LA-CLIP: Lesion-Aware Vision-Language Pretraining for Mammography via ROI-Guided Contrastive Learning
European Congress of Radiology (ECR) / 2026 / conference
Bhandary Panambur A, Nguyen T-T, Bayer S, Maier A
conference / 2026
Medical AI visual summary
Scientific summary
LA-CLIP studies lesion-aware mammography vision-language pretraining through ROI-guided contrastive representation learning.
Abstract
TODO: Add the ECR abstract when the public poster or proceedings page is available.
Why it matters
Lesion-aware pretraining can encourage mammography vision-language models to encode localized pathology rather than relying on global image shortcuts.
Contribution
The title indicates a pretraining method that injects lesion-region information into image-text representation learning.
Method overview
TODO: Add the verified ROI-guided contrastive learning details from the accepted abstract.
Key findings
- Accepted for ECR 2026 according to the existing repository content.
- TODO: Add public results after the ECR page is available.
Citation
Bhandary Panambur A, Nguyen T-T, Bayer S, Maier A. (2026). LA-CLIP: Lesion-Aware Vision-Language Pretraining for Mammography via ROI-Guided Contrastive Learning. European Congress of Radiology (ECR).
TODO: Add BibTeX.
