Abstract
Surgical Phase Recognition (SPR) automates surgical video analysis for intraoperative guidance and skill training. Most surgical procedures are performed following standardized workflows, which impose a well-defined structure on phase order and transition. However, existing SPR methods typically predict phases from visual features alone and rarely exploit this structured prior. Therefore, we propose Bridging Implicit and Explicit Modeling (BIEM), a hybrid framework coupling a Conditional Random Field for modeling phase transition probabilities with a Transformer-based implicit encoder. Our contributions are threefold: (i) proposing BIEM with a joint cross-entropy (CE) and negative log-likelihood (NLL) loss for stable end-to-end optimization; (ii) designing Sliding-Window Conditional Random Field (SWCRF) to incorporate surgical workflow priors; (iii) introducing HipScope39, the first dedicated hip arthroscopy dataset comprising 39 procedures. Experimental results demonstrate that BIEM achieves state-of-the-art accuracy of 89.3% on HipScope39 and competitive accuracy of 94.2% (relaxed) on Cholec80 and 86.0% on AutoLaparo, demonstrating robust cross-domain generalizability.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/EMA_028.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: Not Available
BibTex
@InProceedings{ZhuYon_BIEM_MICCAISAT2026,
author = { Zhu, Yongkang AND Du, Xinzhe AND Shao, Shuwei AND Ma, Shixing AND Sha, Shiyu AND Liu, Yi AND Yin, Qingfeng AND Liu, Wenguang AND Min, Zhe},
title = { { BIEM: Bridging Implicit and Explicit Temporal Modeling for Surgical Phase Recognition } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17265},
month = {pending},
page = {pending}
}