Research map / Label-efficient and robust learning / Semi-supervised learning
Semi-supervised medical segmentation
Combines foundation models, density-aware contrast and variance analysis for medical image segmentation.
50 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2019–2026) · an approach in Semi-supervised learning, keyword cluster “pseudo labels · segmentation · pseudo labeling”
State History Benchmarks Tried Checklist Preprints Who Export
Where it stands
Still moving The best reported result is still improving on 4 of 4 tracked benchmarks.
15 papers in 2023–24 and 24 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 9.4% to 15.2%.
Open questions the data points to
- ACDC 10% labeled (Dice): best went from 88.84 (up to 2024) to 90.91 in 2025 — Neighbor Does Matter: Density-Aware Contrastive Learning for Medical Semi-supervised Segmentation (AAAI 2025 · 8 citations)
- ACDC 5% labeled (Dice): best went from 87.59 (up to 2024) to 90.12 in 2025 — Neighbor Does Matter: Density-Aware Contrastive Learning for Medical Semi-supervised Segmentation (AAAI 2025 · 8 citations)
- LA, 10% labeled (Dice (%)): best went from 89.62 (up to 2024) to 92.77 in 2025 — DyCON: Dynamic Uncertainty-aware Consistency and Contrastive Learning for Semi-supervised Medical Image Segmentation (CVPR 2025 · 36 citations)
- LA, 5% labeled (Dice (%)): best went from 88.02 (up to 2024) to 91.18 in 2025 — DyCON: Dynamic Uncertainty-aware Consistency and Contrastive Learning for Semi-supervised Medical Image Segmentation (CVPR 2025 · 36 citations)
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2019 · Data Augmentation Using Learned Transformations for One-Shot Medical Image Segmentation (CVPR 2019 · 459 citations)
- 2020 · FocalMix: Semi-Supervised Learning for 3D Medical Image Detection (CVPR 2020 · 153 citations)
- 2021 · Semi-supervised Medical Image Segmentation through Dual-task Consistency (AAAI 2021 · 689 citations)
- 2022 · BoostMIS: Boosting Medical Image Semi-Supervised Learning With Adaptive Pseudo Labeling and Informative Active Annotation (CVPR 2022 · 130 citations)
- 2023 · Bidirectional Copy-Paste for Semi-Supervised Medical Image Segmentation (CVPR 2023 · 412 citations)
- 2024 · Adaptive Bidirectional Displacement for Semi-Supervised Medical Image Segmentation (CVPR 2024 · 110 citations)
- 2025 · DyCON: Dynamic Uncertainty-aware Consistency and Contrastive Learning for Semi-supervised Medical Image Segmentation (CVPR 2025 · 36 citations)
- 2026 · Bidirectional Channel-selective Semantic Interaction for Semi-Supervised Medical Segmentation (AAAI 2026 · 1 citation)
Earliest papers here
- Data Augmentation Using Learned Transformations for One-Shot Medical Image Segmentation (CVPR 2019 · 459 citations)
- A Two-Stream Mutual Attention Network for Semi-Supervised Biomedical Segmentation with Noisy Labels (AAAI 2019 · 78 citations)
- FocalMix: Semi-Supervised Learning for 3D Medical Image Detection (CVPR 2020 · 153 citations)
Benchmark records
The best number each year that beat every earlier one, among papers of this approach reporting their own method. Numbers are extracted by AI from each paper's text; check the paper before citing.
ACDC 10% labeled · Dice (higher is better)
4 papers, 2023–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2023 | 88.84 | Bidirectional Copy-Paste for Semi-Supervised Medical Image Segmentation (CVPR 2023 · 412 citations) |
| 2025 | 90.91 | Neighbor Does Matter: Density-Aware Contrastive Learning for Medical Semi-supervised Segmentation (AAAI 2025 · 8 citations) |
ACDC 5% labeled · Dice (higher is better)
4 papers, 2023–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2023 | 87.59 | Bidirectional Copy-Paste for Semi-Supervised Medical Image Segmentation (CVPR 2023 · 412 citations) |
| 2025 | 90.12 | Neighbor Does Matter: Density-Aware Contrastive Learning for Medical Semi-supervised Segmentation (AAAI 2025 · 8 citations) |
LA, 10% labeled · Dice (%) (higher is better)
4 papers, 2023–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2023 | 89.62 | Bidirectional Copy-Paste for Semi-Supervised Medical Image Segmentation (CVPR 2023 · 412 citations) |
| 2025 | 92.77 | DyCON: Dynamic Uncertainty-aware Consistency and Contrastive Learning for Semi-supervised Medical Image Segmentation (CVPR 2025 · 36 citations) |
LA, 5% labeled · Dice (%) (higher is better)
4 papers, 2023–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2023 | 88.02 | Bidirectional Copy-Paste for Semi-Supervised Medical Image Segmentation (CVPR 2023 · 412 citations) |
| 2025 | 91.18 | DyCON: Dynamic Uncertainty-aware Consistency and Contrastive Learning for Semi-supervised Medical Image Segmentation (CVPR 2025 · 36 citations) |
