Abstract
Clinical histopathology is inherently multi‑view: a single patient case is represented by multiple whole‑slide images (WSIs) sampled from different tissue blocks or depths. Most computational pathology pipelines remain slide‑centric, producing independent slide‑level predictions and aggregating them post hoc via heuristics. Recent evidence shows that combining information across WSIs improves patient‑level assessment, yet existing approaches aggregate independently processed slides late in the pipeline. We propose \emph{TumorGraphs}, a framework that models a patient’s histopathology data as a single graph integrating patches from all WSIs. Within each slide, patches are connected via spatial adjacency, preserving local tissue topology. Across slides, cross-slide edges connect patches from different WSIs of the same patient based on embedding-space similarity, establishing feature-level correspondences that enable information exchange across anatomically distinct sections. We investigate three relational designs—homo-ge-ne-ous graphs, heterogeneous graphs with typed intra‑ and inter‑slide relations, and hypergraphs connecting morphologically similar patches—under a controllable connectivity budget. Experiments on renal clear cell carcinoma and lung cancer cohorts for binary 12-month survival prediction show that explicit patient‑level integration consistently outperforms slide‑level baselines. A semantic audit using a histology foundation model provides evidence that learned hyperedges connect morphologically coherent tissue across slides. These results support a shift from slide‑centric aggregation toward structured patient‑level modelling in multi‑WSI computational pathology.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/GRAIL_011.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: https://papers.miccai.org/miccai-2026-sat/supp/GRAIL_011_supp.pdf
Link to Open Review
Open Review Page: https://openreview.net/forum?id=e7ktQ7jGhj
BibTex
@InProceedings{VerMar_TumorGraphs_MICCAISAT2026,
author = { Verdelho, Maria Rita Fonseca AND Dubrawski, Artur AND Bernardino, Alexandre AND Barata, Catarina},
title = { { TumorGraphs: Patient-Level Fusion of Multi-WSI Histopathology } },
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}
}