Reproduce “DWARF: Disease-weighted network for attention map refinement” 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
Haozhe Luo, Aurélie Pahud de Mortanges, Oana Inel, Abraham Bernstein, Mauricio Reyes
DWARF trains disease-specific chest X-ray classifiers to align highlighted regions with annotated disease regions. It uses a pretrained vision-language model and a disease-specific head to convert cross-attention into segmentation-style maps, training classification and map alignment together. On full 13-finding ChestX-Det, it reports AUC 81.94 ± 0.37% and Max Dice 18.24 ± 0.18%, versus GAIN’s 80.90% and 13.90%; these are benchmark, not clinical, results.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.