<strong>CONFLUX is a conditional rectified-flow model for full-volume 3D chest-CT synthesis, with GRPO reinforcement-learning post-training that sharpens control over the requested clinical findings.</strong></p>\n<p>Flow matching trains the generator to match the data distribution but never checks that an individual sample realizes its requested attributes; the GRPO stage optimizes that agreement directly — to our knowledge, the first GRPO post-training of a 3D medical flow model.</p>\n<ul>\n<li>Full 216×176×200 volumes from a 42-dimensional conditioning vector: 18 CT-RATE abnormality findings, sex, age, and reconstruction kernel.</li>\n<li>Outperforms strong 3D baselines (MAISI, GenerateCT) on distribution-level quality.</li>\n<li>GRPO post-training measurably improves how reliably the requested findings appear, verified by an independent held-out judge — recovering 47% of the gap to real scans.</li>\n<li>Release of ~200,000 labeled synthetic chest CTs for cohort augmentation and controlled study design at a scale unavailable in real corpora.</li>\n<li>Interactive demo for generating volumes from arbitrary clinical profiles.</li>\n</ul>\n<p>📄 Paper: <a href=\"https://arxiv.org/abs/2607.02998\" rel=\"nofollow\">https://arxiv.org/abs/2607.02998</a><br>🤗 Model: <a href=\"https://huggingface.co/gevaertlab/conflux\">https://huggingface.co/gevaertlab/conflux</a><br>🤗 Dataset (200k): <a href=\"https://huggingface.co/datasets/gevaertlab/conflux-chest-ct\">https://huggingface.co/datasets/gevaertlab/conflux-chest-ct</a><br>🩻 Demo: <a href=\"https://huggingface.co/spaces/mxvp/conflux-chest-ct-demo\">https://huggingface.co/spaces/mxvp/conflux-chest-ct-demo</a></p>\n<p>Feel free to reach out!</p>\n","updatedAt":"2026-07-07T07:16:58.282Z","author":{"_id":"68210ad4f29d70e1cccc86be","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/peHjJRYBM1-tjTORiyASd.png","fullname":"Max Van Puyvelde","name":"mxvp","type":"user","isPro":true,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.8217685222625732},"editors":["mxvp"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/peHjJRYBM1-tjTORiyASd.png"],"reactions":[],"isReport":false}},{"id":"6a4cdc5534fd55c4f6b9fba5","author":{"_id":"69a0b539f2314740e527db21","avatarUrl":"/avatars/779fb51f480a14ce0acce76d163b1a56.svg","fullname":"Lee Seungju","name":"nugejus","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false},"createdAt":"2026-07-07T11:00:37.000Z","type":"comment","data":{"edited":true,"hidden":false,"latest":{"raw":"You misspelled the title","html":"<p>You misspelled the title</p>\n","updatedAt":"2026-07-07T11:00:51.913Z","author":{"_id":"69a0b539f2314740e527db21","avatarUrl":"/avatars/779fb51f480a14ce0acce76d163b1a56.svg","fullname":"Lee Seungju","name":"nugejus","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false}},"numEdits":1,"identifiedLanguage":{"language":"nl","probability":0.2956317961215973},"editors":["nugejus"],"editorAvatarUrls":["/avatars/779fb51f480a14ce0acce76d163b1a56.svg"],"reactions":[],"isReport":false},"replies":[{"id":"6a4ce5fb6f126cfd7024f8cf","author":{"_id":"68210ad4f29d70e1cccc86be","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/peHjJRYBM1-tjTORiyASd.png","fullname":"Max Van Puyvelde","name":"mxvp","type":"user","isPro":true,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false},"createdAt":"2026-07-07T11:41:47.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"Will be updated in the next arXiv release, thx!","html":"<p>Will be updated in the next arXiv release, thx!</p>\n","updatedAt":"2026-07-07T11:41:47.110Z","author":{"_id":"68210ad4f29d70e1cccc86be","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/peHjJRYBM1-tjTORiyASd.png","fullname":"Max Van Puyvelde","name":"mxvp","type":"user","isPro":true,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.7640962600708008},"editors":["mxvp"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/peHjJRYBM1-tjTORiyASd.png"],"reactions":[],"isReport":false,"parentCommentId":"6a4cdc5534fd55c4f6b9fba5"}}]}],"primaryEmailConfirmed":false,"paper":{"id":"2607.02998","authors":[{"_id":"6a4ca22625849b193a8343e3","user":{"_id":"68210ad4f29d70e1cccc86be","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/peHjJRYBM1-tjTORiyASd.png","isPro":true,"fullname":"Max Van Puyvelde","user":"mxvp","type":"user","name":"mxvp"},"name":"Max Van