Research map / Label-efficient and robust learning / Semi-supervised learning
Self-Paced Contrastive Learning for Semi-supervised Medical Image Segmentation with Meta-labels
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Apply self-paced meta-label contrastive learning in both pre-training and semi-supervised training, with Mean Teacher and multiple meta-labels | U-Net segmentation · ACDC, PROMISE12, and MMWHS; few labeled scans | 3D DSC | Helped |
| Add self-paced weighting to unsupervised contrastive pre-training (ablation) | Unsupervised contrastive pre-training · ACDC, 1 scan | 3D DSC: 65.14 (2.53) -> 67.35 (1.98) | Helped |
| Use self-paced slice-position meta-label contrastive pre-training instead of ordinary meta-label contrastive pre-training (ablation) | Contrastive pre-training · ACDC, 1 scan | 3D DSC: 70.57 (0.96) -> 73.99 \ (1.27) | Helped |
| Add slice-position self-paced contrastive loss during semi-supervised training without pre-training (ablation) | Supervised U-Net training · ACDC, 1 scan | 3D DSC: 57.53\ (1.18) -> 67.34\ (0.74) | Helped |
| Use self-paced contrastive learning in both pre-training and semi-supervised training rather than either stage alone (ablation) | U-Net segmentation · ACDC, 1 scan | 3D DSC: 73.99 \ (1.27) -> 75.66 (1.94) | Helped |
| Add Mean Teacher to self-paced contrastive learning used in both stages (ablation) | SP-Con (both) · ACDC, 1 scan | 3D DSC: 75.66 (1.94) -> 78.76 (0.26) | Helped |
| Combine slice position, patient identity, and cardiac phase meta-label losses instead of using slice position alone (ablation) | SP-Con (both) + Mean Teacher · ACDC, 1 scan | 3D DSC: 78.76 (0.26) -> 79.80\ (0.33) | Helped |
| Use linear rather than hard self-paced regularization (ablation) | Self-paced contrastive learning · ACDC validation | 3D DSC | Helped |
| Use slice-position meta-labels rather than patient identity or cardiac phase alone (ablation) | Meta-label contrastive learning · ACDC | 3D DSC | Helped |
| Use self-paced meta-label contrastive loss during pre-training rather than self-paced loss only during semi-supervised training (ablation) | SP-Con · ACDC, PROMISE12, and MMWHS | 3D DSC | Mixed |
| Use unsupervised contrastive pre-training rather than training from scratch (ablation) | U-Net segmentation · ACDC, PROMISE12, and MMWHS | 3D DSC | Helped |
| Increase the self-paced learning rate rapidly during the early part of training (ablation) | Self-paced contrastive learning · ACDC | 3D DSC | Helped |
Closest papers by meaning
- Systematic Comparison of Semi-supervised and Self-supervised Learning for Medical Image Classification (CVPR 2024 · similarity 0.89)
- Semi-supervised TEE Segmentation via Interacting with SAM Equipped with Noise-Resilient Prompting (AAAI 2024 · similarity 0.89)
- SGTC: Semantic-Guided Triplet Co-training for Sparsely Annotated Semi-Supervised Medical Image Segmentation (AAAI 2025 · similarity 0.88)
- Enhancing SAM with Efficient Prompting and Preference Optimization for Semi-supervised Medical Image Segmentation (CVPR 2025 · similarity 0.87)
- Neighbor Does Matter: Density-Aware Contrastive Learning for Medical Semi-supervised Segmentation (AAAI 2025 · similarity 0.87)
- Annotation Ambiguity Aware Semi-Supervised Medical Image Segmentation (CVPR 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.