Reproduce “DBF-Net: A Dual-Branch Network with Feature Fusion for Ultrasound 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

Guoping Xu, Ximing Wu, Wentao Liao, Xinglong Wu, Qing Huang, Chang Li

DBF-Net, a dual-branch network, improves ultrasound image segmentation by explicitly modeling and fusing lesion body and boundary information. The method achieved superior Dice Similarity Coefficients of 81.05% on BUSI, 76.41% on UNS, and 87.75% on UHES datasets, along with lower Hausdorff Distances, outperforming various CNN and Transformer-based approaches.

Key insights

  • Explicitly modeling and fusing the relationship between lesion body and boundary features significantly improves segmentation accuracy and boundary delineation in US images.
  • Auxiliary supervision for both body and boundary maps at multiple decoder stages effectively enhances the network's overall performance.
  • The Feature Fusion Module (FFM) helps balance the contributions of body and boundary components during training, preventing one from dominating the other and leading to more coherent feature integration.

Summary by alphaXiv.

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