Explainable Brain Tumor Diagnosis Using Fine Tuned VGG16 Convolutional Neural Networks

Abstract –This paper presents a deep learning-based framework for automatic brain tumor detection using MRI images, combining transfer learning with AI. A pre-trained VGG16 model was fine-tuned to classify MRI scans into four categories: Glioma, Meningioma, Pituitary Adenoma, and No Tumor. Grayscale MRI inputs were preprocessed and converted to threechannel format to align with the VGG16 architecture. Class imbalance was addressed through data augmentation and class weighting. GradCAM visualizations were employed to provide interpretable results, highlighting tumor regions. A userfriendly Streamlit-based UI and a Flask REST API were developed to support interactive and scalable deployment. The system's effectiveness was assessed using metrics such as accuracy, precision, recall, F1-score, and a confusion matrix, all of which indicated strong performance and promising clinical applicability. This modular solution offers a robust tool for medical image analysis and future research expansion. Keywords—Brain Tumor Classification, Magnetic Resonance Imaging, VGG16 Architecture, GradCAM Visualization, Streamlit Interface, Flask Web API