CAMELYON17
收藏资源简介:
CAMELYON17数据集是第二届CAMELYON病理学挑战赛的核心数据,旨在评估和推动乳腺癌淋巴结转移的自动检测与分类算法。该任务具有高度临床相关性,传统上依赖病理学家对全切片图像进行显微镜评估,而自动化方案有望减轻病理学家工作负担并降低诊断主观性。数据集包含来自荷兰5个医疗中心(Radboud大学医学中心、Canisius-Wilhelmina医院、乌得勒支大学医学中心、Rijnstate医院、Oost-Nederland病理实验室)的1000张全切片图像(WSI),格式为TIFF。数据分为训练集和测试集,每个中心在两个集合中各包含20个患者。患者pN分期和切片级标签存储在stages.csv文件中。此外,从每个中心选取10张切片进行详尽标注,标注以XML格式提供,兼容ASAP软件,包含“Tumor”(肿瘤区域)和“Exclusion”(非肿瘤区域,从肿瘤标注中切割出的区域)两组标注。掩膜文件覆盖50张标注WSI和50张随机阴性WSI,掩膜中像素值1表示正常组织,2表示肿瘤区域(正常组织自动检测自参考方法,可能存在误差;肿瘤区域由标注多边形转换而来)。数据中有一张图像(patient_103_node_1.tif)因文件损坏被替换为来自同一淋巴结且转移类别相同的新图像。所有文件均提供MD5校验和以确保完整性,并附带许可文件。
The CAMELYON17 dataset is the core data of the second CAMELYON pathology challenge, aiming to evaluate and promote automatic detection and classification algorithms for breast cancer lymph node metastasis. This task is highly clinically relevant, as it traditionally relies on pathologists microscopic evaluation of whole-slide images, while automated solutions are expected to reduce pathologists workload and diagnostic subjectivity. The dataset consists of 1,000 whole-slide images (WSIs) in TIFF format from five medical centers in the Netherlands (Radboud University Medical Center, Canisius-Wilhelmina Hospital, University Medical Center Utrecht, Rijnstate Hospital, and Pathologisch Laboratorium Oost-Nederland). The data is divided into a training set and a test set, with each center contributing 20 patients to each set. Patient pN stage and slide-level labels are stored in the stages.csv file. Additionally, 10 slides from each center are exhaustively annotated, with annotations provided in XML format compatible with ASAP software, containing two groups: Tumor (tumor regions) and Exclusion (non-tumor regions cut out from tumor annotations). Mask files cover 50 annotated WSIs and 50 random negative WSIs, with pixel values of 1 indicating normal tissue and 2 indicating tumor regions (normal tissue is automatically detected from a reference method, which may have errors; tumor regions are converted from annotated polygons). One image (patient_103_node_1.tif) was replaced due to file corruption with a new image from the same lymph node and the same metastasis category. All files are provided with MD5 checksums to ensure integrity, along with a license file.
CAMELYON17 数据集概述
1. 总体概览
CAMELYON17 是一个医学图像相关的数据集,包含 1156 条记录/幻灯片(slide)。该数据集主要用于淋巴结转移检测相关任务,包含图像、掩膜、注释等多种类型的数据文件。
2. 存储结构
数据集存储于 okbro1234/CAMELYON17 仓库中,主要包含以下内容:
- README.md:说明文件
- annotations/:XML 格式的注释文件(如
patient_004_node_4.xml等),共 50 个文件
数据结构中还包括 images、masks 子文件夹以及 evaluation 相关文件,完整文件清单见 manifest 文件。
3. 数据清单(Manifest)及字段
数据清单包含 5 个字段,均无缺失值:
| 字段 | 数据类型 | 缺失值数量 |
|---|---|---|
| key | 对象 | 0 |
| rel_path | 对象 | 0 |
| subfolder | 对象 | 0 |
| file_size | 整数 | 0 |
| last_modified | 对象 | 0 |
4. 分类标签
根据 subfolder 字段的数据分布如下:
| 类别 | 记录数 | 占比 |
|---|---|---|
| images | 1000 | 86.51% |
| masks | 100 | 8.65% |
| annotations | 50 | 4.33% |
| (root) | 4 | 0.35% |
| evaluation | 2 | 0.17% |
5. 局限性与注意事项
- 数据类别分布存在严重不平衡:
images类占比高达 86.51%,而其他类别占比很低,可能影响模型训练效果或数据分析的公正性。




