Kinetic and radiomic features on DCE-MRI as a predictor for axillary lymph node metastasis burden in T1 and T2 stage breast cancer

Background: Magnetic resonance imaging (MRI) is increasingly used to evaluate axillary lymph node (ALN) status in breast cancer. However, the correlation between MRI features of the primary tumor and the ALN metastasis (ALNM) burden remains poorly understood. This study aimed to develop a non-invasive MRI-based model to preoperatively distinguish between low (≤2 nodes) and high (>2 nodes) ALNM burden in T1 and T2 stage breast cancer.

Methods: This retrospective single-center study included 185 patients, categorized by ALNM burden [≤ 2 nodes (n = 149) or >2 nodes (n = 36)]. The kinetic and radiomic features were extracted from the segmented whole tumor on dynamic contrast-enhanced MRI (DCE-MRI). A forward-stepwise feature selection method was employed based on the ANOVA F-score from the training cohort. Features were added according to F-values and logistical regression model was built iteratively. The final model, trained on the entire training set, was evaluated on the independent test cohort.

Results: The model incorporated five kinetic and three radiomic features, demonstrating moderate predictive performance. The model achieved an area under the receiver operating characteristic curve (AUC) of 0.705 in the test cohort. It showed a sensitivity of 72.7% and a specificity of 77.8%. The negative predictive value (NPV) was 92.1%.

Conclusion: The kinetic and radiomic features from DCE-MRI showed potential for predicting ALNM burden (≤2 or > 2 nodes) in T1 and T2 stage breast cancer. The high NPV particularly supported their utility as a non-invasive tool to identify candidates for less invasive axillary procedures.

1 Introduction

Breast cancer remains one of the most common malignant tumors among women worldwide (1, 2). Axillary lymph node metastasis (ALNM) is a key determinant of overall recurrence and survival in breast cancer patients (3). Since the ACOSOG Z0011 trial (4), there has been a significant change in the management of axillary lymph nodes (ALN). This trial demonstrated that ALN dissection (ALND) could be omitted in T1 or T2 stage breast cancer patients with two or fewer metastatic sentinel lymph nodes (SLNs). A non-invasive method to preoperatively determine ALNM burden (≤2 or > 2 nodes) could facilitate more individualized treatment plans, such as guiding the choice between SLN biopsy (SLNB) alone and direct ALND, or identifying candidates for neoadjuvant chemotherapy.

Magnetic resonance imaging (MRI) is playing an increasingly important role in evaluating both the primary tumor and ALN status in breast cancer (5). Recently, radiomics has gained considerable attention for its ability to extract quantitative imaging features that may reveal invisible disease characteristics and tumor heterogeneity (6). Many previous studies have focused on MRI-based radiomic features derived from the primary breast tumor to predict ALN status. Evidence suggested that radiomic features extracted from MRI could serve as a non-invasive quantitative method to predict ALNM or SLN metastases in breast cancer (7–16). When combined with or without clinicopathological variables, radiomic models reported area under the receiver operating characteristic (ROC) curve (AUC) values varying from 0.60 to 0.91 for predicting ALNM or SLN metastases (7–16). Few studies have focused on MRI radiomics in evaluating ALNM burden. In one study, a radiomic signature achieved an AUC of 0.79 in distinguishing the number of metastatic ALNs (≤2 or > 2 nodes) (7).

The kinetic features of the primary tumor from dynamic contrast-enhanced MRI (DCE-MRI) have also been investigated for their relationship with ALNM. Several studies (17–20) have reported significant correlations of ALNM with kinetic parameters, such as the skewness of K_ep, K^trans value, enhancement maximum slop, time to peak (TTP) and initial peak enhancement (PE) (all p < 0.05). Furthermore, one study (21) found that combined metrics—PE multiplied by tumor volume (PE × volume) and total percent persistent enhancement—could predict the presence of four or more metastatic ALNs, with a combined model achieving an AUC of 0.79. Conversely, Choi EJ et al. (22) concluded that semi-quantitative kinetic parameters (e.g., PE percentage and TTP) were not useful for predicting SLN metastases, as did Rahbar H et al. (23). Our previous study (24) in a relatively small sample found no association between ALNM and quantitative kinetic parameters, including their histogram and texture features. The integration of kinetic parameters with radiomics has been explored in very few studies. To date, the relationship between quantitative kinetic features and the specific number of metastatic ALNs (≤2 or >2 nodes) remains uninvestigated.

