Abstract
We describe our submission to the Pathologist REasoningGuided REport Generation Challenge (REG2 2026), which asks an algorithm to turn a hematoxylin and eosin (H&E)-stained whole-slide image (WSI) into a pathology report together with the reasoning trajectory of question–answer pairs that leads to it, and to answer per-region visual-grounding questions. Rather than treating this as free-form generation, we adopt an impression-first, modular design that follows how a pathologist reads a slide: a frozen foundation model encodes each tile once, and attention-based multiple-instance-learning (MIL) heads first form a slide-level diagnostic impression. Conditioned on that impression, a visual-question-answering module answers the structured questions, proposes the clinically appropriate follow-up questions, and detects secondary findings. The trajectory is grown by a deterministic breadth-first traversal in which a learned reasoning module expands the diagnostic graph and answerer models answer each node. Out-of-distribution cases, flagged by the impression head’s entropy, are handed to a slide-level vision–language model adapted with low-rank adaptation (LoRA) for a zero-shot organ and coarse impression. The report is then assembled by templating the established fields: organ, procedure, histological subtype, grade, and secondary findings. On in-distribution validation it attains a Workflow-Reasoning score of 0.9234; on the held-out Test Phase 1 leaderboard, under distribution shift, an overall 0.7652 (Workflow-Reasoning 0.7110, Visual-Grounding 0.8917), ranking ninth.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/REG2026_003.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: https://papers.miccai.org/miccai-2026-sat/supp/REG2026_003_supp.pdf
Link to Open Review
Open Review Page: https://openreview.net/forum?id=ZDkp7UaPUT
BibTex
@InProceedings{ChaChi_SlideImpressionGuided_MICCAISAT2026,
author = { Chang, Chia-Ping AND Yeh, Yi-Chen AND Liang, Wen-Yih},
title = { { Slide-Impression-Guided Modular Reasoning for Pathology Report Generation } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17251},
month = {pending},
page = {pending}
}