WDM: 3D Wavelet Diffusion Models for High-Resolution Medical Image Synthesis

TL;DR

This paper introduces WDM, a wavelet-based 3D diffusion model for high-resolution medical image synthesis that achieves state-of-the-art results on large 3D datasets using limited GPU memory.

Contribution

The paper presents a novel wavelet-based diffusion framework that scales 3D medical image synthesis to high resolutions efficiently and effectively.

Findings

Achieves state-of-the-art FID and MS-SSIM scores on BraTS and LIDC-IDRI datasets.

Capable of generating high-quality 3D images at 256x256x256 resolution.

Operates efficiently on a single 40 GB GPU.

Abstract

Due to the three-dimensional nature of CT- or MR-scans, generative modeling of medical images is a particularly challenging task. Existing approaches mostly apply patch-wise, slice-wise, or cascaded generation techniques to fit the high-dimensional data into the limited GPU memory. However, these approaches may introduce artifacts and potentially restrict the model's applicability for certain downstream tasks. This work presents WDM, a wavelet-based medical image synthesis framework that applies a diffusion model on wavelet decomposed images. The presented approach is a simple yet effective way of scaling 3D diffusion models to high resolutions and can be trained on a single \SI{40}{\giga\byte} GPU. Experimental results on BraTS and LIDC-IDRI unconditional image generation at a resolution of $128 \times 128 \times 128$ demonstrate state-of-the-art image fidelity (FID) and sample diversity (MS-SSIM) scores compared to recent GANs, Diffusion Models, and Latent Diffusion Models. Our proposed method is the only one capable of generating high-quality images at a resolution of $256 \times 256 \times 256$, outperforming all comparing methods.