Reproduce “Data-Centric Diet: Effective Multi-center Dataset Pruning for Medical Image Segmentation” 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
Yongkang He, Mingjin Chen, Zhijing Yang, Yongyi Lu
This research introduces the Dynamic Average Dice (DAD) score, a data pruning metric tailored for medical image segmentation that effectively identifies and removes less informative examples from multi-center datasets. The method often improves segmentation accuracy while significantly reducing the required training data volume.
Key insights
- Combining multiple medical image datasets does not guarantee improved segmentation performance, as increased data variance and noise can be detrimental.
- For dense labeling tasks like medical image segmentation, metrics that focus on foreground regions are essential for accurately assessing learning difficulty, unlike whole-image metrics which are dominated by background.
- Retaining 'ambiguous' examples (mid-range DAD scores) generally yields better segmentation performance than focusing solely on 'hard-to-learn' examples when pruning medical image datasets.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.