Translating MRI to PET through Conditional Diffusion Models with Enhanced Pathology Awareness

This paper presents PASTA, an AI system that generates synthetic PET scans from MRI images to help diagnose neurodegenerative diseases like Alzheimer's. PET scans are expensive and involve radiation, while MRI is safer and cheaper but less sensitive for diagnosis; PASTA bridges this gap by using advanced diffusion models that preserve both the structural details and disease patterns visible in the original image. The synthetic PET scans produced by PASTA achieve diagnostic accuracy nearly matching real PET scans, improving diagnosis rates by 4% over MRI alone.