Fine Tune LLm Model for MRI Report Generation

The MRI Report Generation System is a graduation-level, production-oriented deep learning application that automates the end-to-end clinical workflow for brain tumor analysis from MRI scans. Given a raw MRI image, the system:

Classifies the tumor type (Glioma, Meningioma, Pituitary, or No Tumor) using a fine-tuned Xception CNN.

Segments the tumor region pixel-by-pixel using a custom UNet architecture.

Generates a structured radiological report (TECHNIQUE / FINDINGS / IMPRESSION) using a LoRA-fine-tuned DeepSeek-R1 language model.

Answers follow-up clinical questions through a RAG-powered chatbot backed by Cohere embeddings and Groq's Llama-3.3-70B.

The system is served via a unified FastAPI REST API that orchestrates all components and delivers results — including base64-encoded overlay images and structured JSON — to any client application.