Abstract
Pathological staging of rectal cancer depends on tumour invasion and regional lymph node involvement, which guide treatment and prognosis. Predicting stage from haematoxylin and eosin (H&E) whole-slide images is difficult because labels are available only at the patient level, while the relevant histological patterns are spread across large tissue regions. Most weakly supervised methods aggregate image patches independently and do not model tissue interactions or spatial organisation.We propose a tissue-aware graph learning framework for pathological T-stage prediction from whole-slide images. Each tissue patch is encoded using deep histological features and a pretrained tissue prior, allowing each graph node to represent both tissue type and visual appearance. We construct a patient-specific graph by connecting neighbouring tissue regions. Multi-scale graph attention captures local and global tissue interactions, while graph-informed sparse pooling selects coherent tissue regions linked to tumour progression. A shared graph representation is then used for multi-task prediction of pathological T-stage status.Experiments on the TCGA-READ cohort show that modelling tissue identity and spatial relationships improves pathological staging from routine H&E whole-slide images. The proposed framework achieves an AUC of 0.886 for pathological T-stage prediction and produces interpretable graph representations that highlight tissue regions associated with disease stage.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/CaPTion_007.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/profile?id=~Sezal_Rana1
BibTex
@InProceedings{RanSez_Learning_MICCAISAT2026,
author = { Rana, Sezal AND Chauhan, Garima Ketan AND Sree, M. Rahul AND Singh, Sneha},
title = { { Learning Tissue Interactions for Pathological T-Staging of Rectal Cancer from Whole-Slide Images } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17278},
month = {pending},
page = {pending}
}