Related Experiment Video
Updated: Sep 20, 2025
07:15
Machine Learning Algorithms for Early Detection of Bone Metastases in an Experimental Rat Model
Published on: August 16, 2020
6.9K
Interpretable Machine Learning Models for Differentiating Glioblastoma From Solitary Brain Metastasis Using Radiomics
Xueming Xia1, Wenjun Wu2, Qiaoyue Tan3
1Division of Head & Neck Tumor Multimodality Treatment, Cancer Center, West China Hospital, Sichuan University, Chengdu, China (X.X., Q.G.).
Academic Radiology
|May 28, 2025
Summary
Machine learning models using radiomics features from MRI effectively differentiate glioblastoma from brain metastasis. High-order features significantly improve model accuracy, with gradient boosting models showing superior performance and interpretability.
Area of Science:
- Radiology
- Machine Learning
- Oncology
Background:
- Differentiating glioblastoma (GB) from solitary brain metastasis (SBM) is crucial for treatment planning.
- Accurate diagnosis impacts patient prognosis and therapeutic strategies.
Purpose of the Study:
- To develop and validate interpretable machine learning models for differentiating GB from SBM using radiomics features from contrast-enhanced T1-weighted MRI (CE-T1WI).
- To compare the impact of low-order versus high-order radiomic features on model performance and interpretability.
Main Methods:
- Retrospective analysis of 434 patients with histopathologically confirmed GB or SBM.
- Radiomic feature extraction from CE-T1WI, followed by feature selection using minimum redundancy maximum relevance and least absolute shrinkage and selection operator regression.
- Training and validation of machine learning models (GradientBoost, lightGBm) using low-order and high-order features, with performance assessed by AUC, accuracy, sensitivity, and specificity.
- SHapley Additive Explanations (SHAP) analysis for model interpretability.
Main Results:
- Machine learning models, particularly lightGBm and GradientBoost, demonstrated strong discriminative power with AUCs exceeding 0.9.
- The lightGBm model showed excellent stability and generalizability.
- GradientBoost achieved the highest AUC (0.927) in the validation group, while lightGBm showed the highest test accuracy (86.2%).
- Models utilizing high-order features consistently outperformed those using low-order features across all metrics.
- SHAP analysis provided insights into feature importance and classification contributions.
Conclusions:
- Radiomics-based machine learning models, especially gradient boosting tree-based methods like lightGBm, can effectively distinguish GB from SBM.
- High-order radiomic features significantly enhance model accuracy and robustness.
- SHAP analysis improves the interpretability and transparency of these models for brain tumor classification.
