Enhancing brain tumor classification by integrating radiomics and deep learning features: A comprehensive study...

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Oncology
• Machine Learning for Diagnostics

Background:

• Accurate brain tumor classification is crucial for effective treatment planning.
• Distinguishing between Glioma, Meningioma, and Pituitary Tumors can be challenging with standard MRI analysis.
• Radiomics features (RFs) and deep learning features (DFs) offer novel quantitative approaches to image analysis.

Purpose of the Study:

• To evaluate the effectiveness of combining RFs and DFs for brain tumor classification.
• To compare the performance of individual and combined feature sets using various machine learning models.
• To investigate the impact of ensemble learning techniques on classification accuracy.

Main Methods:

• Analysis of 3064 T1-weighted contrast-enhanced brain MRI scans.
• Extraction of RFs using Pyradiomics and DFs via a 3D convolutional neural network (CNN).
• Training and evaluation of multiple machine learning models (SVM, DT, RF, AdaBoost, Bagging, KNN, MLP) with ensemble methods (Stacking, Voting, Boosting), employing LASSO feature selection and cross-validation.

Main Results:

• The combined RFs + DFs approach significantly outperformed individual feature sets.
• Ensemble methods, particularly Boosting, achieved the highest performance with 95.0% accuracy, 0.92 AUC, 88% sensitivity, and 90% specificity.
• Individual RFs and DFs yielded lower AUCs of 0.82 and 0.85, respectively.

Conclusions:

• Integrating RFs and DFs with ensemble learning enhances brain tumor classification accuracy and reliability from MRI data.
• This combined approach demonstrates significant clinical potential for improved diagnosis.
• Future research can further refine generalizability and precision with additional MRI sequences and advanced ML techniques.