Tag
This paper proposes a token-centric dual-view learning framework that unifies prompt-based adaptation and cross-view fusion within a frozen vision transformer to improve breast cancer classification from mammography images, achieving consistent improvements on VinDr-Mammo and CMMD datasets.