乳腺癌研究数据集|图像处理数据集

<p>BCI数据集是由北京朝阳医院和首都医科大学合作创建的，专注于乳腺癌免疫组化图像生成。该数据集包含4872对已结构级对齐的H&amp;E和IHC染色图像，用于研究从H&amp;E到IHC染色图像的转换算法。数据集的构建过程包括切片准备、扫描、投影变换、elastix注册、图像精化和补丁选择。BCI数据集的应用领域主要集中在通过深度学习技术生成高质量的IHC染色图像，以辅助乳腺癌的诊断和治疗计划制定。</p><p><br></p><p><br></p><ul><li>数据特点：该数据集主要包含 51 对 WSI（ Whole Slide Imaging，全切片成像）图像对，涉及 HER2（人表皮生长因子受体 2）的四种表达水平（0、1+、2+、3+），包含 HE（苏木精 - 伊红）染色图像和 IHC（免疫组织化学）染色图像，且所有数据均经过脱敏处理，已去除隐私信息。</li><li>数据规模：基于 51 对 WSI pairs 构建，具体单张图像数量未明确提及，但包含训练集、验证集和测试集等数据划分。该数据集包含 4870 组注册图像对，涵盖了各种 HER2 表达水平。基于 BCI，作为一个次要贡献，我们进一步构建了一种金字塔 Pix2pix 图像生成方法，其 HE 到 IHC 的转换结果优于其他当前流行的算法。大量实验表明，BCI 为现有的图像转换研究提出了新的挑战。此外，BCI 也为基于合成的 IHC 图像的未来病理学研究 HER2 表达评估打开了大门。</li><li>应用场景：主要用于乳腺癌免疫组织化学图像生成相关的学术研究、教学、科学出版物或个人实验等非商业用途，可支持如 Pyramid Pix2pix 等模型的训练与测试，以实现 HE 图像到 IHC 图像的转换等任务。</li></ul>

<p>BCI Dataset was co-developed by Beijing Chaoyang Hospital and Capital Medical University, focusing on the generation of breast cancer immunohistochemical (IHC) staining images. This dataset contains 4872 structurally aligned pairs of H&E and IHC stained images, which is intended for research on translation algorithms converting H&E-stained images to IHC-stained images. The construction process of the dataset includes slide preparation, scanning, projection transformation, elastix registration, image refinement and patch selection. The main application scenarios of the BCI dataset center on generating high-quality IHC-stained images via deep learning technologies to assist breast cancer diagnosis and treatment planning.</p><p><br></p><p><br></p><ul><li>Data Characteristics: This dataset mainly includes 51 pairs of Whole Slide Imaging (WSI) image pairs, involving four expression levels of HER2 (Human Epidermal Growth Factor Receptor 2): 0, 1+, 2+ and 3+. It contains both H&E (hematoxylin-eosin) stained images and IHC (immunohistochemistry) stained images, and all data has been de-identified with private information fully removed.</li><li>Data Scale: It is constructed based on 51 WSI pairs. The exact number of individual images is not specified, but it includes standard data splits such as training, validation and test sets. This dataset contains 4870 registered image pairs covering all HER2 expression levels. As a secondary contribution derived from the BCI dataset, we further developed a pyramid Pix2pix image generation method, whose HE-to-IHC translation performance outperforms other current mainstream algorithms. Extensive experiments demonstrate that BCI poses new challenges to existing image translation research. In addition, BCI also opens the door for future pathological research on HER2 expression assessment using synthetic IHC images.</li><li>Application Scenarios: It is primarily designed for non-commercial purposes including academic research, teaching, scientific publications and personal experiments related to breast cancer immunohistochemical image generation. It can support the training and testing of models such as Pyramid Pix2pix to accomplish tasks including H&E image-to-IHC image translation.</li></ul>