Abstract
Perineural invasion (PNI) is a pivotal prognostic indicator in pancreatic cancer. However, conventional computational pathology methods, which often rely on random patch sampling, may fail to capture the complex tumor–nerve topology and hierarchical spatial organization of the tissue microenvironment. We propose a Nerve-Centric Heterogeneous Graph (NCHG) framework that explicitly models nerve-oriented invasion patterns for survival prediction. Our framework constructs a hierarchical tissue-level graph centered on neural structures and integrates interactions among three types of nodes: nerve nodes, tumor nodes, and microenvironment nodes. To characterize neural involvement, we introduce intra-neural tumor nodes based on their geometric location within the reconstructed nerve and their proximity to the nerve boundary, as defined by the boundary-adjacent area. In addition, a biologically inspired dual-attention mechanism, consisting of Structural Attention and Feature Similarity Attention, is designed to capture spatially directional tumor–nerve interactions and infiltration patterns. Extensive experiments across three multicenter cohorts demonstrate the robustness and generalizability of the proposed framework. NCHG achieves a C-index of 0.6437 and an AUC of 0.6857 on TCGA-PAAD, and 0.6504 and 0.7041 on GL, respectively. On the independent SZY cohort, it achieves a C-index of 0.6243 and an AUC of 0.6625, demonstrating superior performance and strong cross-center generalization.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/CMMCA2026_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/forum?id=cSEWPenigk
BibTex
@InProceedings{HanWen_NerveCentric_MICCAISAT2026,
author = { Han, Wennan AND Cao, Ying AND Wang, Bingxue AND Cai, Zhenghua AND Cao, Yingying AND Zhang, Xiuyuan AND Qian, Chunjun AND Gui, Luying},
title = { { Nerve-Centric Heterogeneous Graph Learning for Tumor–Nerve Interaction Modeling in Survival Prediction } },
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}
}