Identification of Molecular Subtypes of Breast Cancer Using Machine Learning Models Based on Multimodal MRI
10.3969/j.issn.1005-5185.2025.10.004
- VernacularTitle:基于多模态MRI的机器学习模型识别乳腺癌分子亚型
- Author:
Mengying XU
1
;
Pan ZHANG
1
;
Chunhua LI
1
;
Jian LI
1
;
Zihan HONG
1
;
Bing CHEN
1
Author Information
1. 宁夏医科大学总医院放射科,宁夏 银川 750004
- Publication Type:Journal Article
- Keywords:
Breast neoplasms;
Magnetic resonance imaging;
Synthetic magnetic resonance;
Dynamic contrast-enhanced imaging;
Diffusion weighted imaging;
Machine learning;
Diagnosis,differential
- From:
Chinese Journal of Medical Imaging
2025;33(10):1043-1048,1055
- CountryChina
- Language:Chinese
-
Abstract:
Purpose To explore the value of machine learning models based on synthetic MRI,dynamic contrast-enhanced MRI(DCE-MRI)and diffusion weighted imaging(DWI)parameters in identifying molecular subtypes of breast cancer.Materials and Methods A retrospective analysis was conducted on the data of 292 patients who underwent synthetic MRI,DCE-MRI and DWI examinations from September 2020 to September 2024 in Ningxia Medical University General Hospital before surgery and were pathologically confirmed to have breast cancer postoperatively.Patients were randomly divided into training and test sets using a ratio of 7:3.Multiple parameters were obtained from the synthetic MRI,DCE-MRI and DWI images.Variance analysis were used to screen the characteristic parameters among molecular subtype groups.Five machine learning models were established based on the selected characteristic parameters,and receiver operating characteristic curves were plotted to calculate the area under the curve among the molecular subtype groups.Results The support vector machine model exhibited the highest overall performance,with an area under the curve of 0.972,accuracy of 82.5％,specificity of 94.76％and sensitivity of 82.14％in the test set.This model's area under the curve values for differentiating luminal A,luminal B,human epidermal growth factor receptor-2 overexpression,and triple-negative groups in the training set were 0.979,0.925,0.971 and 0.982,respectively;in the test set,the area under the curve values were 0.973,0.873,0.956 and 0.955,respectively.Conclusion Machine learning models based on multimodal MRI parameters can assist clinicians in preoperatively determining the molecular subtypes of breast cancer and the support vector machine model shows relatively high comprehensive performance.
