Research map / Vision-language and multimodal / Cross-modal learning
Posterior Samplings are Missing Modalities Generators for Medical Image Translation
Jonghun Kim
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 |
|---|---|---|---|
| use many-to-one instead of many-to-many sampling (all-scenarios training) (ablation) | Ours · BraTS T1->T2 | FID: 65.75 -> 54.15 | Improved |
| use many-to-one sampling on full-modal-only trained model (ablation) | Ours · BraTS T1->T2 | PSNR: 22.28 -> 22.19 | No clear effect |
| train on all missing-modality scenarios instead of full-modal only (MTO) (ablation) | Ours · BraTS T1->T2 | PSNR: 22.19 -> 23.56 | Improved |
| train on all scenarios with MTM sampling instead of full-modal only (ablation) | Ours · BraTS T1->T2 | PSNR: 22.28 -> 22.26 | No clear effect |
| use PSLD instead of DPS posterior sampling (ablation) | Ours · BraTS T2->T1 | PSNR: 24.21 -> 25.56 | Improved |
| use PSLD instead of FlowDPS (ablation) | Ours · BraTS T1->T2 | PSNR: 22.11 -> 23.56 | Improved |
| use PSLD instead of DDNM (ablation) | Ours · BraTS T1->T2 | SSIM: 86.09 -> 88.47 | Improved |
| increase number of source modalities from one to three (ablation) | Ours · BraTS T1 target | PSNR: 25.56 -> 27.84 | Improved |
| replace baselines with posterior-sampled joint prior | Ours vs MMHVAE · BraTS T1->T2 | PSNR: 20.53 -> 23.56 | Improved |
| single-source SSIM vs baselines | Ours · BraTS single-source | SSIM: T1ce->T1: 88.17 vs MMHVAE 89.15 | Mixed |
| reduce expectation approximation samples N_EX to 10 -> 1 (ablation) | Ours · BraTS tumor segmentation | Dice: average Dice 76.94% at N_EX=1 | Hurt |
| increase N_EX (ablation) | Ours · BraTS T1->T2 | perceptual quality: fidelity up, perceptual quality down | Hurt |
Closest papers by meaning
- Tackling Dual-stage Missing Modalities in Brain Tumor Segmentation via Robust Modality Reconstruction and Prompt-guided Modality Adaptation (AAAI 2026 · similarity 0.90)
- MAISI-v2: Accelerated 3D High-Resolution Medical Image Synthesis with Rectified Flow and Region-specific Contrastive Loss (AAAI 2026 · similarity 0.88)
- SphereFlow: Missing Modality Imputation via Geometric Transport on Hypersphere (NeurIPS 2026 · similarity 0.87)
- Simultaneous Super-Resolution and Cross-Modality Synthesis of 3D Medical Images Using Weakly-Supervised Joint Convolutional Sparse Coding (CVPR 2017 · similarity 0.87)
- MRGen: Segmentation Data Engine For Underrepresented MRI Modalities (ICCV 2025 · similarity 0.87)
- UniMRSeg: Unified Modality-Relax Segmentation via Hierarchical Self-Supervised Compensation (NeurIPS 2025 · similarity 0.87)
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.
