A Unified Framework for Brain Tumour Localisation in MRI Images Using Diffusion Models and Advanced Segmentation Techniques

Abstract

The application of artificial intelligence (AI) in medical imaging has significantly en- hanced diagnostic accuracy and efficiency. This thesis proposes a novel framework for the automated detection and segmentation of brain tumors in MRI scans using diffusion mod- els, the Segment Anything Model (SAM), and the Grounding DINO model. The diffusion model generates counterfactual images of the healthy brain, facilitating the identification of anomalies. SAM and Grounding DINO use point and text prompts to accurately seg- ment the tumors. Our proposed pipeline consistently outperforms the individual baseline models, demonstrating high performance across various evaluation metrics. By providing multiple output images, this system aids radiologists in making more informed decisions. Crucially, this framework is intended to assist, not replace, medical professionals, en- hancing their diagnostic capabilities and supporting improved patient outcomes.