Whole-phase DCE-MRI radiomics for differentiating benign and malignant BI-RADS 4 breast lesions: a comparative machine learning study
- 1Department of Radiology, The First Affiliated Hospital of Zhejiang Chinese Medical University (Zhejiang Provincial Hospital of Chinese Medicine), Hangzhou, Zhejiang China
- 2The First School of Clinical Medicine, Zhejiang Chinese Medical University, Hangzhou, Zhejiang China
Abstract
Background: Breast Imaging Reporting and Data System (BI-RADS) category 4 breast lesions have a wide probability of malignancy, with substantial overlap in imaging findings between benign and malignant lesions. Whole-phase dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) radiomics may provide information on tumor enhancement and heterogeneity. This study primarily aimed to evaluate whole-phase DCE-MRI radiomics for differentiating benign from malignant BI-RADS 4 breast lesions and secondarily explored its feasibility for HER-2 status prediction in malignant lesions.
Methods: This retrospective study included 484 women with BI-RADS 4 breast lesions who underwent DCE-MRI at the First Affiliated Hospital of Zhejiang Chinese Medical University between January 2022 and October 2024, including 264 benign and 220 malignant lesions. Among the malignant lesions, 52 were HER-2 positive and 168 were HER-2 negative. Patients were randomly divided into training and internal test sets at a ratio of 7:3. Whole-phase DCE-MRI radiomics features were extracted. Eight machine-learning algorithms were used to build radiomics models, clinical models, and combined models. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, specificity, positive predictive value, negative predictive value, and F1 score. Decision curve analysis (DCA) was performed as an exploratory assessment of potential net benefit.
Results: For differentiating benign from malignant BI-RADS 4 lesions, the support vector machine (SVM) radiomics model achieved the best radiomics performance, with a test-set AUC of 0.893. The combined model further improved the AUC to 0.913. In the secondary exploratory HER-2 analysis, the random forest (RF) radiomics model yielded a test-set AUC of 0.879, whereas the combined model showed a lower AUC of 0.830. Owing to the limited number of HER-2-positive cases and the internal validation design, these HER-2 findings should be interpreted as preliminary and hypothesis-generating.
Conclusions: Whole-phase DCE-MRI radiomics showed favorable internal performance for differentiating benign from malignant BI-RADS 4 breast lesions. The SVM-based combined model performed favorably for the primary malignancy task, whereas the HER-2 analysis should be regarded only as a secondary exploratory observation requiring further validation.
Trial registration: Not applicable. This was a retrospective single-center imaging study and did not prospectively assign participants to any intervention.
Supplementary Information: The online version contains supplementary material available at https://doi.org/10.1186/s12880-026-02609-4.
Related
- Quantitative Pharmacokinetic Analysis of Dynamic Contrast-Enhanced Perfusion in Multiparametric Prostate MRI: An Exploratory Intra-Patient Study Differentiating Malignant Lesions from Normal Tissue
- Computed tomography multi-phase radiomics models for differentiating benign and malignant thyroid nodules
- Role of Serum HE4, Selected Clinical Features and Diagnostic Prediction Models in Differentiating Benign and Malignant Endometrial Lesions in Polish Women: A Prospective Cohort Study
- Ultrasound-based deep learning radiomics with explainable machine learning for predicting benign and malignant thyroid nodules in lymphocytic thyroiditis: a multicenter study
- MRI-based deep-learning radiomics nomogram for differentiating FIGO 2023 stage I from stage II–IV endometrial cancer: a multicenter study
BibTeX
@article{yang2026wholephase,
title = {{Whole-phase DCE-MRI radiomics for differentiating benign and malignant BI-RADS 4 breast lesions: a comparative machine learning study}},
author = {Yantao Yang and Xuewei Ding and Ruixuan Wang and Shiwei Wang and Mengxin Liu and Huan Lu},
journal = {BMC Medical Imaging},
year = {2026},
month = jul,
doi = {10.1186/s12880-026-02609-4},
url = {https://doi.org/10.1186/s12880-026-02609-4},
note = {CC BY-NC-ND 4.0},
}
RIS
TY - JOUR
TI - Whole-phase DCE-MRI radiomics for differentiating benign and malignant BI-RADS 4 breast lesions: a comparative machine learning study
AU - Yantao Yang
AU - Xuewei Ding
AU - Ruixuan Wang
AU - Shiwei Wang
AU - Mengxin Liu
AU - Huan Lu
PY - 2026
DA - 2026/07/29
JO - BMC Medical Imaging
DO - 10.1186/s12880-026-02609-4
UR - https://doi.org/10.1186/s12880-026-02609-4
AB - Background: Breast Imaging Reporting and Data System (BI-RADS) category 4 breast lesions have a wide probability of malignancy, with substantial overlap in imaging findings between benign and malignant lesions. Whole-phase dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) radiomics may provide information on tumor enhancement and heterogeneity. This study primarily aimed to evaluate whole-phase DCE-MRI radiomics for differentiating benign from malignant BI-RADS 4 breast lesions and secondarily explored its feasibility for HER-2 status prediction in malignant lesions. Methods: This retrospective study included 484 women with BI-RADS 4 breast lesions who underwent DCE-MRI at the First Affiliated Hospital of Zhejiang Chinese Medical University between January 2022 and October 2024, including 264 benign and 220 malignant lesions. Among the malignant lesions, 52 were HER-2 positive and 168 were HER-2 negative. Patients were randomly divided into training and internal test sets at a ratio of 7:3. Whole-phase DCE-MRI radiomics features were extracted. Eight machine-learning algorithms were used to build radiomics models, clinical models, and combined models. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, specificity, positive predictive value, negative predictive value, and F1 score. Decision curve analysis (DCA) was performed as an exploratory assessment of potential net benefit. Results: For differentiating benign from malignant BI-RADS 4 lesions, the support vector machine (SVM) radiomics model achieved the best radiomics performance, with a test-set AUC of 0.893. The combined model further improved the AUC to 0.913. In the secondary exploratory HER-2 analysis, the random forest (RF) radiomics model yielded a test-set AUC of 0.879, whereas the combined model showed a lower AUC of 0.830. Owing to the limited number of HER-2-positive cases and the internal validation design, these HER-2 findings should be interpreted as preliminary and hypothesis-generating. Conclusions: Whole-phase DCE-MRI radiomics showed favorable internal performance for differentiating benign from malignant BI-RADS 4 breast lesions. The SVM-based combined model performed favorably for the primary malignancy task, whereas the HER-2 analysis should be regarded only as a secondary exploratory observation requiring further validation. Trial registration: Not applicable. This was a retrospective single-center imaging study and did not prospectively assign participants to any intervention. Supplementary Information: The online version contains supplementary material available at https://doi.org/10.1186/s12880-026-02609-4.
SN - 1471-2342
ER -