TGIAlign: Text-guided dual-branch bidirectional framework for cross-modal semantic alignment in medical vision-language.

Medical image-text alignment remains challenging due to subtle lesion patterns, heterogeneous vision-language semantics, and the lack of lesion-aware guidance during visual encoding. Existing methods typically introduce textual information only after…

Preview only. Read the full abstract at the source