Glomerulus in multiple stains obtained via large-scale web image mining

We collected over 10,000 compound figures through the NIH Open-i search engine with the keywords “glomerular OR glomeruli OR glomerulus”. Then we used our compound figure separation network to process all compound images to get over 36,000 glomeruli pathologies obtained by different microscopy or in different stains with a confidence threshold of 0.7. We split the data into 3 classes by different microscopy which are light microscopy, electron microscopy, and fluorescent.

本研究借助美国国立卫生研究院（National Institutes of Health，NIH）的Open-i图像搜索引擎，以"glomerular（肾小球的）、glomerulus（肾小球，单数）、glomeruli（肾小球，复数）"为关键词，检索收集到10000余张复合图像（compound figure）。随后，我们采用自研的复合图像分离网络（compound figure separation network）处理所有复合图像，设置置信度阈值（confidence threshold）为0.7，最终得到36000余张来自不同显微镜技术或不同染色方式的肾小球病理图像。最后，我们按照所用显微镜类型将数据集划分为3个类别，分别为光镜（light microscopy）、电子显微镜（electron microscopy）与荧光显微镜（fluorescent microscopy）。