Abstract
Automated Critical View of Safety (CVS) assessment during laparoscopic cholecystectomy could reduce bile duct injuries, but existing methods require additional supervision (segmentation masks, bounding boxes, or language prompts), increasing annotation effort and limiting large-scale deployment. We ask whether a general-domain vision foundation model (FM) can remove this dependence on extra supervision. To answer this, we benchmark five pretrained encoders under identical conditions on two CVS benchmarks, Endoscapes 2023 and SAGES 2024, and find EVA-02 L the strongest. Fine-tuned per frame without additional supervision, this FM alone matches or surpasses prior methods. On top of the frozen FM, PercEVA-CVS adds a keyframe-anchored Perceiver aggregator that contributes a smaller, dataset-dependent gain and lets us characterise the temporal context CVS assessment requires. A sufficiently strong FM removes the need for additional supervision, and the frozen PercEVA-CVS pipeline builds on this to set a new state of the art on both benchmarks. Results are reported as mean ± std over three seeds.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/SafeSurg_021.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: https://openreview.net/forum?id=WChUTL774d
BibTex
@InProceedings{CañSer_AnnotationEfficient_MICCAISAT2026,
author = { Cañar, Sergio AND Vera Rincon, Javier Santiago AND Tovar Sánchez, Isabel Sofía AND Arbelaez, Pablo},
title = { { Annotation-Efficient Critical View of Safety Assessment with Vision Foundation Models } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS pending},
month = {pending},
page = {pending}
}