AI Paper Atlas

Research map / Image generation / Autoregressive and latent generation

Foundation VAE for CT Reconstruction, Augmentation, and Generation

Qi Chen, Shuhan Ding, Yu Gu, Nan Liu, Jiang Bian, Alan Yuille, Zongwei Zhou, Jingjing Fu

ICML 2026 · Image generation › Autoregressive and latent generation · approach: Hierarchical VQ-VAE generation

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Train segmenter on frozen Foundation VAE reconstructions instead of real CTnnU-Net (WAN2.1) · MSD Task06 LungNSD: 70.2 -> 75.3Improved
Use CVVAE reconstructions for segmentation training instead of real CTnnU-Net · MSD Task07 Pancreas (tumor)NSD2: 39.8 -> 34.5Hurt
Replace Foundation VAE with MedVAE in latent diffusion (ablation)conditional latent diffusion · CT-RATE normal splitFID: 11.28 -> 2.19Improved
Use Foundation VAE instead of MedVAE (ablation)conditional latent diffusion · CT-RATE normal splitCT-CLIP: 20.76 -> 59.35Improved
Add a single disease mask and single-finding prompt to organ-mask conditioning (ablation)conditional latent diffusion · CT-RATE disease splitCT-CLIP Avg: 43.40 -> 46.14Improved
Use multi-disease masks with full report prompts (ablation)conditional latent diffusion · CT-RATE disease splitCT-CLIP Avg: CT-CLIP 46.14 -> 51.49; FID 3.76 -> 4.35Mixed
Add disease-mask-guided synthetic CT to classifier trainingmulti-label classifier · CT-RATE 18-label classificationmean AUC: 67.95 -> 70.71Improved
Condition generation on organ masks (MAISI-style masks) with text and disease maskOurs vs MAISI · Organ-mask adherencevessel Dice: 13.50 -> 63.38Improved
Use CT-specific MedVAE for reconstruction on MSD dataMedVAE · MSD Task07 PancreasPSNR: PSNR 36.00 vs IVVAE 40.43Hurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy