Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

LAND: Lung and Nodule Diffusion for 3D Chest CT Synthesis with Anatomical Guidance

Anna Oliveras, Roger Marí, Rafael Redondo, Oriol Guardià, Ana Tost, Bhalaji Nagarajan, Carolina Migliorelli, Vicent Ribas, Petia Radeva

TL;DR

The paper tackles data scarcity in 3D chest CT imaging by introducing LAND, a memory-efficient latent diffusion model that synthesizes $256^3$ CT volumes at 1 mm isotropic resolution conditioned on lung and nodule masks. LAND encodes volumes with a 3D VAE to a latent space of size $64\times64\times64\times4$ and performs diffusion with a 3D U-Net, incorporating cross-attention to conditioning masks and a Min-SNR loss; it operates on a single 20 GB GPU and supports unseen NLST-derived masks with $T=1000$ timesteps. Key findings show that conditioning on combined lung and nodule masks yields anatomically plausible outputs, while nodules-only conditioning leads to implausible placements; the method achieves favorable fidelity (FID) and competitive diversity (MS-SSIM) relative to state-of-the-art baselines. The approach holds practical value for data augmentation and clinical workflows, with future work including downstream task evaluation and a mask-generation module to broaden anatomical diversity.

Abstract

This work introduces a new latent diffusion model to generate high-quality 3D chest CT scans conditioned on 3D anatomical masks. The method synthesizes volumetric images of size 256x256x256 at 1 mm isotropic resolution using a single mid-range GPU, significantly lowering the computational cost compared to existing approaches. The conditioning masks delineate lung and nodule regions, enabling precise control over the output anatomical features. Experimental results demonstrate that conditioning solely on nodule masks leads to anatomically incorrect outputs, highlighting the importance of incorporating global lung structure for accurate conditional synthesis. The proposed approach supports the generation of diverse CT volumes with and without lung nodules of varying attributes, providing a valuable tool for training AI models or healthcare professionals.

LAND: Lung and Nodule Diffusion for 3D Chest CT Synthesis with Anatomical Guidance

TL;DR

The paper tackles data scarcity in 3D chest CT imaging by introducing LAND, a memory-efficient latent diffusion model that synthesizes CT volumes at 1 mm isotropic resolution conditioned on lung and nodule masks. LAND encodes volumes with a 3D VAE to a latent space of size and performs diffusion with a 3D U-Net, incorporating cross-attention to conditioning masks and a Min-SNR loss; it operates on a single 20 GB GPU and supports unseen NLST-derived masks with timesteps. Key findings show that conditioning on combined lung and nodule masks yields anatomically plausible outputs, while nodules-only conditioning leads to implausible placements; the method achieves favorable fidelity (FID) and competitive diversity (MS-SSIM) relative to state-of-the-art baselines. The approach holds practical value for data augmentation and clinical workflows, with future work including downstream task evaluation and a mask-generation module to broaden anatomical diversity.

Abstract

This work introduces a new latent diffusion model to generate high-quality 3D chest CT scans conditioned on 3D anatomical masks. The method synthesizes volumetric images of size 256x256x256 at 1 mm isotropic resolution using a single mid-range GPU, significantly lowering the computational cost compared to existing approaches. The conditioning masks delineate lung and nodule regions, enabling precise control over the output anatomical features. Experimental results demonstrate that conditioning solely on nodule masks leads to anatomically incorrect outputs, highlighting the importance of incorporating global lung structure for accurate conditional synthesis. The proposed approach supports the generation of diverse CT volumes with and without lung nodules of varying attributes, providing a valuable tool for training AI models or healthcare professionals.
Paper Structure (4 sections, 3 figures, 1 table)

This paper contains 4 sections, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. Method
  3. Experimental Results
  4. Conclusion

Figures (3)

  • Figure 1: LAND uses a 3D VAE to encode CT volumes into a latent space, where a 3D U-Net performs diffusion on the latent samples $\mathbf{x}$, optionally conditioned on anatomical masks $\mathbf{m}$ (lungs/nodules).
  • Figure 2: Comparison of unconditional (left) and conditional (right) CT generation using LAND and baseline methods PatchDDM bieder2024memory and WDM friedrich2024wdm. Mask overlays are shown where applicable.
  • Figure 3: LAND samples conditioned on nodule+lung+texture masks, with increasing texture scores.