Predicting molecular subtypes of breast cancer based on multi-parametric MRI dataset using deep learning method.
Wanqing Ren1, Xiaoming Xi2, Xiaodong Zhang3
1Department of Radiology, Jinan Third People's Hospital, Jinan, China.
Magnetic Resonance Imaging
|December 16, 2024
Summary
A new multi-parametric MRI model accurately predicts breast cancer molecular subtypes, outperforming individual MRI sequences. The Fast-Scan T1-weighted imaging (FS-T1WI) model also shows promise for subtype prediction.
Area of Science:
- Radiology
- Oncology
- Medical Imaging
Background:
- Accurate prediction of breast cancer molecular subtypes is crucial for guiding treatment decisions.
- Preoperative MRI plays a vital role in breast cancer assessment.
- Existing MRI techniques have limitations in differentiating molecular subtypes.
Purpose of the Study:
- To develop and validate a multi-parametric MRI model for predicting breast cancer molecular subtypes.
- To evaluate the performance of individual MRI sequences and a combined model in subtype prediction.
Main Methods:
- Retrospective analysis of MRI data from 325 breast cancer patients.
- Utilized five MRI sequences: FS-T1WI, T2WI, T1-C, DWI, and ADC.
- Developed a multi-parametric model by fusing outputs from five ResNeXt50 base models using ensemble learning.
- Classified subtypes into Luminal A, Luminal B, HER2-positive, and Triple-Negative (TN).
Main Results:
- The multi-parametric MRI model achieved an AUC of 0.859-0.912 in predicting breast cancer molecular subtypes.
- Individual models showed lower AUCs: FS-T1WI (0.632-0.814), T2WI (0.641-0.788), T1-C (0.621-0.709), DWI (0.620-0.701), and ADC (0.611-0.785).
Conclusions:
- The developed multi-parametric MRI model significantly outperformed individual base models in predicting breast cancer molecular subtypes.
- FS-T1WI demonstrated potential as a valuable sequence for breast cancer molecular subtype prediction.
