Bridging radiology and pathology: domain-generalized cross-modal learning for clinical applications

1

The proposed framework integrates mammography and histopathology for improved breast cancer diagnosis using a cross-modal learning approach.
2

It employs a shared vision transformer encoder and a weakly supervised contrastive alignment module for cross-modal correspondences.
3

Domain generalization strategies enhance the model's robustness across different institutions, achieving a mean AUC of 0.90.
4

The framework generates reasoning-guided attention maps that connect mammographic findings with histopathological evidence.
5

This study advances multimodal integration and explainability, aiming for clinically deployable AI systems in diagnostic decision support.

Npj Digital Medicine

by Xiang Zhong, Zhuo Gu, Manimurugan Shanmuganathan, Meng Li, Hao Sun, Mingming Du, Qian Chen, Guoqin Jiang
February 16, 2026

1