The primary aim of this study was to validate the association between quantitative kinetic features and ALNM in a larger sample size. The second was to develop and validate a non-invasive MRI-based prediction model that combined both kinetic and radiomic features from the primary tumor, aiming to preoperatively distinguish between low (≤2 nodes) and high (>2 nodes) ALNM burden.

2 Methods

2.1 Subjects

The study received approval from our hospital’s institutional review board (SCCHEC2015029), and all enrolled patients provided written informed consent. Consecutive patients at Sichuan Cancer Hospital were retrospectively recruited during the period from 01/08/2015 to 31/03/2018. Patients were included if they met the following criteria: 1) pathologically confirmed primary breast cancer; 2) T1 and T2 stage breast cancer, as well as ductal carcinoma in situ (DCIS); 3) DCE-MRI examination before biopsy or surgery; 4) DCE-MRI examination on the same 3.0-T MRI scanner in our department; and 5) pathological diagnosis for ALN by SLNB and/or ALND. Patients were excluded based on the following: 1) no surgical intervention and thus no pathological ALN diagnosis, or receiving neoadjuvant chemotherapy (NAC); 2) advanced breast cancer; 3) receiving chemotherapy or radiotherapy before the MRI examination; 4) interval between MRI scan and surgery exceeding 1 week; and 5) unavailable quantitative parameters due to data processing failures or unsatisfactory imaging quality. The following clinicopathological parameters were collected from medical records: age, menopausal status, tumor size, axillary surgical approach, histopathological type, estrogen receptor (ER), progesterone receptor (PR), human epidermal growth factor receptor 2 (HER2) status, and the Ki-67 proliferation index.

2.2 MRI acquisition

All patients underwent MRI scanning in the prone position using the same 3.0-T MRI device (Siemens Healthcare, Erlangen, Germany) equipped with a dedicated 16-channel breast array coil. The detailed imaging protocol is provided in Supplementary Table S1. The dynamic scanning involved 26 consecutive phases employing the CAIPIRINHA-Dixon-Twist-Vibe sequence.

2.3 Post-processing of MRI images

2.3.1 Whole tumor segmentation

Raw DCE-MRI data were initially processed using dedicated post-processing software, Omni-Kinetics (GE Healthcare, Milwaukee, WI, USA). The reference region model (25) was employed using the pectoralis major muscle as the reference region. Two radiologists (with more than 8 years of experience), blinded to the pathological information of ALN, manually delineated the entire tumor slice by slice on the early post-contrast DCE-MRI images (Figure 1). When a tumor presented with spiculations, they were included in the region of interest (ROI) delineation; however, the peritumoral tissue was excluded. For patients with multiple lesions, the largest tumor was selected for analysis. All ROIs were subsequently reviewed and confirmed by a third radiologist with 15 years of experience. The final ROIs were merged into a single three-dimensional (3D) volume of interest (VOI). As noted in our previous study (11), to ensure the reproducibility of the extracted kinetic and radiomic features, images from 20 randomly selected patients were used for intra- and inter-observer agreement analyses. One radiologist performed the VOI segmentation twice at 1-month intervals to assess intra-observer consistency, while a second radiologist independently segmented the same set to evaluate inter-observer agreement. The reliability was calculated using the intraclass correlation coefficient (ICC). Features with ICC greater than 0.75 were retained for further analysis.

Figure 1

Figure 1. A 51-year-old woman diagnosed with invasive carcinoma of no special type (NST) with negative axillary lymph nodes. (A) Tumor region of interest (ROI) outlined on the early-phase DCE-MRI images. (B) K_ep color-coded map. (C) K^trans color-coded map. DCE-MRI, dynamic contrast-enhanced magnetic resonance imaging.

