trans as independent predictive factors for ALN metastasis. When PD-L1 was positive, lesion margin was circumscribed, D≤0.86×10⁻³ mm²/s, and K^trans ≥0.49 (min⁻¹), the ALN metastasis tended to be high risk. Furthermore, in both training and test cohorts, a combined predictive model integrating these MRI and pathological biomarkers demonstrated robust potential as a clinical tool, exhibiting superior predictive performance compared to models relying on individual MRI or pathological parameters alone (14,15). The proposed nomogram outperformed single parameter models, enhancing clinical applicability by providing more precise preoperative risk stratification and potentially reducing unnecessary invasive lymph node biopsies. Future studies should aim to validate these findings in larger, prospective cohorts to confirm the clinical utility and reliability of the developed predictive model.

The findings from this retrospective analysis indicate that clinicopathological biomarkers in breast cancer, such as Ki-67, HIF-1α, PD-L1, and LVI, could serve as predictors of ALN metastasis. Numerous studies indicate that elevated Ki-67 levels correlate significantly with increased ALN metastasis risk, emphasizing the prognostic value of this biomarker (16,17). LVI is proposed to stimulate proliferation of breast cancer cells, potentially triggering ALN metastasis. This sequence of events is believed to commence with lymphangiogenesis, followed by LVI, ultimately culminating in lymph node metastasis (18). PD-L1 expression mediates immune evasion by binding to the PD-1 receptor on lymphocytes, reducing anti-tumor immune responses and facilitating aggressive tumor behaviors, including invasion and lymph node metastasis (19). Additionally, HIF-1α expression under hypoxic conditions promotes tumor cell survival, angiogenesis, and metastatic capacity through transcriptional activation of downstream targets, contributing significantly to tumor progression (20).

Conventional MRI features were also identified as valuable predictors of ALN metastasis. Particularly, unclear or blurred tumor margins indicated aggressive tumor growth and invasion into surrounding tissues, including lymphatic structures, potentially leading to distant metastasis (21). We revealed that tumors displaying ALN metastasis predominantly presented with heterogeneous or rim enhancement. The hypothesis derived from pathological findings is correlated with core necrosis of the tumor, which is surrounded by significant angiogenesis and proliferation of fibrous tissue (22,23). However, due to the subjectivity of analysis of conventional MRI features, the diagnostic efficacy of these models for ALN metastasis of breast cancer remains low. Our experiments revealed an AUC of only 0.673, 0.740 for the conventional MRI model, which was comparable to the diagnostic efficacy reported by Zhao and co-workers (24).

Quantitative MRI parameters, especially those derived from IVIM, provided additional insight into the morphological and functional tumor characteristics (25). Consistent with prior research by Zhao et al. (26), our results revealed significantly lower D values and higher D* values in breast cancers with ALN metastasis. Rapid proliferation and adherence to adjacent structures characteristic of metastatic tumors are thought to restrict water molecule diffusion significantly. Although Meng et al. (27) reported no association between the IVIM-derived D* parameter and lymph node metastasis, differences in findings might be attributed to methodological variations. Specifically, our study utilized a greater number of low b-values (<200 s/mm²), enhancing the stability and reliability of D* and f parameters representing microperfusion. The ROC analysis further supported the superior diagnostic value of the D parameter compared to D*, potentially reflecting the complexity and variability in microvascular perfusion among breast cancer subtypes.

Quantitative DCE-MRI provides additional parameters reflecting tumor vascularity and permeability (28). Our findings indicated that K^trans and Kep were significantly elevated in patients with ALN metastasis, whereas no significant differences in Ve were observed between groups. This observation aligns with prior findings by Luo et al. (29) highlighting that increased K^trans and Kep are associated with elevated tumor angiogenesis and vascular permeability in metastatic breast cancers. ROC analysis revealed that K^trans exhibited greater diagnostic efficiency than Kep, D, and D*, highlighting its potential as a superior quantitative marker for detecting ALN metastasis. Furthermore, the DCE-MRI model achieved a significantly superior AUC of 0.939 compared to the pathological model (AUC =0.782), conventional MRI model (AUC =0.673), and IVIM model (AUC =0.797) in the training validation.

This research has several limitations that should be taken into consideration. Firstly, the relatively small sample size, together with the restricted range of pathological types (the majority being invasive ductal carcinoma), could lead to selection bias. This study included a small group of patients with noninvasive breast tumors (n=11), all of them had lower rates of axially involvement than invasive ductal carcinoma (30). This study was performed in a retrospective diagnostic population, the ALN metastasis group was not further divided into different pathological subtypes to demonstrate which subtype of breast cancer was prone to ALN metastasis, so future studies may need a broader prospective screening setting. Given that the vast majority (92.6%) of patients in our cohort had invasive ductal carcinoma, the findings and the predictive model are primarily applicable to this common histological subtype. Additionally, information bias could arise from technical limitations, particularly related to inadequate coverage of ALN by MRI coils and selection of ROIs primarily within solid tumor areas rather than whole-lesion evaluation. Future research endeavors should focus on high-quality large-scale investigations employing advanced technologies to validate our results.

