Abstract
Out-of-body (OoB) detection identifies frames in which the laparoscopic camera exits the patient. It is a prerequisite for privacy preserving surgical video anonymisation and downstream AI tasks, yet no standardized evaluation protocol exists for the robotic domain under annotation scarcity. We address this gap with three contributions. First, a lightweight timestamp annotation protocol requiring only 15–20 minutes per video, enabling institutions to annotate robotic procedures without per-frame clinical review. Second, a controlled comparative evaluation of 17 da Vinci robotic procedures across 12 procedure types (≈24h26 min; 5 training, 12 test), on which OoBNet, IODA (zero-shot and supervised), and four ablated variants are all evaluated under identical training conditions. Third, GatedTextFusion-LSTM, a strong baseline that fuses SurgVLP visual embeddings with dynamically gated class-specific text prompts and models the sequence with a stateful LSTM, achieving F1smooth =0.9849 and AP=0.9956 on the held-out test set, outperforming OoBNet supervised (F1smooth =0.9390, AP=0.9572) and IODA supervised (F1smooth=0.9663, AP=0.9883) with 24%fewer parameters compared to IODA. Notably, our model yields only 1 missed (false negative) OoB segment across the 12 test videos, and this residual error occurs at an IB↔OoB transition frame rather than
mid-segment inside an out-of-body scene. A semi-supervised extension via LSTM self-training on 15 pseudo-labelled videos further improves AP to 0.9964, demonstrating that the evaluation protocol enables label efficient research. On the gynaecological procedure, the model achieves perfect segment-level accuracy (zero FP/FN), addressing the heightened privacy sensitivity of such cases for women’s health surgery.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/COLAS_016.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=06dJ7gYCbr
BibTex
@InProceedings{KumSuy_SurgAnon_MICCAISAT2026,
author = { Kumar, Suyash AND Frey, Sébastien AND Ekmekci, Ezem Sura AND Lalli, Duccio AND Rudondy, Quentin AND Berthet-Rayne, Pierre AND Delingette, Herve AND Bremond, Francois AND Ayache, Nicholas},
title = { { SurgAnon: Out-of-Body Detection in Robotic Surgical Video via Gated Multi-Prompt Text Fusion with Temporal Modeling } },
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}
}