Abstract
Segmenting the mitral valve in intraoperative images remains challenging because blood, glare, surgical instruments, and visually similar surrounding tissue often produce multiple plausible segmentation masks. The challenge is therefore not generating candidate masks, but determining which prediction is best supported by the surgical evidence. We propose RATS-SAM, a lightweight reasoning layer that refines uncertain predictions by integrating relational scene cues and sparse temporal evidence. Anatomical structures and surgical context provide spatial evidence for where the valve should appear, while neighboring sparsely sampled frames offer complementary temporal evidence when their correspondence is reliable. Rather than freely overriding the current prediction, a bounded confidence gate selectively incorporates contextual evidence only when it is sufficiently consistent and reliable. On one held-out surgical video, RATS-SAM improves the positive-frame Dice score from 0.7829 to 0.7976 while reducing the negative-frame false-positive fraction from 0.0514 to 0.0027. Combined with a complementary appearance ensemble and conservative inference rules, the final system achieves a Dice score of 0.8338 without producing false positives on target-negative frames. Although these results are obtained on a validation set rather than an independent test set, they demonstrate that reasoning with surgical evidence can improve both the accuracy and reliability of proposal-based surgical segmentation.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MVAA_093.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=aghUfOsosy
BibTex
@InProceedings{LeePin_RATSSAM_MICCAISAT2026,
author = { Lee, Ping-Lun AND Kao, Yun-Ching AND Lin, Cheng-Kuan AND Tseng, Yu-Chee AND Huang, Eng-Yen AND Huang, Tzu-Ting AND Huang, Kuan-Ching},
title = { { RATS-SAM: Reasoning with Surgical Evidence for Mitral Valve Segmentation } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17268},
month = {pending},
page = {pending}
}