Published Paper Details:

The integration of artificial intelligence (AI) into medical imaging has opened new frontiers for accurate, efficient, and scalable brain tumor diagnostics. This study evaluates and compares the performance of Logistic Regression (LR), Random Forest (RF), and Artificial Neural Networks (ANN) in differentiating glioblastoma multiforme (GBM), lymphoma, and metastases using radiomics features derived exclusively from contrast-enhanced T1-weighted MRI scans. A retrospective cohort of 85 patients (62 GBM, 23 lymphoma) was analyzed, with radiomic features extracted via the IBSI-compliant PyRadiomics framework following manual segmentation. To address natural class imbalance, the Synthetic Minority Oversampling Technique (SMOTE) was applied, creating balanced datasets for robust evaluation. Models were trained and validated using stratified 10-fold cross-validation, with performance measured by accuracy, sensitivity, specificity, and area under the ROC curve (AUC). Results demonstrated that ANN consistently outperformed classical methods, achieving 79% accuracy (AUC 0.83) on the original dataset and 87% accuracy (AUC 0.90) on the balanced dataset, compared with RF (74% and 84.5%) and LR (42% and 63%). In multiclass classification, ANN achieved an overall accuracy of 79.4% with a macro-AUC of 0.887, effectively distinguishing GBM, lymphoma, and metastases. These findings confirm the superiority of deep learning in handling complex, high-dimensional imaging data, particularly when supported by preprocessing and data balancing strategies. By demonstrating the strengths and limitations of different approaches, this study highlights the potential of AI-driven radiomics to enhance diagnostic reliability, reduce variability, and improve patient triage in neuro-oncology.