RealDeal: Enhancing Realism and Details in Brain Image Generation via Image-to-Image Diffusion Models

TL;DR

This paper introduces RealDeal, an image-to-image diffusion approach that refines brain MRI images generated by latent diffusion models, adding realistic details, textures, and noise to improve their authenticity.

Contribution

The work presents a novel image-to-image diffusion framework that enhances the realism and detail of brain images generated by existing models, addressing their smoothness and lack of fine features.

Findings

Improved image realism as measured by FID and LPIPS.

Enhanced anatomical detail and noise in generated images.

New metrics demonstrating increased realism in texture and noise distribution.

Abstract

We propose image-to-image diffusion models that are designed to enhance the realism and details of generated brain images by introducing sharp edges, fine textures, subtle anatomical features, and imaging noise. Generative models have been widely adopted in the biomedical domain, especially in image generation applications. Latent diffusion models achieve state-of-the-art results in generating brain MRIs. However, due to latent compression, generated images from these models are overly smooth, lacking fine anatomical structures and scan acquisition noise that are typically seen in real images. This work formulates the realism enhancing and detail adding process as image-to-image diffusion models, which refines the quality of LDM-generated images. We employ commonly used metrics like FID and LPIPS for image realism assessment. Furthermore, we introduce new metrics to demonstrate the realism of images generated by RealDeal in terms of image noise distribution, sharpness, and texture.