The features of Tissues and Patches for "Predicting microsatellite instabilitiy from histology images with a three-level hierarchical graph fusion model"

This repository contains features and corresponding coordinates of patches and tissues extracted from 430 and 326 histologic images from patients with colorectal and gastric cancers from the TCGA cohort (original whole section SVS images are freely available at https://portal.gdc.cancer.gov/). All images in this library are from formalin-fixed paraffin-embedded (FFPE) diagnostic sections (“DX” on the GDC Data Portal). This blog explains this in detail: http://www.andrewjanowczyk.com/download-tcga-digital-pathology-images-ffpe/ Preprocessing. All SVS slices were pre-processed as follows. According to “Deep learning can predict microsatellite instability directly from histology in gastrointestinal cancer” these histology images were categorized into The histology images were classified as “MSS” (microsatellite stable) or “MSIMUT” (microsatellite unstable or highly mutated) according to “Deep learning can predict microsatellite instability directly from histology in gastrointestinal cancer”, which corresponds to the division of the training and test sets in the article. Patches were extracted at 40x objective magnification and 20x objective magnification, respectively, and the corresponding features were extracted by pre-training resnet48, respectively The features of Tissues are thumbnails obtained at 2.5x objective magnification and further extracted by MedSAM after extracting the masks of the tissues.本仓库包含从TCGA队列的结直肠癌患者的430张组织学图像、胃癌患者的326张组织学图像中提取的斑块与组织的特征及对应坐标，原始全切片SVS图像可于https://portal.gdc.cancer.gov/免费获取。本库中所有图像均来自福尔马林固定石蜡包埋（FFPE）诊断切片（GDC数据门户中标注为"DX"）。以下博客对此进行了详细说明：http://www.andrewjanowczyk.com/download-tcga-digital-pathology-images-ffpe/ 预处理流程 所有SVS切片均按照以下流程进行预处理： 依据《Deep learning can predict microsatellite instability directly from histology in gastrointestinal cancer》一文，本研究将组织学图像划分为微卫星稳定（MSS，microsatellite stable）与微卫星不稳定/高突变（MSIMUT，microsatellite unstable or highly mutated）两类，该分类规则与原文中训练集与测试集的划分方式一致。 分别以40倍物镜放大倍率与20倍物镜放大倍率提取图像斑块，并分别通过预训练ResNet48提取对应特征。 组织特征来源于以2.5倍物镜放大倍率获取的缩略图，在提取组织掩码后，再通过MedSAM进一步提取特征。