Research map / Label-efficient and robust learning / Self-supervised and contrastive
A Unified Visual Information Preservation Framework for Self-supervised Pre-training in Medical Image Analysis
Hong-Yu Zhou, Chixiang Lu, Chaoqi Chen, Sibei Yang, Yizhou 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 |
|---|---|---|---|
| remove skip connections (nsUNet) (ablation) | PCRLv2 · NIH ChestX-ray (5% labels) | mean AUROC: 75.4 -> 76.6 | Improved |
| remove skip connections (nsUNet) (ablation) | PCRLv2 · BraTS (10% labels) | Dice: 71.5 -> 73.0 | Improved |
| remove skip connections (nsUNet) (ablation) | PCRLv2 · LiTS (10% labels) | Dice: 77.6 -> 79.0 | Improved |
| use cross-scale instead of pairwise siamese comparison (ablation) | PCRLv2 · NIH ChestX-ray (5% labels) | mean AUROC: 76.6 -> 76.0 | Hurt |
| add feature comparison to pixel restoration (ablation) | pixel restoration · BraTS WT/ET, NIH transfer | Dice / AUROC: WT 74.2 -> 76.2, ET 64.9 -> 64.6, NIH 78.2 -> 80.9 | Improved |
| add pixel restoration to feature comparison (ablation) | feature comparison · BraTS WT/ET, NIH transfer | Dice / AUROC: WT 76.4 -> 76.2, ET 63.8 -> 64.6, NIH 78.5 -> 80.9 | Mixed |
| add 3 feature scales (ablation) | Res+Comp · BraTS WT/ET, NIH transfer | Dice / AUROC: WT 76.2 -> 76.9, ET 64.6 -> 66.1, NIH 80.9 -> 81.5 | Improved |
| increase scales from 3 to 5 (ablation) | Res+Comp · BraTS WT/ET, NIH transfer | Dice / AUROC: WT 76.9 -> 77.2, ET 66.1 -> 66.8, NIH 81.5 -> 82.0 | Improved |
| add multi-crop in 3D (ablation) | Res+Comp, 5 scales · BraTS WT/ET, NIH transfer | Dice / AUROC: NIH 82.0 -> 82.5, BraTS fail | Mixed |
| replace multi-crop with sub-crop (ablation) | Res+Comp, 5 scales · BraTS WT/ET, NIH transfer | Dice / AUROC: WT 77.2 -> 77.7, ET 66.8 -> 67.2, NIH 82.0 -> 82.7 | Improved |
| PCRLv2 vs PCRLv1 (multi-scale, nsUNet, simpler design) | PCRLv1 · NIH ChestX-ray semi-supervised (5% labels) | mean AUROC: 74.1 -> 76.6 | Improved |
| PCRLv2 vs PCRLv1 | PCRLv1 · LiTS transfer (10% labels) | Dice: 77.3 -> 79.0 | Improved |
Closest papers by meaning
- Preservational Learning Improves Self-Supervised Medical Image Models by Reconstructing Diverse Contexts (ICCV 2021 · similarity 0.88)
- FOCUS: Knowledge-enhanced Adaptive Visual Compression for Few-shot Whole Slide Image Classification (CVPR 2025 · similarity 0.86)
- PathVQ: Reforming Computational Pathology Foundation Model for Whole Slide Image Analysis via Vector Quantization (NeurIPS 2025 · similarity 0.86)
- Revisiting MAE Pre-training for 3D Medical Image Segmentation (CVPR 2025 · similarity 0.86)
- Unleashing the Potential of SAM for Medical Adaptation via Hierarchical Decoding (CVPR 2024 · similarity 0.85)
- Unsupervised Foundation Model-Agnostic Slide-Level Representation Learning (CVPR 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.
