Reproduce “Semi-supervised Medical Image Segmentation via Query Distribution Consistency” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Rong Wu, Dehua Li, Cong Zhang
A semi-supervised framework, Dual-KMax UX-Net (DKUXNet), improves 3D medical image segmentation by introducing a dual-level consistency loss across segmentation outputs and learned query distributions. This approach enables the model to match or exceed the performance of fully supervised methods using only 10-20% of the training data annotations.
Key insights
- Explicitly leveraging labeled data to guide the information extraction process from unlabeled samples, rather than merely for initial training, enhances semi-supervised learning in medical imaging.
- Enforcing consistency on both fine-grained pixel-level segmentation and high-level query distributions (cluster centers) synergistically improves model robustness and representation learning.
- Integrating modern architectural components like ConvNeXt-based encoders and KMax decoders (using k-means cross-attention) is effective for advanced 3D medical image segmentation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.