Research map / Vision-language and multimodal / Cross-modal learning
Towards a Universal 3D Medical Multi-modality Generalization via Learning Personalized Invariant Representation
Zhaorui Tan, Xi Yang, Tan Pan, Tianyi Liu, Chen Jiang, Xin Guo, Qiufeng Wang, Anh Nguyen, Yuan Qi, Kaizhu Huang, Yuan Cheng
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 the learnable prior O to equivariance + invariance constraints (ablation) | PUIR · AutoPET-II segmentation | Dice: 42.42 -> 48.20 | Improved |
| Add prior O to contrastive + decomposition + equivariance (ablation) | PUIR · AutoPET-II segmentation | Dice: 40.85 -> 46.31 | Improved |
| Add prior O to contrastive + decomposition + invariance (ablation) | PUIR · AutoPET-II segmentation | Dice: 44.34 -> 44.42 | No clear effect |
| Add invariance to equivariance without prior O (ablation) | PUIR · AutoPET-II segmentation | Dice: 44.34 -> 42.42 | Hurt |
| Add invariance after O to equivariance + O (ablation) | PUIR · AutoPET-II segmentation | Dice: 46.31 -> 48.20 | Improved |
| Use PUIR pre-training in SwinUNETR instead of prior SSL pre-training | SwinUNETR · AutoPET-II segmentation | Dice: 44.06 -> 48.20 | Improved |
| Use PUIR pre-training instead of training from scratch | SwinUNETR · AutoPET-II segmentation | Dice: 43.45 -> 48.20 | Improved |
| Use PUIR instead of MS-SPADE for one-to-all modality transfer | MS-SPADE · BRATS23 | PSNR: 26.4225 -> 35.4378 | Improved |
| Use 96^3 instead of 192^3 volumes (ablation) | PUIR · BRATS23 | PSNR: 35.4378 -> 33.7418 | No clear effect |
| Use PUIR pre-training instead of SwinUNETR pre-training for 3 missing modalities | SwinUNETR pre-training · BRATS18 enhancing tumor | Dice: 55.79 -> 60.19 | Improved |
| Use PUIR instead of mmFormer on full modality enhancing tumor | mmFormer · BRATS18 enhancing tumor | Dice: 79.91 -> 70.64 | Hurt |
| Use PUIR instead of SwinUNETR for NAC-PET to CT translation | SwinUNETR · HNSCC | SSIM: 0.5853 -> 0.6939 | Improved |
Closest papers by meaning
- Beyond Instance-Level Self-Supervision in 3D Multi-Modal Medical Imaging (ICML 2026 · similarity 0.88)
- Multi-modal Vision Pre-training for Medical Image Analysis (CVPR 2025 · similarity 0.86)
- Learning General-purpose Biomedical Volume Representations using Randomized Synthesis (ICLR 2025 · similarity 0.86)
- Modality-Agnostic Structural Image Representation Learning for Deformable Multi-Modality Medical Image Registration (CVPR 2024 · similarity 0.85)
- Why Invariance is Not Enough for Biomedical Domain Generalization and How to Fix It (NeurIPS 2026 · similarity 0.85)
- Modalities Contribute Unequally: Enhancing Medical Multi-modal Learning through Adaptive Modality Token Re-balancing (ICML 2025 · similarity 0.85)
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.