2.3.2 Kinetic feature extraction

The software automatically generated a comprehensive set of quantitative and semi-quantitative kinetic parameters from the VOI, as illustrated in Figure 1. These parameters included the following: 1) K^trans (min⁻¹), the volume-transfer constant, indicative of vascular permeability and perfusion; 2) K_ep (min⁻¹), the washout-rate constant, reflecting contrast agent reflux back to the vessels; 3) volume fraction of plasma (Vp); 4) voxel value; 5) TTP; 6) blood flow (BF); 7) blood volume (BV); 8) maximum concentration (MAX Conc); 9) MAX Slope; 10) AUC; and 11) mean transit time (MTT). For each of these parameters, the software also provided a range of statistical metrics, including the maximum, minimum, median, mean, Std, and area, as well as the 10%, 25%, 50%, 75%, and 90% values.

2.3.3 Radiomic feature extraction

The radiomic features of the tumors were automatically extracted from the early post-contrast DCE-MRI images using the defined VOI. The software generated a total of 77 features, categorized as follows: morphology metrics (n = 9), first-order statistical features (n = 29), gray-level co-occurrence matrix (GLCM) features (n = 13), Haralick features (n = 10), and run-length matrix (RLM) features (n = 16).

2.3.4 Feature selection and model building

The dataset comprised 185 cases, and we used a random split method to divide the cohort into training (70%) and test (30%) sets, ensuring that the test cohort was completely independent from the training process. The training set consisted of 129 cases (25, >2 metastatic nodes; 104, ≤2 metastatic nodes), while the independent test set included 56 cases (11, >2 metastatic nodes; 45, ≤2 metastatic nodes). Prior to model development, the feature matrix underwent normalization. Specifically, we scaled each feature vector by its L2 norm, effectively transforming it into a unit vector. To address the high dimensionality of the feature space, we performed a feature reduction step. We computed the cosine similarity for all feature pairs; if the cosine value between any two features exceeded 0.86, we removed one of them. This procedure effectively reduced the feature space dimension and ensured greater independence among the remaining features. We used a forward-stepwise feature selection method to select the set of features. We implemented analysis of variance (ANOVA) on the training cohort, and we estimated the F-value for each feature. Then, we added each feature according to the F-value and built a logistic regression model iteratively. For each set of features, we used 10-fold cross-validation to evaluate the performance and to determine the selected features. Then, we built the model based on all training cases and evaluated the model on the independent test cohort. To prevent overfitting and ensure a rigorous evaluation of the model’s generalizability, we strictly confined all feature selection and model-building steps to the training cohort. We then evaluated the final model on the completely independent test cohort, the results of which were reported herein. Figure 2 shows the study workflow.

Figure 2

Figure 2. The study workflow. MRI, magnetic resonance imaging; ALNM, axillary lymph node metastasis; VOI, volume of interest; DCIS, ductal carcinoma in situ; DCE-MRI, dynamic contrast-enhanced MRI; ALN, axillary lymph node; SLNB, sentinel lymph node biopsy; ALND, axillary lymph node dissection.

2.4 Statistical analysis

Model performance was assessed using ROC curve analysis, with the AUC calculated for quantification evaluation. Additional metrics—including sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV)—were computed at the optimal cutoff threshold determined by maximizing Youden’s index. The estimation was also boosted 1,000 times, and a paired t-test was applied to give the 95% confidence interval. The entire analytical workflow was executed in Python 3.7.6, utilizing the FeAture Explorer Pro (FAE, v0.6.4, https://github.com/salan668/FAE).

Continuous variables with normal and non-normal distributions were compared using Student’s t-test and the Mann–Whitney U test and were presented as mean ± standard deviation and median [interquartile range (IQR)], respectively. Categorical variables were compared using the chi-square test or Fisher’s exact test to assess for differences in clinical and pathological factors between the ALNM groups (>2 vs. ≤2 nodes). Statistical analyses were performed using the R software (v4.2.3, http://www.r-project.org). A two-tailed p value <0.05 was considered statistically significant.

3 Results

There were 149 patients with ALN metastases of less than or equal to two nodes and 36 patients with ALN metastases of more than two nodes. The clinicopathological characteristics of patients are listed in Table 1. The clinical and pathological characteristics—including age, menopause status, tumor size, histopathological type, ER/PR positivity rate, HER2 positivity rate, and molecular type—did not differ significantly between the ALNM groups (>2 vs. ≤2 nodes). A significantly higher proportion of T2 stage tumors and Ki-67 positivity was found in the group with ALNM > 2 nodes, which was consistent with clinical observations. A higher SLNB rate was observed in the group with ALNM ≤ 2 nodes.

