A Deep Learning-Based MRI-to-CT Image Translation Using Generative Models

Abstract

Magnetic Resonance Imaging (MRI) and Computed Tomography (CT) are both widely used medical imaging, they provide complementary diagnostic clues. Still getting both of them at the same time, ends up taking more time, more money, and for CT also involves radiation exposure, which is not small. Lately, deep learning has really pushed things forward, with cross-modality synthesis where a CT image can be created straight from an MRI scan. In this paper, there is a survey of deep learning based MRI-to-CT translation strategies, with attention on “generate it” style methods such as Generative Adversarial Networks (GANs) and diffusion models. The discussion goes into how models trade accuracy versus realism, plus what kinds of optimization tricks are usually used and what limitations show up in real practice. At the end, it also shares thoughts on upcoming research directions, like how to make the approach more stable and easier to deploy.