Research map / Label-efficient and robust learning / Semi-supervised learning
The Devil is in the Statistics: Mitigating and Exploiting Statistics Difference for Generalizable Semi-supervised Medical Image Segmentation
Muyang Qiu, Jian Zhang, Lei Qi, Qian Yu, Yinghuan Shi, Yang Gao
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 statistics-individual and aggregated branches (SIAB) (ablation) | semi-supervised DG baseline · Fundus | average Dice (%): 83.28 -> 84.56 | Improved |
| add histogram-matching style augmentation (ablation) | baseline with SIAB · Fundus | average Dice (%): 84.56 -> 85.22 | Improved |
| add RandBN feature-level perturbation (ablation) | SIAB + Style · Fundus | average Dice (%): 85.22 -> 85.60 | Improved |
| use both batch and instance statistics in aggregated branch instead of one (ablation) | SAB · Fundus | average Dice (%): 83.50 -> 84.56 | Improved |
| use instance statistics only instead of batch statistics (ablation) | SAB · Fundus | average Dice (%): 83.50 -> 83.31 | Hurt |
| use strong plus style augmentation instead of two strong augmentations (ablation) | consistency regularization · Fundus | average Dice (%): 84.76 -> 85.60 | Improved |
| increase RandBN feature-consistency weight above 0.2 to 1.0 (ablation) | Ours · Fundus | average Dice (%): 85.60 -> 85.50 | No clear effect |
| apply RandBN perturbation (p>0) versus none (ablation) | Ours · Fundus | average Dice (%): 85.18 -> 85.63 | Improved |
| reduce training to 2.01 hours (ablation) | Ours · Prostate (30%) | Dice (%): 88.73 -> 87.98 (CACPS 86.47) | No clear effect |
| vary confidence threshold from 0.85 to 0.99 (ablation) | Ours · Fundus | average Dice (%): 85.30 - 85.60 | No clear effect |
| apply method to typical semi-supervised task with 5% labels | Ours · ACDC 3 cases | Dice (%): BCP 87.59 vs Ours 86.72 | No clear effect |
Closest papers by meaning
- Constructing and Exploring Intermediate Domains in Mixed Domain Semi-supervised Medical Image Segmentation (CVPR 2024 · similarity 0.91)
- Steady Progress Beats Stagnation: Mutual Aid of Foundation and Conventional Models in Mixed Domain Semi-Supervised Medical Image Segmentation (CVPR 2025 · similarity 0.90)
- Why Invariance is Not Enough for Biomedical Domain Generalization and How to Fix It (NeurIPS 2026 · similarity 0.89)
- MAPSeg: Unified Unsupervised Domain Adaptation for Heterogeneous Medical Image Segmentation Based on 3D Masked Autoencoding and Pseudo-Labeling (CVPR 2024 · similarity 0.88)
- Domain Generalized Medical Landmark Detection via Robust Boundary-Aware Pre-Training (AAAI 2025 · similarity 0.88)
- A Unified Degradation-Robust Approach to SSL and UDA for 3D Medical Images (AAAI 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.