Table 1

Table 1. Clinical and pathological characteristics.

The final model, which incorporated five kinetic and three radiomic features, achieved the highest AUC. On the validation set, it yielded an AUC of 0.689, with a sensitivity of 68.0%, a specificity of 71.2%, a PPV of 36.2%, and a NPV of 90.2%. Performance on the test set was comparable, with an AUC of 0.705, a sensitivity of 72.7%, a specificity of 77.8%, a PPV of 44.4%, and a NPV of 92.1%. The diagnostic statistics and selected features are summarized in Tables 2 and 3, respectively. The corresponding ROC curves are shown in Figure 3.

Table 2

Table 2. The performance of the kinetic and radiomic feature model in predicting ALNM burden (≤2 or >2 nodes).

Table 3

Table 3. The selected features and their coefficients in the model.

Figure 3

Figure 3. Receiver operating characteristic (ROC) curves of kinetic and radiomic feature model.

After excluding DCIS cases, 135 patients had ≤2 metastatic ALNs, and 36 patients had >2 metastatic ALNs (clinicopathological characteristics in Supplementary Table S2). The final model on the test set, which incorporated three kinetic and three radiomic features, achieved an AUC of 0.714. However, the predictive performance showed a decrease accordingly, with a sensitivity of 63.6%, a specificity of 57.5%, a PPV of 29.2%, and a NPV of 85.2%. The selected features are listed in Supplementary Table S3, and the corresponding ROC curves are shown in Supplementary Figure S1.

4 Discussion

In this study, the kinetic and radiomic features of tumors demonstrated moderate accuracy (AUC: 0.705 and 0.689 in the test and validation sets, respectively) to differentiate ALNM burden (≤2 or >2 nodes) in T1 and T2 stage breast cancer. However, the kinetic and radiomic features of the tumors exhibited a high NPV (92.1% or 90.2% in the test dataset and validation dataset, respectively), demonstrating their potential as a reliable rule-out test. The high NPV particularly supported their utility as a non-invasive tool to identify candidates for less invasive axillary procedures.

Our result was similar to that reported by Han Lu et al. (7), who demonstrated that a radiomic signature on MRI could moderately predict ALNM burden (≤2 or >2 nodes) (AUC of 0.79). However, our model’s performance was lower than that in another study (13), which reported an AUC of 0.88 for differentiating patients with one to two metastatic nodes from those with ≥3 metastatic nodes. A key methodological difference in our study was the incorporation of quantitative kinetic features alongside radiomic features. However, this addition did not enhance the overall predictive performance in our study. This result was different from that of another study (8), which found that combining kinetic and radiomic features yielded the highest performance for predicting the presence of ALNM (AUC of 0.91), outperforming the use of radiomic features alone. The discrepancy in conclusions may be attributed to a difference in endpoints: their combined model predicted ALNM status (positive/negative), whereas ours aimed to stratify by specific nodal burden (≤2 or >2 nodes).

Furthermore, as noted in the Introduction, studies using kinetic parameters to predict ALNM in breast cancer have yielded conflicting results (17–24). Our previous study (24) showed that quantitative kinetic parameters of the tumors with histogram and texture features were not associated with ALNM in 94 patients. The current study, despite its larger sample size and five kinetic features to be included in the model, still demonstrated moderate predictive performance. The distinct differences in the tumor microenvironment between the ALN and primary tumor site may account for this discrepancy, as suggested by prior research (10). Therefore, future efforts to improve MRI-based diagnostic accuracy for ALNM may benefit from focusing on kinetic or radiomic features derived directly from the ALNs themselves.

