MammoClean: Toward Reproducible and Bias-Aware AI in Mammography through Dataset Harmonization

Yalda Zafari, Hongyi Pan, Gorkem Durak, Ulas Bagci, Essam A. Rashed, Mohamed Mabrok · 2025

Read MammoClean: Toward Reproducible and Bias-Aware AI in Mammography through Dataset Harmonization by Yalda Zafari, Hongyi Pan, Gorkem Durak, Ulas Bagci, Essam A. Rashed, Mohamed Mabrok (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-02400.txt.

Description

Yalda Zafari, Hongyi Pan, Gorkem Durak, Ulas Bagci, Essam Rashed, and Mohamed Mabrok build shared infrastructure for fairer AI in breast cancer screening. Public mammography datasets differ wildly in image quality, metadata habits, and patient populations, so models absorb dataset-specific biases and fail elsewhere, a core obstacle to clinical use. Their public framework, MammoClean, standardises case selection and image processing, including laterality and intensity correction, and unifies metadata into one consistent multi-view structure.

Alongside a review of breast anatomy, imaging traits, and the public datasets, the team applied the framework to three heterogeneous collections. They measured large shifts in breast density and abnormality rates between them. The sharpest demonstration was direct: models trained on corrupted versions of the data performed notably worse than those trained on curated versions. Clean, harmonised corpora built this way gave better generalisation across domains.

Catalog record

CREATORYalda Zafari, Hongyi Pan, Gorkem Durak, Ulas Bagci, Essam A. Rashed, Mohamed Mabrok
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE72 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2511-02400

Browse the full library · About the Tower of Babel