Leveraging Vision Transformers and Diffusion Models for Medical Imaging Diagnosis- A Study on Weakly Supervised and Zero-Shot Learning for Rare Diseases

The emergence of Vision Transformers (ViTs) and diffusion models has transformed the landscape of computer vision, particularly in medical imaging diagnosis. This research examines the utilization of ViTs and diffusion models for diagnosing conditions through X-rays, MRI, and CT scans, specifically emphasizing rare diseases. It investigates the effectiveness of weakly supervised and zero- shot learning approaches to tackle the challenges associated with the limited availability of annotated data for uncommon pathologies. The study introduces an innovative hybrid architecture combining ViTs and diffusion models, thereby improving accuracy and generalization. Experimental findings indicate enhanced diagnostic capabilities, a decreased dependence on extensive labelled datasets, and promising potential for application in clinical environments.