Indeed, several studies (26–28) using radiomic and kinetic features of ALNs have shown relatively high diagnostic performance for ALNM, with AUC values of 0.88, 0.86, and 0.979, respectively. Two previous studies (10, 12) also found that the radiomic signature of the ALN on MRI achieved the best performance in predicting SLN or ALN status, with AUC values of 0.906 and 0.85, respectively. Moreover, their analysis revealed that the ALN-derived radiomic signature outperformed both the combined (tumor+ALN) signature and the tumor signature alone in the test or validation cohorts. A fundamental obstacle in using ALN-derived features stems from the clinical dilemma of node-to-node matching. Although previous studies have demonstrated promising results, their inherent selection bias cannot be overlooked. For instance, to define positive lymph nodes, they applied a dual criterion requiring both a high metastatic burden (n > 8) confirmed by ALND and the presence of up to three highly suspicious lymph nodes on MRI (26, 27). Consequently, while these findings are encouraging, their broad generalizability remains uncertain. That was why we based our study on tumor features, rather than ALN characteristics, to predict ALNM burden.

Compared with previous studies that relied on tumor radiomics to differentiate metastatic ALNs from non-metastatic ALNs (not ALNM burden), our model demonstrated a comparable AUC to a few of them (7, 10, 12, 16). However, the model demonstrated comparatively lower performance (in terms of AUC, sensitivity, and specificity) than those reported in other studies (8, 9, 11, 13–15). This discrepancy may be attributed to several factors. 1) Feature extraction platform: our study relied on the Omni-Kinetics software, primarily for extracting kinetic features, which resulted in a relatively limited set of radiomic features. In contrast, comprehensive tools such as PyRadiomics or RadCloud are recommended in the field (29). 2) Feature selection method: the majority of studies utilized the least absolute shrinkage selection operator (LASSO) regression for feature selection, while our approach employed ANOVA with stepwise regression. Although this method was interpretable and allowed us to identify a parsimonious set of features, we acknowledged that more contemporary techniques like LASSO regression may offer advantages in handling multicollinearity. 3) Model building: numerous studies have documented improved AUC when clinical or pathological parameters were incorporated with radiomic data (9, 11, 13, 15). Our study focused on developing an imaging-based model and thus did not incorporate clinical parameters. 4) Cohort characteristics: unlike most prior studies, our cohort design —which was confined to T1 and T2 stage breast cancer and stratified by ALNM burden (≤2 or >2 nodes), resulted in an imbalanced dataset (for more details, see Supplementary Table S4). All the above factors also represented barriers to the clinical adoption of radiomics—including segmentation challenges, limited generalizability, and poor reproducibility (30).

This study had certain limitations. First, the manual segmentation may decrease reproducibility. Future studies could benefit from adopting automated detection and segmentation methods. Including spiculations within the segmentation may reduce feature reproducibility and distort radiomic features; however, research by Cama I. et al. has found that segmentation accuracy does not diminish their predictive capability (31). Second, our feature selection employed ANOVA with stepwise regression. We acknowledged that more contemporary techniques like LASSO regression may offer advantages in handling multicollinearity. However, the use of an independent test set for validation provides confidence that our model’s performance is generalizable. Third, the study was restricted to kinetic and radiomic features from MRI. The inclusion of clinical and pathological features (e.g., Ki-67 data) may improve the predictive performance. Fourth, the inclusion of pure DCIS cases may influence model performance, particularly by inflating the NPV. We included them to address a common preoperative dilemma that a core needle biopsy cannot definitively rule out concurrent invasive carcinoma. Given this diagnostic uncertainty, patients with an initial DCIS diagnosis thus represent a clinically relevant population requiring preoperative nodal assessment. Fifth, a high SLNB rate was observed in the group with ALNM ≤ 2 nodes, and its false-negative rate may have introduced potential bias into the results. Finally, this was a retrospective single-center investigation with a limited sample size. Thus, future research involving larger, multi-center cohorts is warranted to validate and extend our findings.

5 Conclusion

The kinetic and radiomic features from DCE-MRI showed potential for predicting ALNM burden (≤2 or >2 nodes) in T1 and T2 stage breast cancer. The high NPV particularly supported their utility as a non-invasive tool to identify candidates for less invasive axillary procedures.

Data availability statement

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

Ethics statement

The studies involving humans were approved by Sichuan cancer hospital’s institutional review board (SCCHEC2015029). The studies were conducted in accordance with the local legislation and institutional requirements. The participants provided their written informed consent to participate in this study.

