A histopathological image dataset for endometrial disease diagnosis

Recent studies have shown that computer-aided diagnosis (CADx) systems based on deep learning can perform even better than human experts in detecting some common diseases, such as diabetic retinopathy and pneumonia. However, the scarcity of labeled medical imaging data impedes a broad application of this technique to diagnosis of other malignant tumors that receive relatively little attention, especially for women. For example, uterine cancer (also known as endometrial cancer) can seriously affect the female reproductive organs, but it receives less attention than breast cancer and lung cancer. We developed a histopathological image dataset of the endometrium to help alleviate the shortage of pathologically proven cases in using deep learning algorithms to train classification models for endometrial diseases. The dataset comprises ~3,300 joint-photographic-experts-group (JPEG) images from ~500 endometrial specimens stained with hematoxylin and eosin (H&E), each of which has a unique class label derived from the corresponding diagnosis made by three expert pathologists collectively. Note that we will continue to enrich this dataset in the future. All the related source code is available at https://github.com/ssea-lab/DL4ETI.

已有研究表明，基于深度学习的计算机辅助诊断（Computer-aided diagnosis, CADx）系统，在糖尿病视网膜病变、肺炎等常见疾病的检测任务中，性能甚至优于人类专家。然而，标注医学影像数据的稀缺性，阻碍了该技术在其他受关注度较低的恶性肿瘤诊断中的广泛应用，这一局限在女性群体中尤为显著。例如，子宫癌（亦称子宫内膜癌，endometrial cancer）会严重侵袭女性生殖器官，但其受关注程度远不及乳腺癌与肺癌。为此，我们构建了一套子宫内膜组织病理学图像数据集，以期缓解利用深度学习算法训练子宫内膜疾病分类模型时，病理确诊病例匮乏的问题。该数据集包含约3300张联合图像专家小组（Joint Photographic Experts Group, JPEG）格式的图像，素材源自约500份经苏木精-伊红（hematoxylin and eosin, H&E）染色的子宫内膜组织样本；每张图像均配有唯一类别标签，标签基于三位病理专家共同会诊得出的诊断结果生成。请注意，我们后续将持续扩充该数据集。所有相关源代码均可在https://github.com/ssea-lab/DL4ETI获取。