Publication Date

Spring 2026

Degree Type

Master's Project

Degree Name

Master of Science in Computer Science (MSCS)

Department

Computer Science

First Advisor

Navrati Saxena

Second Advisor

Saptarshi Sengupta

Third Advisor

Pooja Shyamsundar

Keywords

Agentic AI, Explainable AI (XAI), Brain Tumor MRI, CNN Classification, Tumor Localization, Retrieval-Augmented Generation (RAG), Clinical Decision Support Systems, Prognosis Prediction, Treatment Recommendation

Abstract

Gliomas are one of the most aggressive primary brain tumors, and the only way to deal with them is early diagnosis and personalized treatment planning which could improve overall survival rate. Current studies using AI to solve this problem largely focus on classification or segmentation. They lack comprehensive prognosis analysis, treatment recommendation, and explainability for diagnostic decision support. In this study, we aim to develop an end-to-end explainable agentic AI system that combines deep learning and retrieval augmented system to perform glioma subtype classification from MRI images, tumor localization using heatmaps like Score-CAM and Grad-CAM, similar patient retrieval using FAISS based embedding search ultimately leading to prognosis and personalized treatment recommendation using a large language model. On the MU-Glioma-Post dataset, which represents multimodal and longitudinal 3D MRI data with abundant clinical information, this study has been able to develop an automated diagnostic model with a classification rate of 80.79%, and the study has shown the possibility of generating evidence-based glioma prognosis and treatment narratives using the decision support tool as a proof of concept

Download

Available for download on Monday, May 24, 2027

Share

COinS