Hybrid Radiomics and Machine Learning for Brain Tumors Multi-Task Classification: An Exploratory Study on Integrating...

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Medicine
• Computational Pathology

Background:

• Accurate brain tumor classification is crucial for treatment planning.
• Manual MRI interpretation is subjective and time-consuming.
• This study explores machine learning for objective tumor classification.

Purpose of the Study:

• To develop and evaluate a machine learning model for classifying glioblastomas (GBM), meningiomas, brain metastases, and normal controls (NC).
• To integrate radiomic features from contrast-enhanced T1-weighted MRI scans for improved classification accuracy.

Main Methods:

• Utilized a dataset of 154 patients and 170 normal controls.
• Applied data augmentation to address class imbalance.
• Extracted radiomic features using Gray-Level Co-occurrence Matrix (GLCM) and Curvelet transformations.
• Performed feature selection with Least Absolute Shrinkage and Selection Operator (LASSO) and Principal Component Analysis (PCA).
• Trained and evaluated nine machine learning classifiers.

Main Results:

• Random Forest and CatBoost achieved the highest accuracy (95.2%) using LASSO-selected features.
• Combining GLCM and Curvelet features outperformed individual feature sets.
• PCA-based dimensionality reduction also showed strong performance.

Conclusions:

• Integrating Curvelet and GLCM radiomics with optimized machine learning models significantly improves diagnostic accuracy for brain tumors.
• This approach offers a valuable tool to assist radiologists in accurate tumor classification.