Reproduce “Do Saliency Maps Fail Dense Breasts? A Chance-Normalized, Confounder-Aware Audit of Explanation Fairness in Mammography” 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
Ahmed Jaheen
The authors audit mammography saliency maps by comparing lesion localization across density groups, correcting for chance and case mix. In 1,345 malignant views from CBIS-DDSM, Grad-CAM peaks hit within 5 mm in 44% of dense versus 58% of non-dense cases before adjustment. After adjustment, occlusion retains a significant gap, but an exploratory model finds none; findings are dataset-specific.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.