Author contributions

YL: Data curation, Software, Writing – original draft, Writing – review & editing. CW: Data curation, Software, Writing – original draft, Writing – review & editing. RM: Data curation, Writing – original draft. HL: Software, Validation, Writing – review & editing. XC: Data curation, Writing – original draft. SW: Formal analysis, Software, Writing – review & editing. JR: Conceptualization, Validation, Writing – review & editing. PZ: Conceptualization, Methodology, Supervision, Writing – review & editing. XZ: Conceptualization, Methodology, Project administration, Supervision, Writing – review & editing.

Funding

The author(s) declared that financial support was not received for this work and/or its publication.

Acknowledgments

The authors would like to thank Shaoyu Wang and Ting Yin, PhD (Siemens Healthineers), for the statistical analysis.

Conflict of interest

Author SW was employed by Siemens Healthineers.

The remaining authors declared that this work was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Generative AI statement

The author(s) declare that Generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Supplementary material

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fonc.2025.1700248/full#supplementary-material

References

1. Torre LA, lami F, Siegel RL, Ward EM, and Jemal A. Global cancer in women: burden and trends. Cancer Epidemiol Biomarkers Prev. (2017) 26:444–57. doi: 10.1158/1055-9965.EPI-16-0858

PubMed Abstract | Crossref Full Text | Google Scholar

2. Chen W, Zheng R, Baade PD, Zhang S, Zeng H, Bray F, et al. Cancer statistics in China, 2015. CA Cancer J Clin Actions. (2016) 66:115–32. doi: 10.3322/caac.21338

PubMed Abstract | Crossref Full Text | Google Scholar

3. Liao GS, Chou YC, Hsu HM, Dai MS, and Yu JC. The prognostic value of lymph node status among breast cancer subtypes. Am J Surg. (2015) 209:717–24. doi: 10.1016/j.amjsurg.2014.05.029

PubMed Abstract | Crossref Full Text | Google Scholar

4. Giuliano AE, Ballman KV, McCall L, Beitsch PD, Brennan MB, Kelemen PR, et al. Effect of axillary dissection vs no axillary dissection on 10-year overall survival among women with invasive breast cancer and sentinel node metastasis: the ACOSOG Z0011 (Alliance) randomized clinical trial. JAMA. (2017) 318:918–26. doi: 10.1001/jama.2017.11470

PubMed Abstract | Crossref Full Text | Google Scholar

5. Choi HY, Park M, Seo M, Song E, Shin SY, and Sohn YM. Preoperative axillary lymph node evaluation in breast cancer: current issues and literature review. Ultrasound Q. (2017) 33:6–14. doi: 10.1097/RUQ.0000000000000277

PubMed Abstract | Crossref Full Text | Google Scholar

6. Gillies RJ, Kinahan PE, and Hricak H. Radiomics: images are more than pictures, they are data. Radiology. (2016) 278:563–77. doi: 10.1148/radiol.2015151169

PubMed Abstract | Crossref Full Text | Google Scholar

7. Han L, Zhu Y, Liu Z, Yu T, He C, Jiang W, et al. Radiomic nomogram for prediction of axillary lymph node metastasis in breast cancer. Eur Radiol. (2019) 29:3820–29. doi: 10.1007/s00330-018-5981-2

PubMed Abstract | Crossref Full Text | Google Scholar

8. Chai R, Ma H, Xu M, Arefan D, Cui X, Liu Y, et al. Differentiating axillary lymph node metastasis in invasive breast cancer patients: A comparison of radiomic signatures from multiparametric breast MR sequences. J Magn Reson Imaging. (2019) 50:1125–32. doi: 10.1002/jmri.26701

PubMed Abstract | Crossref Full Text | Google Scholar

9. Tan H, Gan F, Wu Y, Zhou J, Tian J, Lin Y, et al. Preoperative prediction of axillary lymph node metastasis in breast carcinoma using radiomics features based on the fat-suppressed T2 sequence. Acad Radiol. (2020) 27:1217–25. doi: 10.1016/j.acra.2019.11.004