Conclusions

In conclusion, specific MRI and pathological features of PD-L1, margin, D, and K^trans may serve as potential risk markers for breast cancer with metastasis to ALN. The combined model demonstrates potential in not only assisting healthcare providers with assessment of ALN metastasis prior to surgical intervention for breast cancer but also developing suitable treatment plans.

Acknowledgments

We thank International Science Editing (http://www.internationalscienceediting.com) for editing this manuscript.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://gs.amegroups.com/article/view/10.21037/gs-2025-215/rc

Data Sharing Statement: Available at https://gs.amegroups.com/article/view/10.21037/gs-2025-215/dss

Peer Review File: Available at https://gs.amegroups.com/article/view/10.21037/gs-2025-215/prf

Funding: This work was supported by Research Fund for Academician Lin He New Medicine in Jining Medical University (No. JYHL2022FMS07), Jining Medical and Health key Research and Development Plan Project (No. 2024YXNS059), Jining Medical College Affiliated Hospital Medical Talent Training Program Project (No. 2022-yxyc-010), and Practical teaching education research project of Jining Medical University (No. JYSJ2022B17).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://gs.amegroups.com/article/view/10.21037/gs-2025-215/coif). X.Y. is currently an employee of Philips Healthcare, Beijing, China. J.Z. is currently an employee of GE Healthcare PDX GMS Medical Affairs Department, Shanghai, China. The other 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 the Affiliated Hospital of Jining Medical University (No. 2021C019), and informed consent was taken from all the patients.

