Search papers, labs, and topics across Lattice.
The paper introduces CONFLUX, a latent diffusion model designed for synthesizing 3D chest CT images that are both high-fidelity and controllable based on specified clinical attributes. By employing a 3D variational autoencoder for volume compression and a rectified-flow transformer for latent space generation, the model achieves significant improvements in volumetric quality, as evidenced by a tri-planar Frechet distance (FID) of 32.3 compared to 74.6 for existing methods. Additionally, an online reinforcement-learning post-training stage enhances the model's ability to reliably generate images that correspond to requested clinical findings, effectively reducing the reliability gap by 47% when evaluated against real scans.
CONFLUX not only synthesizes high-quality 3D chest CT images but also allows for precise control over clinical attributes, significantly improving reliability in medical imaging.
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.
