Implementation of Convolutional Neural Networks with Transfer Learning for Brain Tumor Detection Using Large-Scale Magnetic Resonance Image Datasets

Fast and accurate brain tumor identification via MRI is crucial for treatment. This research presents a CAD pipeline using CNNs and transfer learning to classify axial T1-weighted brain MRI scans. Using a public dataset of 3,264 scans (glioma, meningioma, pituitary, non-tumor), we preprocessed and augmented the data. We compared a custom CNN with a fine-tuned VGG16. The transfer-learning model achieved superior performance (98% accuracy, 0.98 precision, 0.97 recall, 0.97 F1, 0.99 AUC) with fewer epochs, outperforming the baseline. McNemar’s test confirmed statistical significance (α = 0.05). This suggests pre-trained networks with augmentation offer state-of-the-art results even with limited medical data, paving the way for deployable brain tumor screening tools.