Reproduce “VasoMIM: Vascular Anatomy-Aware Masked Image Modeling for Vessel 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

De-Xing Huang, Xiao-Hu Zhou, Mei-Jiang Gui, Xiao-Liang Xie, Shi-Qi Liu, Shuang-Yi Wang, Tian-Yu Xiang, Rui-Ze Ma and 2 more

VasoMIM introduces a vascular anatomy-aware Masked Image Modeling (MIM) framework that integrates explicit anatomical knowledge into the pre-training process for X-ray angiograms. It achieves superior vessel segmentation performance and improved generalizability on both in-domain and out-of-domain datasets, outperforming existing self-supervised learning methods.

Key insights

  • Explicitly integrating vascular anatomical knowledge into MIM pre-training significantly improves representation learning for vessel segmentation, overcoming challenges posed by class imbalance.
  • A gradual 'weak-to-strong' anatomical guidance strategy in masking effectively balances initial model exploration with focused attention on anatomically relevant regions.
  • Employing a semantic-level consistency loss, beyond pixel-level reconstruction, is crucial for learning robust and discriminative features that maintain the topological integrity of vessels.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.