CONFLUX: A Latent Diusion Model for 3D Chest-CT Synthesis with RL Post-Training

2026-07-03

AI Summary

Existing approaches to controllable 3D medical image generation struggle to simultaneously achieve high fidelity, native 3D structural integrity, and precise control over clinical attributes. This work proposes a latent-space 3D diffusion model that first compresses chest CT volumes using a 3D variational autoencoder and then integrates a rectified flow Transformer with adaptive layer normalization to enable structured radiological metadata conditioning in the latent space. To further enhance attribute controllability, the method incorporates an online reinforcement learning post-training stage based on group relative policy optimization. The proposed approach substantially outperforms baseline methods, achieving a three-plane FID of 32.3 compared to 74.6 and improving clinical attribute recognition accuracy by 47%. Additionally, the study releases a large-scale dataset comprising approximately 200,000 synthetic chest CT scans with rich annotations.

Abstract

Controllable generative models of 3D medical images can synthesize volumes with specified clinical attributes, but this demands samples that are simultaneously high-fidelity, natively 3D, and faithful to the requested conditioning. We present CONFLUX, a latent diffusion model for chest computed tomography (CT): a 3D variational autoencoder compresses each volume, and a rectified-flow transformer generates in the latent space. Generation is conditioned on structured radiological metadata (18 abnormality findings, sex, age, and reconstruction kernel) through adaptive layer normalization. The model leads strong volumetric baselines on tri-planar Frechet distance (FID 32.3 vs. 74.6 for MAISI) while exposing direct control over clinical attributes. To strengthen that control we add an online reinforcement-learning post-training stage (group-relative policy optimization) that rewards how reliably a classifier recovers the requested findings from each generated volume. Judged by a separate, independent classifier, post-training removes 47% of the shortfall relative to real-scan reliability. We release the model and a ~200k synthetic chest-CT dataset with conditioning metadata spanning a wide variety of clinical findings.

Problem

• 3D medical image synthesis
• chest CT
• controllable generation
• clinical attribute conditioning
• high-fidelity volumetric imaging

Innovation

• latent diffusion model
• 3D medical image synthesis
• rectified-flow transformer
• reinforcement learning post-training
• controllable generation