PubMed Abstract | Crossref Full Text | Google Scholar

10. Yu Y, Tan Y, Xie C, Hu Q, Ouyang J, Chen Y, et al. Development and validation of a preoperative magnetic resonance imaging radiomics-based signature to predict axillary lymph node metastasis and disease-free survival in patients with early-stage breast cancer. JAMA Netw Open. (2020) 3:e2028086. doi: 10.1001/jamanetworkopen.2020.28086

PubMed Abstract | Crossref Full Text | Google Scholar

11. Wang C, Chen X, Luo H, Liu Y, Meng R, Wang M, et al. Development and internal validation of a preoperative prediction model for sentinel lymph node status in breast cancer: Combining radiomics signature and clinical factors. Front Oncol. (2021) 11:754843. doi: 10.3389/fonc.2021.754843

PubMed Abstract | Crossref Full Text | Google Scholar

12. Ma M, Jiang Y, Qin N, Zhang X, Zhang Y, Wang X, et al. A radiomics model for preoperative predicting sentinel lymph node metastasis in breast cancer based on dynamic contrast-enhanced MRI. Front Oncol. (2022) 12:884599. doi: 10.3389/fonc.2022.884599

PubMed Abstract | Crossref Full Text | Google Scholar

13. Chen W, Lin G, Kong C, Wu X, Hu Y, Chen M, et al. Non-invasive prediction model of axillary lymph node status in patients with early-stage breast cancer: a feasibility study based on dynamic contrast-enhanced-MRI radiomics. Br J Radiol. (2024) 97:439–50. doi: 10.1093/bjr/tqad034

PubMed Abstract | Crossref Full Text | Google Scholar

14. Wang Q, Lin Y, Ding C, Guan W, Zhang X, Jia J, et al. Multi-modality radiomics model predicts axillary lymph node metastasis of breast cancer using MRI and mammography. Eur Radiol. (2024) 34:6121–31. doi: 10.1007/s00330-024-10638-2

PubMed Abstract | Crossref Full Text | Google Scholar

15. Li X, Hong M, Lu Z, Liu Z, Lin L, and Xu H. Radiomics models to predict axillary lymph node metastasis in breast cancer and analysis of the biological significance of radiomic features. Front Oncol. (2025) 15:1546229. doi: 10.3389/fonc.2025.1546229

PubMed Abstract | Crossref Full Text | Google Scholar

16. Zhang D, Shen M, Zhang L, He X, and Huang X. Establishment of an interpretabl e MRI radiomics-based machine learning model capable of predicting axillary lymph node metastasis in invasive breast cancer. Sci Rep. (2025) 15:26030. doi: 10.1038/s41598-025-10818-0

PubMed Abstract | Crossref Full Text | Google Scholar

17. Sun K, Zhu H, Chai W, Zhan Y, Nickel D, Grimm R, et al. Whole-lesion histogram and texture analyses of breast lesions on inline quantitative DCE mapping with CAIPIRINHA-Dixon-TWIST-VIBE. Eur Radiol. (2020) 30:57–65. doi: 10.1007/s00330-019-06365-8

PubMed Abstract | Crossref Full Text | Google Scholar

18. Liu F, Wang M, and Li H. Role of perfusion parameters on DCE-MRI and ADC values on DWMRI for invasive ductal carcinoma at 3. 0 Tesla. World J Surg Oncol. (2018) 16:239. doi: 10.1186/s12957-018-1538-8

PubMed Abstract | Crossref Full Text | Google Scholar

19. Wang C, Wei W, Santiago L, Whitman G, and Dogan B. Can imaging kinetic parameters of dynamic contrast-enhanced magnetic resonance imaging be valuable in predicting clinicopathological prognostic factors of invasive breast cancer? Acta Radiol. (2018) 59:813–21. doi: 10.1177/0284185117740746

PubMed Abstract | Crossref Full Text | Google Scholar

20. Loiselle CR, Eby PR, DeMartini WB, Peacock S, Bittner N, Lehman CD, et al. Dynamic contrast-enhanced MRI kinetics of invasive breast cancer: a potential prognostic marker for radiation therapy. Int J Radiat Oncol Biol Phys. (2010) 76:1314–9. doi: 10.1016/j.ijrobp.2009.03.053

