Related Experiment Video
Updated: Jun 20, 2026
Tracking the Mammary Architectural Features and Detecting Breast Cancer with Magnetic Resonance Diffusion Tensor Imaging
Published on: December 15, 2014
The Role of Multiparametric MRI Radiomics for Preoperative Prediction of Axillary Lymph Node Metastasis in Patients
Qingcong Kong1, Yongxin Chen2, Yi Sui3
1Department of Radiology, The Third Affiliated Hospital Sun Yat-Sen University Guangzhou Guangdong China.
Background:
The predictive value of different MRI sequences for axillary lymph node metastasis (ALNM) in patients with invasive breast cancer remains unclear. This study compared the performance of radiomics models based on individual and combined MRI sequences for the preoperative prediction of ALNM and evaluated the clinical application value of the optimal model.
Methods:
This retrospective study included 454 patients (mean ± SD age 50.9 ± 10.7 years) diagnosed with invasive breast cancer from two centers, with 382 patients from Center 1 (training cohort) and 72 patients from Center 2 (external test cohort). Tumor segmentation and radiomics feature extraction were performed on T2-weighted imaging (T2WI), diffusion-weighted imaging (DWI), and dynamic contrast-enhanced (DCE) images. The least absolute shrinkage and selection operator with 10-fold cross-validation was used for feature selection and radiomics score construction. Three single-sequence models and one multi-sequence radiomics model were developed, and the optimal model was combined with conventional MRI features to create a combined MRI model. The combined model's performance was compared to radiologists' diagnoses. A nomogram was developed based on the optimal model and correlated with prognosis using the Kaplan-Meier curve and Cox proportional hazard regression. Model performance was evaluated using area under the curve (AUC); DeLong's test was used for comparison.
Results:
In the external test cohort, the DCE model showed the highest performance (AUC = 0.76) but was not significantly different from T2WI (AUC = 0.72) and DWI (AUC = 0.70) (all p > 0.05). The combined radiomics model achieved an AUC of 0.82, outperforming DWI and T2WI (p < 0.05), but was not significantly different from the DCE model (p > 0.05). The combined MRI model demonstrated the highest AUC of 0.84 and notably improved radiologist diagnostic accuracy. A nomogram based on the combined MRI model was developed to assist clinical decision-making by providing individualized risk predictions. The higher-risk group based on the model's predictive probability showed a significantly worse prognosis (p < 0.001).
Conclusion:
The combined radiomics model outperformed single-sequence models in predicting ALNM. The combined MRI model demonstrated the highest performance, improving diagnostic accuracy and showing potential for prognostic prediction.
