CellHIST-Bench/CellHIST-Bench

CellHIST-Bench是一个用于基于组织学的空间基因表达推断的基准数据集。它提供了配对的整张切片组织病理学图像、空间转录组学标签、细胞分割结果、以点为中心的补丁元数据以及细胞到补丁的对应信息。该数据集支持两种空间转录组学注释级别：点分辨率空间转录组学，其中每个点与局部组织学补丁和点级别基因表达向量相关联；单细胞分辨率空间转录组学，其中每个细胞与单细胞基因表达向量相关联，并可以链接到图像检测到的细胞。CellHIST-Bench旨在支持从组织病理学图像预测空间基因表达的弱监督学习方法的基准测试。CellHIST-Bench is a benchmark dataset for histology-based spatial gene expression inference. It provides paired whole-slide histopathology images, spatial transcriptomics labels, cell segmentation results, spot-centered patch metadata, and cell-to-patch correspondence information. The dataset supports two levels of spatial transcriptomics annotations: spot-resolution spatial transcriptomics, where each spot is associated with a local histology patch and a spot-level gene expression vector; and single-cell-resolution spatial transcriptomics, where each cell is associated with a single-cell gene expression vector and can be linked to image-detected cells. CellHIST-Bench is designed to support benchmarking of weakly supervised learning methods for predicting spatial gene expression from histopathology images.