Glioma Grade and Molecular Markers: Comparing Machine-Learning Approaches Using VASARI (Visually AcceSAble Rembrandt...

Area of Science:

• Neuro-oncology
• Radiomics
• Machine Learning

Background:

• Gliomas are primary brain tumors with varied molecular subtypes impacting prognosis and treatment.
• Accurate prediction of glioma grade, IDH mutation status, and MGMT methylation is crucial for personalized patient management.
• Visually AcceSAble Rembrandt Images (VASARI) offers a standardized set of radiological features from MRI scans.

Purpose of the Study:

• To evaluate the efficacy of machine learning models trained on VASARI features for predicting glioma grade, IDH mutation status, and MGMT methylation.
• To identify the most effective machine learning model and key VASARI features for these predictions.

Main Methods:

• A retrospective analysis of 107 glioma patients' MRI and molecular data.
• Extraction of 27 VASARI radiological features from MRI scans.
• Training and internal cross-validation of four machine learning models (Random Forest, Elastic-Net, MARS, XGBoost).
• Performance assessment using Area Under the Curve (AUC), sensitivity, and specificity.

Main Results:

• The XGBoost model demonstrated superior performance in predicting glioma grade (AUC=0.978).
• XGBoost also showed predictive capability for IDH mutation status (AUC=0.806) and MGMT methylation (AUC=0.580).
• Key VASARI features like tumor enhancement, location, and size were identified as important predictors.

Conclusions:

• Combining VASARI features with machine learning, particularly XGBoost, shows significant potential for glioma classification.
• While glioma grade prediction is robust, further improvements in sensitivity are needed for IDH mutation and MGMT methylation prediction.
• Larger datasets and advanced AI are recommended for enhanced accuracy in future studies.