Abstract
Gastric cancer precursor conditions, such as H. pylori infection,
intestinal metaplasia, and atrophy, frequently co-occur in patients,
as modeled by the Correa Cascade. However, current models for wholeslide
image classification are not optimized for multi-label settings, failing
to model potentially useful inter-pathology dependencies or provide
per-class interpretability. We introduce QLabelMIL, a multiple-instance
learning aggregator that uses class-specific queries and a transformer decoder
to model patch-to-label relationships, while natively supporting
joint multi-label classification and per-class heatmaps. A Graph Convolutional
Network variant further leverages the prior distribution of label
co-occurrences to score per-label features. It is evaluated on one of the
largest gastric histopathology datasets, comprising 5764 WSIs from 3412
patients, against several established MIL frameworks like CLAM, Trans-
MIL, and MambaMIL. QLabelMIL achieved better predictive performance
with a macro-averaged AUROC of up to 0.9186 and competitive
calibration metrics, offering an efficient and interpretable solution.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/COMPAYL_060.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: https://papers.miccai.org/miccai-2026-sat/supp/COMPAYL_060_supp.pdf
Link to Open Review
BibTex
@InProceedings{NetPed_QLabelMIL_MICCAISAT2026,
author = { Neto, Pedro C. AND Lopes, Rita N. AND Prado e Castro, Lígia},
title = { { QLabelMIL: Inter-Pathology Query Decoding for Multi-Label Gastric Histopathology } },
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}
}