Abstract
A key challenge in the automatic generation of pathology reports from gigapixel WSI images is the black-box nature of the reasoning process. In this study, in accordance with the MICCAI 2026 REG2 Challenge, we propose a two-module framework that mimics the thought process of a pathologist. The reasoning module uses a minimum path coverage algorithm to extract and learn optimal diagnostic pathways from a DAG. The grounding module performs region-of-interest (ROI) identification that is robust to image perturbations through binary classification using consistency loss. The overall score for Test Phase 1 in the official challenge system is 0.5444. The code is available at https://github.com/sk-jp/REG2026.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/REG2026_010.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=GAghpM2XI3
BibTex
@InProceedings{KonSat_PRGRGen_MICCAISAT2026,
author = { Kondo, Satoshi AND Kasai, Satoshi},
title = { { PRG-RGen: Pathologist Reasoning-Guided Report Generation on WSIs via Multimodal LLMs } },
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}
}