Abstract
In order to produce a trustworthy pathology report based on a whole slide image (WSI), it is necessary to go beyond diagnos- tic accuracy, clearly explaining how histomorphological evidence con- nects to the final diagnosis. We present BEAVER, a branch-aware, confidence-gated reasoner for whole slide images, developed for the MIC- CAI 2026 REG2 challenge, comprising 11,220 H&E stained WSIs from seven organs across multiple international datasets. BEAVER translates WSIs into diagnostic reports and provides a chain-of-thought (CoT) that articulates how the diagnostic criteria support the final diagnosis. By using logical rules for decision branches and standardized frameworks for diagnostic verification, the system maintains consistency without us- ing large language models. BEAVER also applies confidence-gated ver- ification to criteria whose certainty exceeds a certain threshold. With the visual encoder frozen, BEAVER trains only 2.15 M task-specific parameters and constructs reports in 0.01 s from slide features. Also, end-to-end processing takes approximately 20 s per slide on one GPU. On the external Test Phase 1, BEAVER ranks 5th with an overall score of 0.794 (Metric A = 0.705, Metric B = 1.000), with a 0.0187 point gap from the leading entry. In Test Phase 2, it scored 0.730 overall (Metric A = 0.614, Metric B = 1.000). Code is available at https://github.com/kim-kyunga/beaver.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/REG2026_012.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=tBh1nEVVva
BibTex
@InProceedings{KimKyu_ABranchAware_MICCAISAT2026,
author = { Kim, Kyung A. AND Eom, Irene Y. AND Shim, Hyo Sup},
title = { { A Branch-Aware, Confidence-Gated Reasoning Framework for 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}
}