Development and validation of a DCE-MRI radiomics-based machine learning model for predicting HER-2 status in breast cancer
10.13491/j.issn.1004-714X.2025.06.005
- VernacularTitle:基于DCE-MRI影像组学构建并验证预测乳腺癌HER-2状态的机器学习模型
- Author:
Yan ZHANG
1
;
Zhijian ZHU
1
;
Jihua HAN
1
;
Honglei LUO
1
;
Yaqi SONG
1
;
Wei HUANG
2
Author Information
1. Department of Radiotherapy Oncology, The Affiliated Huaian No.1 People's Hospital of Nanjing Medical University, Huai'an 223300, China.
2. Department of Medical Imaging, The Affiliated Huaian No.1 People's Hospital ofNanjing Medical University, Huai'an 223300, China.
- Publication Type:OriginalArticles
- Keywords:
Breast cancer;
HER-2;
Radiomics;
Machine learning;
Predictive model;
Dynamic contrast-enhanced MRI;
XGBoost
- From:
Chinese Journal of Radiological Health
2025;34(6):811-818
- CountryChina
- Language:Chinese
-
Abstract:
Objective To analyze dynamic contrast-enhanced MRI (DCE-MRI) radiomic features using machine learning algorithms, and to develop and validate a predictive model for HER-2 status in breast cancer. Methods The DCE-MRI images of 272 treatment-naive female patients with breast cancer between 2020 and 2022 were included in this study. Regions of interest (ROIs) were manually segmented using 3d-Slicer software, and radiomic features were extracted. All patients were randomly divided into training sets or validation sets at a ratio of 4∶1. The least absolute shrinkage and selection operator (LASSO) algorithm was used for feature screening on the training set, followed by the development of predictive models using six machine learning algorithms. Internal cross-validation was performed to compare the performance differences between the models. The best-performing model was selected, trained on the training set, and evaluated on the validation set. Evaluation metrics included area under the curve (AUC), sensitivity, specificity, precision, and recall rate. Results The clinical data of patients in the training set and validation set showed no significant differences. Five features were identified by the LASSO algorithm. With these features, six machine learning models were developed on the training set, and their predictive performance was internally cross-validated using the bagging method. XGBoost model had the highest mean AUC (0.696), followed by RF model (0.690); XGBoost model had the highest mean precision (0.756), followed by LR and RF models. Therefore, XGBoost was the optimal model. An HER-2 predictive model was built using the XGBoost algorithm on the training set and applied to the validation set. The AUC, precision, sensitivity, and specificity of the predictive model on the validation set were calculated, and ROC curves, precision-recall curves, calibration curves, and decision-making curves were plotted. Conclusion This study constructed and evaluated different DCE-MRI radiomics-based machine learning models for predicting HER-2 status in breast cancer. Among them, XGBoost algorithm performed the best and has the potential to become a new non-invasive method for preoperative prediction of HER-2 status, providing reliable evidence for personalized clinical diagnosis and treatment.
