eess.IVAug 10, 2026

BreastMammo and DenseMammo: Benchmarks for Mammography Domain Generalization

Authors: Hongyi PanGorkem DurakHalil Ertugrul AktasAndrea Mia BejarMustafa Ege SekerNebile AlibeyogluRumeysa GucluRana Gunoz Comert Bozkurt+13 more

Organizations: Northwestern University · University of Wisconsin-Madison · Istanbul University · Namik Kemal University · Istanbul Florence Nightingale Hospital · Acibadem Mehmet Ali Aydinlar University · Qatar University · Boston University · Mohamed bin Zayed University of Artificial Intelligence · NVIDIA

Abstract

Breast density classification is a critical component of breast cancer risk assessment, yet AI models often struggle to generalize across clinical sites due to vendor-specific acquisition styles. In this work, we introduce two new datasets, BreastMammo and DenseMammo, to facilitate robust multi-view mammography research. We propose a domain generalization framework that utilizes a foreground-only histogram matching protocol to resolve the domain shift issue arising from disparate clinical sources. Internal evaluation using a 5-fold cross-validation protocol demonstrates the efficacy of our approach, with the Swin Transformer backbone achieving a peak AUC of 98.32% for density classification. External evaluation on the TNMammo and LUMINA datasets demonstrates that the proposed approach consistently reduces domain shift, significantly outperforming prominent domain generalization paradigms, including MixStyle and Discrete-Fourier-Transform-based frameworks.

Explore similar work

CardsList