Cancer prognostication is a challenging task in computational pathology that requires context-aware representations of histology features to adequately infer patient survival. Despite the advancements made in weakly-supervised deep learning, many approaches are not context-aware and are unable to model important morphological feature interactions between cell identities and tissue types that are prognostic for patient survival. In this work, we present Patch-GCN, a context-aware, spatially-resolved patch-based graph convolutional network that hierarchically aggregates instance-level histology features to model local- and global-level topological structures in the tumor microenvironment. We validate Patch-GCN with 4,370 gigapixel WSIs across five different cancer types from the Cancer Genome Atlas (TCGA), and demonstrate that Patch-GCN outperforms all prior weakly-supervised approaches by 3.58-9.46%. Our code and corresponding models are publicly available at https://github.com/mahmoodlab/Patch-GCN.
翻译:在计算病理学中,癌症预测是一项具有挑战性的任务,需要从环境角度对病理学特征进行描述,才能充分推断病人的存活。尽管在微弱监督的深层学习中取得了进步,但许多方法并非环境意识,无法模拟细胞特性与组织类型之间重要的形态特征相互作用,而细胞特性与组织类型对于病人生存来说是预测性的。在这项工作中,我们介绍了Patch-GCN, 一种环境认知、空间溶解的、基于空间的、基于补丁的图形革命性网络,按等级将实例级的特征汇总到肿瘤微观环境中的当地和全球层次的地形结构模型中。我们用癌症基因组图集图(TCGA)的5种癌症类型4,370千兆像素 WSIs验证了Patch-GCN, 并表明Patch-GCN在3.58-9.46%之前所有受微弱监督的方法上都比强。我们的代码和相应的模型可在https://github.com/mahoodlab/Patch-GCN上公开查阅。