Abstract
The REG2026 challenge requires systems to produce both a structured diagnostic workflow and a pathology report from a single whole-slide image (WSI). End-to-end methods generate reports or workflows. When both are needed, they easily disagree on basic specimen fields like organ and procedure, even when each output is individually plausible. We formalize this as workflow-report field consistency and introduce a framework that predicts a structured clinical context from the WSI using frozen encoders, renders both outputs from this context through template mapping, and enforces consistency through bounded report refinement and consistency verification. Our system ranks first on the REG2026 Test Phase 1 leaderboard with an overall score of 0.8125. On a separate 400-slide cohort, bounded report refinement and consistency verification substantially improve performance under a fixed path. Concatenating multi-encoder representations does not improve the renderer, but removing any encoder degrades the refiner. These results show that multi-encoder representations help refinement, and that consistency is separately measurable and controllable. Code is available at https://github.com/haloooo47/SEUMIA_official_repo_reg2026.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/REG2026_017.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=8CFNlsiYUg
BibTex
@InProceedings{WuYih_Structured_MICCAISAT2026,
author = { Wu, Yihong AND Xue, Jun AND Zhong, Qirui AND Xue, Cheng},
title = { { Structured Context Rendering for Consistent Pathology Workflows and Reports from Whole-Slide Images } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17252},
month = {pending},
page = {pending}
}