Multiparametric MRI Radiomics With Machine Learning for Differentiating HER2-Zero, -Low, and -Positive Breast Cancer:...

Area of Science:

• Oncology
• Radiology
• Machine Learning

Background:

• MRI radiomics is being explored for classifying HER2 expression in breast cancer.
• Understanding the predictive mechanisms of these models is crucial.

Purpose of the Study:

• To develop and test multiparametric MRI radiomics machine learning models for three-tiered HER2 expression classification.
• To interpret model predictions using SHAP analysis.

Main Methods:

• Retrospective study of 737 breast cancer patients from two centers.
• Radiomic features extracted from DCE, T2-weighted, and DWI MRI.
• Support vector machine (SVM) for feature selection and radiomics score (radscore) computation.
• SHAP analysis for local and global model interpretation.

Main Results:

• In the external test set, the combined model achieved an AUC of 0.762 for differentiating HER2-negative from HER2-positive tumors.
• For differentiating HER2-zero from HER2-low tumors, the radscore achieved an AUC of 0.754.
• SHAP analysis highlighted early DCE imaging features as most influential for both tasks.

Conclusions:

• MRI radiomics models demonstrate suboptimal performance for noninvasive HER2 expression characterization.
• SHAP analysis offers valuable insights into the predictions of imaging-based machine learning models.