Research map / Detection and segmentation / CNN features
Keypoint-Augmented Self-Supervised Learning for Medical Image Segmentation with Limited Annotation
Zhangsihao Yang, Mengwei Ren, Kaize Ding, Guido Gerig, Yalin Wang
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 KAF layers at all four encoder scales (ablation) | UNet (random init) · CHD M=15 | Dice: 0.627 -> 0.686 | Improved |
| add KAF layers at all four encoder scales (ablation) | UNet (random init) · ACDC M=6 | Dice: 0.782 -> 0.827 | Improved |
| remove KAF at last scale l4 (Exp B vs F) (ablation) | KAF-UNet · CHD M=15 | Dice: 0.686 -> 0.658 | Hurt |
| reduce self-attention layers in KAF transformer from 9 to 3 (ablation) | KAF-UNet · CHD M=15 | Dice: 0.686 -> 0.677 | Hurt |
| reduce self-attention layers from 9 to 6 (ablation) | KAF-UNet · CHD M=15 | Dice: 0.686 -> 0.690 | No clear effect |
| combine global conv and keypoint SSL losses vs single global loss (ablation) | KAF-UNet SSL · CHD M=15 | Dice: 0.699 -> 0.711 | Improved |
| add local correspondence loss w3=0.01 (ablation) | KAF-UNet SSL · CHD M=15 | Dice: 0.711 -> 0.712 | Improved |
| add local loss w3=0.01 on ACDC (ablation) | KAF-UNet SSL · ACDC M=6 | Dice: 0.865 -> 0.873 | Improved |
| replace SIFT with SuperPoint keypoints (random init) (ablation) | KAF-UNet · CHD M=2 | Dice: 0.686 -> 0.643 | Hurt |
| replace SIFT with SuperPoint keypoints (SSL) (ablation) | KAF-UNet SSL · CHD M=2 | Dice: 0.712 -> 0.703 | No clear effect |
| vary keypoint correspondence threshold 5 to 40 (ablation) | KAF-UNet SSL local loss · CHD M=15 | Dice: 0.684 -> 0.690 | No clear effect |
| widen UNet with duplicated channels (UNet(c2)) vs standard UNet (ablation) | UNet · CHD M=15 | Dice: 0.627 -> 0.646 | Improved |
Closest papers by meaning
- Revisiting MAE Pre-training for 3D Medical Image Segmentation (CVPR 2025 · similarity 0.88)
- EMCAD: Efficient Multi-scale Convolutional Attention Decoding for Medical Image Segmentation (CVPR 2024 · similarity 0.86)
- Repurposing Stable Diffusion Attention for Training-Free Unsupervised Interactive Segmentation (CVPR 2025 · similarity 0.86)
- Enhancing SAM with Efficient Prompting and Preference Optimization for Semi-supervised Medical Image Segmentation (CVPR 2025 · similarity 0.86)
- MAPSeg: Unified Unsupervised Domain Adaptation for Heterogeneous Medical Image Segmentation Based on 3D Masked Autoencoding and Pseudo-Labeling (CVPR 2024 · similarity 0.85)
- Unleashing the Potential of SAM for Medical Adaptation via Hierarchical Decoding (CVPR 2024 · 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.
