Unified Image-to-Image Generation for Diverse Medical Vision Tasks

Abstract

This paper introduces a novel framework that unifies various medical vision tasks, including synthesis, segmentation, denoising, and inpainting, into a single image-to-image generation process. By treating these tasks as conditional image generation problems, the proposed approach enables a generalist model to handle diverse inputs and outputs across different modalities and datasets. The effectiveness of this unification strategy is demonstrated through a comprehensive evaluation on a curated medical vision benchmark, showcasing its potential to simplify and enhance medical image analysis.