Abstract
Pathological diagnosis follows an ordered reasoning process that accompanies the final disease diagnosis. Free-text vision-language generation may introduce unsupported findings and reduce keyword and n-gram scores. In the MICCAI 2026 REG2 challenge corpus, 699 report strings reproduce 11,216 of 11,220 training reports exactly. We therefore formulate the REG2 task as structured slot prediction followed by likelihood-based retrieval using two pathology foundation models and gated-attention multiple-instance learning. On an internal hold-out set of 1,672 slides, this approach achieved a workflow reasoning score of 0.8838, an edge F1 score of 0.9360, an answer similarity of 0.9198 and a report score of 0.8360. The overall scores on the official REG2 test leaderboard were 0.7724 in Test Phase 1 and 0.6850 in Test Phase 2. Exact report reproduction reached 53.3%. Visual grounding used a fixed saturation threshold in the hue, saturation and value (HSV) colour space without a language model. SlotPath demonstrates that structured prediction and retrieval can support reasoning-guided pathology reporting without free-text generation, although generalisation depends on report-bank coverage.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/REG2026_005.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=O8gAIVqUXi
BibTex
@InProceedings{KacSan_SlotPath_MICCAISAT2026,
author = { Kachole, Sanket AND Bakas, Spyridon},
title = { { SlotPath: Structured Slot Prediction and Retrieval for Reasoning-Guided Pathology Reporting } },
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}
}