Multi-Network Stacking Architecture for Enhanced Brain Tumor Detection in Magnetic Resonance Images

Brain tumors present a significant health challenge, making accurate detection and classification via MRI essential for effective diagnosis and treatment. This paper explores a stacking ensemble of 16 neural networks to enhance classification performance across 16 tumor classes, aggregating outputs through maximum selection, averaging, and top-k methods to create a robust feature matrix. We evaluated two meta-classifiers, Random Forest and XGBoost, on these aggregated features, revealing improved classification accuracy. Additionally, we developed a Convolutional Neural Network (CNN) as an alternative meta-classifier, which achieved 100.00% accuracy on a test dataset of 460 images, demonstrating its effectiveness in synthesizing base model outputs. Overall, this research highlights the potential of stacking ensembles and neural network architectures to advance MRI-based brain tumor classification, contributing to more accurate and efficient clinical decision-making.