Puyvelde","status":"claimed_verified","statusLastChangedAt":"2026-07-07T12:10:41.028Z","hidden":false},{"_id":"6a4ca22625849b193a8343e4","name":"Halil Ibrahim Gulluk","hidden":false},{"_id":"6a4ca22625849b193a8343e5","name":"Wim Van Criekinge","hidden":false},{"_id":"6a4ca22625849b193a8343e6","name":"Olivier Gevaert","hidden":false}],"publishedAt":"2026-07-03T00:00:00.000Z","submittedOnDailyAt":"2026-07-07T00:00:00.000Z","title":"CONFLUX: A Latent Diusion Model for 3D Chest-CT Synthesis with RL Post-Training","submittedOnDailyBy":{"_id":"68210ad4f29d70e1cccc86be","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/peHjJRYBM1-tjTORiyASd.png","isPro":true,"fullname":"Max Van Puyvelde","user":"mxvp","type":"user","name":"mxvp"},"summary":"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.","upvotes":3,"discussionId":"6a4ca22725849b193a8343e7","projectPage":"https://huggingface.co/spaces/mxvp/conflux-chest-ct-demo","ai_summary":"A 3D latent diffusion model for chest CT generation that enables controlled synthesis of medical images with clinical attributes through adaptive layer normalization and reinforcement learning post-training.","ai_keywords":["latent diffusion model","3D variational autoencoder","rectified-flow transformer","adaptive layer normalization","group-relative policy optimization","Frechet distance","FID","reinforcement learning","clinical attributes","chest CT","medical image synthesis"],"ai_summary_model":"Qwen/Qwen2.5-Coder-32B-Instruct"},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"666ddefe83571a7a05af7870","avatarUrl":"/avatars/74b88573973b5508e47f5af4044b14a6.svg","isPro":false,"fullname":"Halil Ibrahim Gulluk","user":"gulluk","type":"user"},{"_id":"6997ef2f68950cfdb9f81875","avatarUrl":"/avatars/d99a1f9df211f4162b4e177eded49570.svg","isPro":false,"fullname":"Jcdbzh9olj","user":"jcdbzh9olj","type":"user"},{"_id":"6a2da6c8ca070ee12c6e396c","avatarUrl":"/avatars/0355287dcabaa67dbc7f0b10b87451f9.svg","isPro":false,"fullname":"Joe Mama","user":"JoeMama123123123","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2607/2607.02998.md","query":{}}">
CONFLUX: A Latent Diusion Model for 3D Chest-CT Synthesis with RL Post-Training
Abstract
A 3D latent diffusion model for chest CT generation that enables controlled synthesis of medical images with clinical attributes through adaptive layer normalization and reinforcement learning post-training.
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.
Community
CONFLUX is a conditional rectified-flow model for full-volume 3D chest-CT synthesis, with GRPO reinforcement-learning post-training that sharpens control over the requested clinical findings.
Flow matching trains the generator to match the data distribution but never checks that an individual sample realizes its requested attributes; the GRPO stage optimizes that agreement directly — to our knowledge, the first GRPO post-training of a 3D medical flow model.
- Full 216×176×200 volumes from a 42-dimensional conditioning vector: 18 CT-RATE abnormality findings, sex, age, and reconstruction kernel.
- Outperforms strong 3D baselines (MAISI, GenerateCT) on distribution-level quality.
- GRPO post-training measurably improves how reliably the requested findings appear, verified by an independent held-out judge — recovering 47% of the gap to real scans.
- Release of ~200,000 labeled synthetic chest CTs for cohort augmentation and controlled study design at a scale unavailable in real corpora.
- Interactive demo for generating volumes from arbitrary clinical profiles.
📄 Paper: https://arxiv.org/abs/2607.02998
🤗 Model: https://huggingface.co/gevaertlab/conflux
🤗 Dataset (200k): https://huggingface.co/datasets/gevaertlab/conflux-chest-ct
🩻 Demo: https://huggingface.co/spaces/mxvp/conflux-chest-ct-demo
Feel free to reach out!
Will be updated in the next arXiv release, thx!
Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.
Tap or paste here to upload images
Cite arxiv.org/abs/2607.02998 in a model README.md to link it from this page.
Cite arxiv.org/abs/2607.02998 in a dataset README.md to link it from this page.
Cite arxiv.org/abs/2607.02998 in a Space README.md to link it from this page.
Discussion (0)
Sign in to join the discussion. Free account, 30 seconds — email code or GitHub.
Sign in →No comments yet. Sign in and be the first to say something.