Generative models compared for 3D medical image translation

Researchers have conducted a comprehensive evaluation of seven generative models for 3D medical image-to-image translation, comparing GANs against latent generative models across numerous datasets and anatomical regions. The study found that GANs, particularly SRGAN, generally outperformed latent models in synthesizing medical images. A key finding was that while synthetic images were largely indistinguishable from real ones in a Visual Turing test with physicians, quantitative metrics did not fully align with clinical preference, especially concerning the synthesis of small lesions and intensity values. AI

IMPACT This research standardizes evaluation for medical image translation, potentially improving diagnostic accuracy and reducing patient exposure to radiation.

RANK_REASON The cluster contains an academic paper detailing a comparative evaluation of generative models for a specific research task. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• Brownian Bridge
• CycleGAN
• Flow Matching
• Latent Diffusion Model
• Latent Diffusion Model+ControlNet
• Pix2Pix
• SRGAN

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →