PTCGA200, PCam200, segPANDA200
收藏arXiv2023-06-13 更新2024-06-21 收录
下载链接：
https://drive.google.com/drive/folders/1kPWiFxL5HxVRM4antSPq9FzjBIbmkGWA
下载链接
链接失效反馈官方服务：
资源简介：
本研究汇编了三个用于深度学习预训练和基准测试的公共病理H&E图像数据集，包括一个大型数据集PTCGA200和两个小型数据集PCam200与segPANDA200。PTCGA200由约10000张H&E诊断幻灯片中的500个200μm尺度方形补丁组成，总计511万张图像。PCam200和segPANDA200分别是通过系统地滑动裁剪区域从Camelyon2016挑战数据集和PANDA挑战数据集创建的。这些数据集用于肿瘤分类和组织分割基准，旨在验证大规模预训练在病理图像上的有效性，特别是在自监督学习框架中。
This study compiles three public H&E-stained pathology image datasets for deep learning pre-training and benchmarking, comprising one large-scale dataset PTCGA200 and two small-scale datasets PCam200 and segPANDA200. PTCGA200 consists of 500 square patches at a 200 μm scale sampled from approximately 10,000 H&E-stained diagnostic slides, with a total of 5.11 million images. PCam200 and segPANDA200 were respectively constructed via systematic sliding-window cropping from the Camelyon2016 Challenge Dataset and PANDA Challenge Dataset. These datasets serve as benchmarks for tumor classification and tissue segmentation tasks, aiming to validate the efficacy of large-scale pre-training on pathology images, especially within self-supervised learning frameworks.
提供机构：
山梨大学病理科, 日本; 系统研究与开发中心, 技术局, NS解决方案公司, 日本
创建时间：
2023-03-28
