Research map / 3D vision / 3D generation
Generative Enhancement for 3D Medical Images
Lingting Zhu, Noel Codella, Dongdong Chen, Zhenchao Jin, Lu Yuan, Lequan Yu
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add informed-slice conditioning (GEM-3D + IC vs Make-A-Volume) | Make-A-Volume · BraTS (train samples) | FID-C: 62.63 -> 6.14 | Improved |
| add informed-slice conditioning (GEM-3D + IC vs Make-A-Volume) | Make-A-Volume · AbdomenCT-1K (train samples) | FID-C: 6.68 -> 4.21 | Improved |
| add true informed slice as patient prior at test time | Make-A-Volume · BraTS (test) | MS-SSIM: 0.817 -> 0.887 | Improved |
| add true informed slice as patient prior at test time | Make-A-Volume · AbdomenCT-1K (test) | LPIPS: 0.318 -> 0.204 | Improved |
| generate the initial informed slice (IG) instead of cross-sampling (IC) (ablation) | GEM-3D · BraTS (train samples) | FID-A: 2.07 -> 15.67 | Hurt |
| add 3D mask augmentation (MA) (ablation) | GEM-3D + IC · BraTS (train samples) | FID-C: 6.14 -> 11.76 | Hurt |
| add 3D mask augmentation (MA) (ablation) | Make-A-Volume · BraTS (train samples) | FID-C: 62.63 -> 71.64 | Hurt |
| replace concatenation with ControlNet for condition injection (ablation) | GEM-3D · BraTS (test) | MS-SSIM: 0.887 -> 0.844 | Hurt |
| replace concatenation with Concatenation + CLIP (ablation) | GEM-3D · BraTS (test) | MS-SSIM: 0.887 -> 0.879 | Hurt |
| train segmentation on re-sampled GEM-3D data (3x) instead of baseline reconstruction | nnU-Net · BraTS segmentation | Dice (%): 59.70 -> 71.57 | Improved |
| increase re-sampling from 1x to 3x (ablation) | GEM-3D synthetic training data · BraTS segmentation | Dice (%): 68.53 -> 71.57 | Improved |
| synthetic 3x data versus true data | nnU-Net · BraTS segmentation | Dice (%): 72.90 -> 71.57 | Hurt |
Closest papers by meaning
- TCAM-Diff: Triplane-Aware Cross-Attention Medical Diffusion Model (AAAI 2025 · similarity 0.86)
- Better Tokens for Better 3D: Advancing Vision-Language Modeling in 3D Medical Imaging (NeurIPS 2025 · similarity 0.86)
- Enhancing vision-language models for medical imaging: bridging the 3D gap with innovative slice selection (NeurIPS 2024 · similarity 0.85)
- LiFT: Lifted Inter-slice Feature Trajectories for 3D Image Generation from 2D Generators (NeurIPS 2026 · similarity 0.84)
- DiffusionBlend: Learning 3D Image Prior through Position-aware Diffusion Score Blending for 3D Computed Tomography Reconstruction (NeurIPS 2024 · similarity 0.84)
- LeFusion: Controllable Pathology Synthesis via Lesion-Focused Diffusion Models (ICLR 2025 · similarity 0.84)
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.
