Abstract
The Critical View of Safety is the clinical standard for preventing bile duct injury during laparoscopic cholecystectomy, yet its three anatomical criteria are assessed subjectively and inter-rater agreement among expert surgeons remains limited. Existing computational methods reduce multi-rater annotations to a single aggregated label, discarding the disagreement signal. In this paper, we benchmark seven frozen image encoders across general-purpose and surgical pretraining domains to predict per-frame inter-rater disagreement, evaluating both individual encoders and their combination via a gradient-boosted ensemble. On the Endoscapes-CVS201 cohort of 11,090 frames extracted from 201 cholecystectomy videos, evaluated under a strict video-level split with Cholec80-overlapping videos excluded from the test set, the ensemble of all encoders achieved a Spearman correlation of 0.453, outperforming the handcrafted visual baseline by Δr = 0.349. No single encoder dominated; complementary representations from both surgical foundation models (EndoViT, SurgeNetXL) and general-purpose backbones collectively captured the disagreement signal more effectively than any individual model. Disagreement-guided frame selection recovered 1.7× more high-disagreement frames than random selection at a 20% review budget, with a precision of 0.390 at a 5% budget. These results demonstrate that inter-rater disagreement in CVS assessment is predictable from a single laparoscopic frame, supporting integration into confidence-aware annotation and quality-control pipelines for surgical AI.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/CREATE_007.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=3gPYSme8ls
BibTex
@InProceedings{RajDas_Surgical_MICCAISAT2026,
author = { Raju, Dasari Naga AND Kancharla, Vamshi Krishna AND Kumar, Kaveti Pavan},
title = { { Surgical Foundation Model Ensembles for Predicting Annotator Disagreement in Critical View of Safety Assessments } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17275},
month = {pending},
page = {pending}
}