Integration Of VGG Network For Accurate Brain Tumor Detection Using MRI Scan

Abstract

Brain tumour is one of the most aggressive malignant cancers and early-accurate diagnosis may increase the changes in successful treatment and subsequently improvement in patient's survival. Magnetic Resonance Imaging (MRI) is currently the gold standard diagnostic technique, but manual interpretation becomes slow, unreliable and radiologist skill-dependent when applied to large numbers of images. To overcome these limitations, in this work, we propose an automated deep learning-based brain tumor detection framework using transfer learning with VGG16 and VGG19 models. Both the models were pre-trained and fine-tuned using a brain MRI dataset, and their performance was measured in terms of accuracy, loss, precision, recall and F1-score. The experimental results show a trade-off between performance and efficiency: VGG19 achieves better accuracy at the cost of slower convergence time and heavier computation, whereas VGG16 provides faster convergence rate with less cost. In the interest of transparency and clinical confidence in these predictions, we used explainable AI (XAI) methods – namely Grad-CAM, LIME, and SHAP – to generate visual level as well as feature-level explanations of how each model makes its predictions. The deployed models along with the XAI outputs were implemented with a web-based framework written in Flask, which allows real-time MRI upload, prediction and visualization of interpretability map. We showed the advantage of integrating deep learning, explainability, as well as deployment for robust and interpretable brain tumor diagnostic assistance.