PubMed Abstract | Crossref Full Text | Google Scholar

21. Loiselle C, Eby PR, Kim JN, Calhoun KE, Allison KH, Gadi VK, et al. Preoperative MRI improves prediction of extensive occult axillary lymph node metastases in breast cancer patients with a positive sentinel lymph node biopsy. Acad Radiol. (2014) 21:92–8. doi: 10.1016/j.acra.2013.10.001

PubMed Abstract | Crossref Full Text | Google Scholar

22. Choi EJ, Youk JH, Choi H, and Song JS. 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. doi: 10.1002/jmri.26865

PubMed Abstract | Crossref Full Text | Google Scholar

23. Rahbar H, Conlin JL, Parsian S, DeMartini WB, Peacock S, Lehman CD, et al. Suspicious axillary lymph nodes identified on clinical breast MRI in patients newly diagnosed with breast cancer: can quantitative features improve discrimination of Malignant from benign? Acad Radiol. (2015) 22:430–8. doi: 10.1016/j.acra.2014.10.010

PubMed Abstract | Crossref Full Text | Google Scholar

24. Luo HB, Du MY, Liu YY, Wang M, Qing HM, Wen ZP, et al. Differentiation between luminal A and B molecular subtypes of breast cancer using pharmacokinetic quantitative parameters with histogram and texture features on preoperative dynamic contrast-enhanced magnetic resonance imaging. Acad Radiol. (2020) 27:e35–44. doi: 10.1016/j.acra.2019.05.002

PubMed Abstract | Crossref Full Text | Google Scholar

25. Yankeelov TE, Luci JJ, Lepage M, Li R, Debusk L, Lin PC, et al. Quantitative pharmacokinetic analysis of DCE-MRI data without an arterial input function: a reference region model. Magn Reson Imaging. (2005) 23:519–29. doi: 10.1016/j.mri.2005.02.013

PubMed Abstract | Crossref Full Text | Google Scholar

26. Schacht DV, Drukker K, Pak I, Abe H, and Giger ML. Using quantitative image analysis to classify axillary lymph nodes on breast MRI: a new application for the Z0011 Era. Eur J Radiol. (2015) 84:392–97. doi: 10.1016/j.ejrad.2014.12.003

PubMed Abstract | Crossref Full Text | Google Scholar

27. Shan YN, Xu W, Wang R, Wang W, Pang PP, and Shen QJ. A nomogram combined radiomics and kinetic curve pattern as imaging biomarker for detecting metastatic axillary lymph node in invasive breast cancer. Front Oncol. (2020) 10:1463. doi: 10.3389/fonc.2020.01463

PubMed Abstract | Crossref Full Text | Google Scholar

28. Luo HB, Liu YY, Wang CH, Qing HM, Wang M, Zhang X, et al. Radiomic features of axillary lymph nodes based on pharmacokinetic modeling DCE-MRI allow preoperative diagnosis of their metastatic status in breast cancer. PloS One. (2021) 16:e0247074. doi: 10.1371/journal.pone.0247074

PubMed Abstract | Crossref Full Text | Google Scholar

29. Motiei M, Mansouri SS, Tamimi A, Farokhi S, Fakouri A, Rassam K, et al. The diagnostic accuracy of MRI radiomics in axillary lymph node metastasis prediction: a systematic review and meta-analysis. Int J Surg. (2025) 111:6412–26. doi: 10.1097/JS9.0000000000002588

PubMed Abstract | Crossref Full Text | Google Scholar

30. Huisman M and Akinci D’Antonoli T. What a radiologist needs to know about radiomics, standardization, and reproducibility. Radiology. (2024) 310:e232459. doi: 10.1148/radiol.232459

PubMed Abstract | Crossref Full Text | Google Scholar

31. Cama I, Guzmán A, Campi C, Piana M, Lekadir K, Garbarino S, et al. Segmentation variability and radiomics stability for predicting triple-negative breast cancer subtype using magnetic resonance imaging. J Med Imaging (Bellingham). (2025) 12:54501. doi: 10.1117/1.JMI.12.5.054501

PubMed Abstract | Crossref Full Text | Google Scholar