Unsupervised Domain Adaptation for Calcification Classification in Mammography (arXiv 2507.06826)

Forum topic · 小凯 · 2026-07-09

Summary

A 2025 paper on arXiv (2507.06826) by Xuan Liu, Derek L. Nguyen, and Emily C. Barre proposes an unsupervised domain adaptation framework for classifying malignant versus benign breast calcifications in mammography across multi-site datasets. The framework has two components: (1) an unsupervised domain adaptation module built on style transfer models (AdaIN and CycleGAN) that generates vendor-specific and technique-specific training samples without extra annotations, and (2) a supervised classifier using Swin Transformer V2 as the backbone. The method was cross-validated on OPTIMAM (UK NHS, n=2994) and externally validated on EMBED (Emory University, n=125) and the Duke Calcification Dataset v1 (n=788), covering multiple vendors and both full-field digital mammography and synthetic 2D images from digital breast tomosynthesis. Domain adaptation improved cross-site performance, raising AUC from 0.68 to 0.72 on EMBED and from 0.68 to 0.73 on the Duke dataset, showing it reduces domain shift and improves generalization for calcification classification.

Overview

Field: Computer Vision (medical imaging)
Authors: Xuan Liu, Derek L. Nguyen, Emily C. Barre
Published: 2025-07-09
arXiv: 2507.06826

Key points

Deep learning-based computer-aided diagnosis (CAD) systems perform strongly in breast cancer diagnosis, especially mammography classification, but domain shifts across multi-site datasets remain a major challenge when models are applied to unseen domains.
The authors propose a calcification classification framework for malignant vs. benign breast disease classification with two components:

unsupervised domain adaptation module

AdaIN and CycleGAN

without additional annotations

supervised classification module

Swin Transformer V2

Evaluation

Cross-validation: OPTIMAM (UK National Health Service; n=2994)
External validation: EMBED (Emory University; n=125) and Duke Calcification Dataset v1 (n=788)
Datasets span multiple vendors, including full-field digital mammography and synthetic 2D images derived from digital breast tomosynthesis.

Results

EMBED: AUC improved from 0.68 to 0.72
Duke Calcification Dataset: AUC improved from 0.68 to 0.73

Conclusion

Unsupervised domain adaptation via style transfer reduces domain shift and improves generalization of calcification classification across multi-site mammography datasets.

Original abstract

Deep learning-based computer-aided diagnosis (CAD) systems have shown strong performance in breast cancer diagnosis, particularly for classification tasks in mammography. However, domain shifts across multi-site datasets remain a challenge, especially when models are applied to unseen domains. In this work, we proposed a calcification classification framework to improve malignant versus benign breast disease classification across multi-site mammography datasets. The framework consisted of two components: (1) an unsupervised domain adaptation module based on style transfer models (AdaIN and CycleGAN) to generate vendor-specific and technique-specific training samples without additional annotations, and (2) a supervised classification module using Swin Transformer V2 as the backbone. We eval...

--- *Auto-collected on 2026-07-09*