Abstract
Preclinical toxicological pathology studies generate large collections
of gigapixel whole-slide images (WSIs) that are time-consuming
to analyze manually. Automated AI workflows offer a promising route to
improve scalability, but their deployment is complicated by the frequent
presence of multiple organs or tissue types on a single slide. Robust tissue
identification and spatial segmentation are therefore important prerequisites
for reliable downstream analysis.
In this study, we introduce two thumbnail-resolution semantic segmentation
mask datasets for preclinical tissue segmentation, derived from
9,442 WSIs across nine studies, four animal models, and 53 tissue categories.
We train four SAM2-based segmentation models across two label
granularities and two backbone sizes, and evaluate performance with
and without slide-level metadata using a censored inference strategy. The
models achieve accurate low-latency segmentation performance, with further
gains when metadata are incorporated, suggesting that thumbnailresolution
semantic segmentation can support automated organ identification
and routing of tissue-specific AI models in preclinical pathology
pipelines. Code and models are publicly available at https://github.
com/baerminator/Organ-izer.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/COMPAYL_010.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
BibTex
@InProceedings{MølBjø_Learning_MICCAISAT2026,
author = { Møller, Bjørn Leth AND Hvid, Henning AND Dalsgaard, Charlotte Maria AND Rittscher, Jens},
title = { { Learning to Organ-ize: Low-resolution Tissue Segmentation for Preclinical Toxicological Pathology } },
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}
}