Abstract
REG2026 requires a whole-slide image to be mapped to an ordered pathology workflow under fixed runtime and schema constraints. We use diagnosis-conditioned retrieval. An ensemble predicts the organ from sparse tiles, and organ-specific attention models predict the primary diagnosis from denser tile bags. Cosine search among cases with both predicted labels returns one annotated workflow. On 350 held-out Test Phase 1 cases, the submitted container obtained a workflow score of 0.7087 and an overall score of 0.7961. On Test Phase 2, the workflow and overall scores were 0.5969 and 0.7178. We separately evaluated the candidate restriction with every same-center exemplar excluded. With reference labels, adding diagnosis to organ routing increased workflow scores from 0.6379 to 0.9040 in one center and from 0.6544 to 0.9587 in the other. The retrieved workflow still disagreed with the reference procedure in 43.4% of cases from one center and with grade-related fields in 32.3% of cases from the other. Under correct routing, diagnosis strongly constrained retrieval but did not determine every case-specific workflow field.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/REG2026_008.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=rdOgKZaBaO
BibTex
@InProceedings{CheYix_DiagnosisConditioned_MICCAISAT2026,
author = { Chen, Yixin},
title = { { Diagnosis-Conditioned Retrieval for Structured Pathology Workflows } },
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}
}