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. Qian F, Shen H, Deng C, et al. Establishment of a logistic regression model nomogram for clinicopathological characteristics and risk factors with axillary lymph node metastasis in T1 locally advanced breast cancer: a retrospective study. Gland Surg 2024;13:871-84. [Crossref] [PubMed]
2. Zhang Q, Lou Y, Liu X, et al. DCE-MRI-based machine learning model for predicting axillary lymph node metastasis in breast cancer. Gland Surg 2025;14:228-37. [Crossref] [PubMed]
3. Tang HW, Feng HL, Wang M, et al. In vivo longitudinal and multimodal imaging of hypoxia-inducible factor 1α and angiogenesis in breast cancer. Chin Med J (Engl) 2020;133:205-11. [Crossref] [PubMed]
4. Li X, Yang L, Wang Q, et al. Soft tissue sarcomas: IVIM and DKI correlate with the expression of HIF-1α on direct comparison of MRI and pathological slices. Eur Radiol 2021;31:4669-79. [Crossref] [PubMed]
5. Vranic S, Cyprian FS, Gatalica Z, et al. PD-L1 status in breast cancer: Current view and perspectives. Semin Cancer Biol 2021;72:146-54. [Crossref] [PubMed]
6. Bi Q, Deng Y, Xu N, et al. Differentiation of early-stage endometrial carcinoma from benign endometrial lesions: a comparative study of six diffusion models. Quant Imaging Med Surg 2025;15:121-34. [Crossref] [PubMed]
7. Jin M, Xiao F, Zhao Q, et al. Predicting sentinel lymph node metastatic burden with intravoxel incoherent motion diffusion-weighted imaging and dynamic contrast-enhanced magnetic resonance imaging in clinical early-stage breast cancer patients. Magn Reson Imaging 2025;121:110397. [Crossref] [PubMed]
8. Zhang D, Shen M, Zhang L, et al. Establishment of an interpretable MRI radiomics-based machine learning model capable of predicting axillary lymph node metastasis in invasive breast cancer. Sci Rep 2025;15:26030. [Crossref] [PubMed]
9. Kong Q, Chen Y, Sui Y, et al. The Role of Multiparametric MRI Radiomics for Preoperative Prediction of Axillary Lymph Node Metastasis in Patients With Invasive Breast Cancer: A Comparative Study. Cancer Innov 2025;4:e70022. [Crossref] [PubMed]
10. Sun Z, Zhou Z, Wang L, et al. IVIM and DCE-MRI in Predicting Phenotypic Subtypes and Nottingham Prognostic Index of Breast Cancer. J Coll Physicians Surg Pak 2024;34:400-6. [Crossref] [PubMed]
11. Wang W, Lv S, Xun J, et al. Comparison of diffusion kurtosis imaging and dynamic contrast enhanced MRI in prediction of prognostic factors and molecular subtypes in patients with breast cancer. Eur J Radiol 2022;154:110392. [Crossref] [PubMed]
12. Wu Z, Lin Q, Wang H, et al. Intratumoral and Peritumoral Radiomics Based on Preoperative MRI for Evaluation of Programmed Cell Death Ligand-1 Expression in Breast Cancer. J Magn Reson Imaging 2024;60:588-99. [Crossref] [PubMed]
13. Li X, Hu Y, Xie Y, et al. Whole-tumor histogram analysis of diffusion-weighted imaging and dynamic contrast-enhanced MRI for soft tissue sarcoma: correlation with HIF-1alpha expression. Eur Radiol 2023;33:3961-73. [Crossref] [PubMed]
14. Xue M, Che S, Tian Y, et al. Nomogram Based on Breast MRI and Clinicopathologic Features for Predicting Axillary Lymph Node Metastasis in Patients with Early-Stage Invasive Breast Cancer: A Retrospective Study. Clin Breast Cancer 2022;22:e428-37. [Crossref] [PubMed]
15. Chen ST, Lai HW, Chang JH, et al. Diagnostic accuracy of pre-operative breast magnetic resonance imaging (MRI) in predicting axillary lymph node metastasis: variations in intrinsic subtypes, and strategy to improve negative predictive value-an analysis of 2473 invasive breast cancer patients. Breast Cancer 2023;30:976-85. [Crossref] [PubMed]
16. Chen W, Wang C, Fu F, et al. A Model to Predict the Risk of Lymph Node Metastasis in Breast Cancer Based on Clinicopathological Characteristics. Cancer Manag Res 2020;12:10439-47. [Crossref] [PubMed]
17. Liu G, Xing Z, Guo C, et al. Identifying clinicopathological risk factors for regional lymph node metastasis in Chinese patients with T1 breast cancer: a population-based study. Front Oncol 2023;13:1217869. [Crossref] [PubMed]
18. Coşkun Bilge A, Yaltırık Bilgin E, Bulut ZM, et al. Preoperative Dynamic Contrast-Enhanced and Diffusion-Weighted Breast Magnetic Resonance Imaging Findings for Prediction of Lymphovascular Invasion of the Lesions in Node-Negative Invasive Breast Cancer. Can Assoc Radiol J 2024;75:386-96. [Crossref] [PubMed]
19. Li M, Li A, Zhou S, et al. Heterogeneity of PD-L1 expression in primary tumors and paired lymph node metastases of triple negative breast cancer. BMC Cancer 2018;18:4. [Crossref] [PubMed]
20. Tian Y, Zhao L, Gui Z, et al. Clinical and pathological features analysis of invasive breast cancer with microcalcification. Cancer Med 2023;12:11351-62. [Crossref] [PubMed]
21. Zhou Z, Chen Y, Zhao F, et al. Predictive value of intravoxel incoherent motion combined with diffusion kurtosis imaging for breast cancer axillary lymph node metastasis: a retrospective study. Acta Radiol 2023;64:951-61. [Crossref] [PubMed]
22. Choi EJ, Youk JH, Choi H, et al. Dynamic contrast-enhanced and diffusion-weighted MRI of invasive breast cancer for the prediction of sentinel lymph node status. J Magn Reson Imaging 2020;51:615-26. [Crossref] [PubMed]
23. Dietzel M, Baltzer PA, Vag T, et al. The necrosis sign in magnetic resonance-mammography: diagnostic accuracy in 1,084 histologically verified breast lesions. Breast J 2010;16:603-8. [Crossref] [PubMed]
24. Zhao M, Wu Q, Guo L, et al. Magnetic resonance imaging features for predicting axillary lymph node metastasis in patients with breast cancer. Eur J Radiol 2020;129:109093. [Crossref] [PubMed]
25. Su Y, Zeng K, Yan Z, et al. Predicting the Ki-67 proliferation index in cervical cancer: a preliminary comparative study of four non-Gaussian diffusion-weighted imaging models combined with histogram analysis. Quant Imaging Med Surg 2024;14:7484-95. [Crossref] [PubMed]
26. Zhao M, Fu K, Zhang L, et al. Intravoxel incoherent motion magnetic resonance imaging for breast cancer: A comparison with benign lesions and evaluation of heterogeneity in different tumor regions with prognostic factors and molecular classification. Oncol Lett 2018;16:5100-12. [Crossref] [PubMed]
27. Meng N, Wang XJ, Sun J, et al. Comparative Study of Amide Proton Transfer-Weighted Imaging and Intravoxel Incoherent Motion Imaging in Breast Cancer Diagnosis and Evaluation. J Magn Reson Imaging 2020;52:1175-86. [Crossref] [PubMed]
28. Chen X, Yang Z, Huang R, et al. Development and validation of a point-based scoring system for predicting axillary lymph node metastasis and disease outcome in breast cancer using clinicopathological and multiparametric MRI features. Cancer Imaging 2023;23:54. [Crossref] [PubMed]
29. Luo Q, Yang L, Zhou X. The value of multimodal magnetic resonance imaging in breast cancer and its correlation with pathological features and prognosis. Eur Rev Med Pharmacol Sci 2023;27:8397-403. [Crossref] [PubMed]
30. Yang X, Lu Z, Tan X, et al. Evaluating the added value of synthetic magnetic resonance imaging in predicting sentinel lymph node status in breast cancer. Quant Imaging Med Surg 2024;14:3789-802. [Crossref] [PubMed]