Predicting the molecular subtype of breast cancer and identifying interpretable imaging features using machine
Mengwei Ma1, Renyi Liu1, Chanjuan Wen1
1Department of Radiology, Nanfang Hospital, Southern Medical University, Guangzhou, 510515, Guangdong, China.
European Radiology
|October 14, 2021
Summary
An interpretable machine learning model effectively predicts breast cancer molecular subtypes, aiding radiologists in diagnosis. This tool enhances accuracy for triple-negative and Luminal breast cancer detection.
Area of Science:
- Radiology and Medical Imaging
- Artificial Intelligence in Medicine
- Oncology
Background:
- Accurate breast cancer molecular subtyping is crucial for treatment selection.
- Interpretable machine learning (ML) offers potential for improving diagnostic accuracy.
Purpose of the Study:
- To evaluate the performance of interpretable machine learning models in predicting breast cancer molecular subtypes.
- To assess the impact of an interpretable ML model on radiologist performance.
Main Methods:
- Retrospective analysis of 600 invasive breast carcinoma cases.
- Development and evaluation of five ML models using clinical and imaging data (mammography, ultrasonography).
- Application of Shapley Additive Explanations (SHAP) for model interpretability.
Main Results:
- The decision tree (DT) model achieved high performance (AUC 0.971) in distinguishing triple-negative breast cancer (TNBC).
- Radiologist performance significantly improved with DT model assistance for TNBC and Luminal subtype diagnosis.
- Average sensitivity, specificity, and accuracy increased for both less experienced and more experienced radiologists.
Conclusions:
- An interpretable ML model can effectively differentiate breast cancer molecular subtypes.
- SHAP analysis identified key imaging features for subtype prediction.
- The ML model serves as a valuable assistive tool for radiologists.
