MammoDG: Generalisable Deep Learning Advances Cross-Domain Multi-Center Breast Cancer Screening

Abstract

Breast cancer is a major cause of cancer death among women, emphasising the importance of early detection for improved treatment outcomes and quality of life. Mammography, the primary diagnostic imaging test, poses challenges due to the high variability and patterns in mammograms. Double reading of mammograms is recommended in many screening programmes to improve diagnostic accuracy but increases the workload for radiologists. Therefore, researchers have explored Machine Learning models to support expert decision-making. Stand-alone models have shown comparable or superior performance to radiologists, but some studies observe decreased sensitivity when facing multiple datasets and sites, indicating the need for highly generalisable and robust models. This work devises MammoDG, a novel deep-learning framework for generalisable, robust, and reliable analysis of cross-domain multi-center mammography data. MammoDG introduces a cross-channel cross-view enhancement module to conduct information interaction among multi-view mammograms. Additionally, an instance-level contrastive regularisation is designed to mitigate the distribution gap between multi-center data and thus enhance generalisation capabilities. Extensive validation demonstrates the superiority of MammoDG over existing models, highlighting the critical importance of domain generalisation for trustworthy mammography analysis in the presence of imaging protocol variations.