Graph Attention Networks for Node Classification in Brain Tumor Diagnosis

The detection of brain tumors from MRI scans is crucial for timely and effective medical intervention. Traditional methods based on the convolutional neural network (CNN) are not very effective here, as they can capture spatial relations not of higher order, and they are sensitive to class imbalance in medical data sets. These limitations make them less reliable for identifying tumor regions, particularly those that belong to underrepresented classes. This project proposes a new framework based on Graph Neural Networks (GNNs), specifically Graph Attention Networks (GATs), to address these challenges. By converting MRI scans into superpixel-based graph representations and utilizing advanced graph-level oversampling techniques such as SMOTE and ADASYN, this approach outperforms conventional CNNs in terms of accuracy, F1 scores, and robustness. The proposed framework includes state-of-the-art preprocessing methods, segmentation techniques, and innovative graph-based learning methods, establishing it as the new benchmark for brain tumor detection in medical imaging applications.