What has been tried
38 papers here report 434 measured changes (a component added, removed or swapped, compared with a baseline): helped 79%, no clear effect 3%, hurt 14%, mixed 4%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Add input perturbation | ACDC test, 10% labeled · DSC | 86.45 -> 86.25 | Hurt | Adaptive Bidirectional Displacement for Semi-Supervised Medical Image Segmentation (CVPR 2024 · 110 citations) |
| Add VanillaGD to UCP | Fundus · DC | 85.78 -> 86.07 | Helped | Constructing and Exploring Intermediate Domains in Mixed Domain Semi-supervised Medical Image Segmentation (CVPR 2024 · 45 citations) |
| add UnCL without β to MT | ISLES’22, 10% labeled · Dice (%) | 36.43 -> 58.78 | Helped | DyCON: Dynamic Uncertainty-aware Consistency and Contrastive Learning for Semi-supervised Medical Image Segmentation (CVPR 2025 · 36 citations) |
| add adaptive patch augmentation to supervised-only training | UDIAT 1/4 · Dice (%) | +1.89% | Helped | PH-Net: Semi-Supervised Breast Lesion Segmentation via Patch-wise Hardness (CVPR 2024 · 31 citations) |
| Add low-frequency component exchange with H=30x30 | ACDC test, 10% labeled · Dice | 89.60 -> 89.96 | Helped | beta-FFT: Nonlinear Interpolation and Differentiated Training Strategies for Semi-Supervised Medical Image Segmentation (CVPR 2025 · 16 citations) |
| set αu to 0.3 instead of 0.5 | LIDC-IDRI, 5% labeled · GED | 0.1872 -> 0.1987 | Hurt | Annotation Ambiguity Aware Semi-Supervised Medical Image Segmentation (CVPR 2025 · 13 citations) |
| add preference alignment trained with unannotated data | Chest-Xray · Dice | 75.60 -> 78.87 | Helped | Enhancing SAM with Efficient Prompting and Preference Optimization for Semi-supervised Medical Image Segmentation (CVPR 2025 · 13 citations) |
| Add pixel-level consistency loss Lpixel | ACDC and ISIC, 3% labeled · DSC | ACDC: 0.510 -> 0.715; ISIC: 0.677 -> 0.734 | Helped | Combinatorial CNN-Transformer Learning with Manifold Constraints for Semi-supervised Medical Image Segmentation (AAAI 2024 · 12 citations) |
| Add GAP to EM | ACDC, 1%, 5%, and 10% labeled · Dice (%) | ↑13.9; ↑18.9; ↑3.2 | Helped | GapMatch: Bridging Instance and Model Perturbations for Enhanced Semi-Supervised Medical Image Segmentation (AAAI 2025 · 11 citations) |
| Add data flow decoupling to the basic pseudo-labeling baseline | Synapse (20% labeled) · average Dice (%) | 44.66 ± 2.5 -> 57.4 ± 0.9 | Helped | A Semantic Knowledge Complementarity based Decoupling Framework for Semi-supervised Class-imbalanced Medical Image Segmentation (CVPR 2025 · 11 citations) |
| Synergistically train U-Net and MedSAM with SMC and CDCR | Prostate · DSC | 81.14 -> 87.16 | Helped | Steady Progress Beats Stagnation: Mutual Aid of Foundation and Conventional Models in Mixed Domain Semi-Supervised Medical Image Segmentation (CVPR 2025 · 10 citations) |
| Add plain contrastive learning with random sampling | ACDC 10% labeled · Dice (%) | 85.71±0.9 -> 87.50±0.8 | Helped | Neighbor Does Matter: Density-Aware Contrastive Learning for Medical Semi-supervised Segmentation (AAAI 2025 · 8 citations) |
What reviewers will expect
Datasets and baselines named in the 14 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- ACDC · 6 of 14 papers
- LA · 2 of 14 papers
- Synapse · 2 of 14 papers
- Pancreas-NIH · 2 of 14 papers
- BUSI · 2 of 14 papers
Baselines
- UA-MT · 8 of 14 papers
- BCP · 8 of 14 papers
- SS-Net · 7 of 14 papers
- URPC · 7 of 14 papers
- DTC · 5 of 14 papers
- MC-Net · 5 of 14 papers
- CPS · 4 of 14 papers
- MT · 3 of 14 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 39 papers; from OpenAlex and the papers' affiliations).
- Hong Kong University of Science and Technology · 4 papers
- Shanghai Jiao Tong University · 3 papers
- The Hong Kong Polytechnic University · 3 papers
- University of Science and Technology of China · 3 papers
- Nanjing University · 3 papers
- Shandong Women’s University · 3 papers
- Tencent · 2 papers
- East China Normal University · 2 papers
- The University of Hong Kong · 2 papers
- State Key Laboratory for Novel Software Technology · 2 papers
Use it
Download BibTeX of 36 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- Confidence-based pseudo-label selection (134 papers)
- Consistency training for semi-supervised segmentation (49 papers)
- Pseudo-labeling for specialized tasks (48 papers)
Elsewhere in the topic
- Holistic semi-supervised pipelines (125 papers)
- Self-training theory (86 papers)
- Safe and realistic semi-supervised learning (70 papers)
- GAN-based semi-supervised learning (61 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).