Interpretable Brain Tumor Detection Using VGG16 and Grad-CAM Visualization

Aaqib Rashid Mir, Bachcha Lal Pal

Abstract: Accurate and interpretable detection of brain tumors from MRI scans remains a critical task in medical imaging. In this paper, we present a deep learning-based diagnostic system leveraging a pre-trained VGG16 model with Grad-CAM for explainability. We fine-tuned the VGG16 architecture using a publicly available MRI dataset containing four tumor classes Glioma, Meningioma, Pituitary, and No Tumor. The model achieved a test accuracy of 98.5%, surpassing several existing benchmarks. Grad-CAM visualizations provided insights into the decision-making process of the model, enhancing its trustworthiness for clinical use. Our method addresses the challenge of black-box predictions in AI by offering a highly accurate and transparent brain tumor detection framework. Future improvements include ensemble learning, tumor segmentation, and real-time deployment. The proposed system is deployed on Hugging Face as a real-time web application using Streamlit to facilitate practical and interpretable usage.

Keywords: Brain Tumor Detection, MRI Images, Deep Learning, VGG16, Grad-CAM, Convolutional Neural Network (CNN), Medical Image Analysis, Interpretable AI

How to Cite?: Aaqib Rashid Mir, Bachcha Lal Pal, "Interpretable Brain Tumor Detection Using VGG16 and Grad-CAM Visualization", Volume 14 Issue 7, July 2025, International Journal of Science and Research (IJSR), Pages: 927-932, https://www.ijsr.net/getabstract.php?paperid=SR25714205134, DOI: https://dx.doi.org/10.21275/SR25714205134