四川大学学报（自然科学版）2025，Vol.62Issue(3)：537-547,11.DOI:10.19907/j.0490-6756.240364
金字塔图Transformer全切片病理图像生存预测
Pyramid Graph Transformer for survival prediction from gigapixel whole slide images
摘要
Abstract
Representation learning for Whole Slide Images(WSIs)plays a vital role in automated survival prediction using Graph Neural Network(GNN).The multi-resolution information in WSI,which includes both fine-grained details like cellular phenotypes and coarse-grained characteristics such as tissue structures and global microenvironment,is extensively leveraged in clinical practice for comprehensive analyses.How-ever,existing GNN-based survival prediction methods mainly rely on single resolution images.To address this issue,paper this propose a novel survival prediction framework named PGT:Pyramid Graph Trans-former.PGT hierarchically decomposes WSIs at any resolutions into distinct and heterogeneous graphs,cap-turing and integrating graph representations from local to global to achieve more accurate predictions.The au-thors validate the framework using five public TCGA datasets of various cancer types.The experimental re-sults demonstrate that PGT not only significantly outperforms state-of-the-art models but also exhibits robust generalizability and excels in patient stratification capability.关键词
计算病理/生存预测/全切片图像/图神经网络Key words
Computational pathology/Survival prediction/Whole slide image/Graph neural network分类
信息技术与安全科学引用本文复制引用
李欣洋,张懿,骆梦悦,郑玉玲,王维雯,张海仙..金字塔图Transformer全切片病理图像生存预测[J].四川大学学报（自然科学版）,2025,62(3):537-547,11.基金项目
国家自然科学基金(62476183) （62476183）
四川省科技厅自然科学基金(2024NSFTD0051) （2024NSFTD0051）
