Glioma C6 dataset for cell segmentation

Training and test images of Glioma C6 cells imaged using phase-contrast microscopy for the task of cell segmentation. The example shows a phase-contrast image of Glioma C6 cells and the manually annotated segmentation mask. Data type: Phase-contrast images with corresponding annotations in COCO format. Microscopy data type: 2D phase-contrast images recorded at 24 or 72-hour intervals after cell seeding. Microscope: BestScope BS-2092 microscope in phase-contrast mode, equipped with 10× and 20× objective lenses. Cell type: C6 glioma cells (rat glial tumor cells, ATCC CCL-107). Image size: 2592 × 1944 px² . File format: .tif (8-bit). File naming convention: The file names include the cultivation time (24h or 72h) and the microscope objective lens (10× or 20×), e.g., spec_24h_10x_17.tif. Dataset subsets: Glioma C6-spec: 45 images captured under strictly controlled imaging conditions, divided into training (30 images), validation (4 images), and test (11 images) subsets. Glioma C6-gen: 30 images captured under varied imaging and seeding conditions, designed to test model generalization. Annotations: Over 20,000 annotated objects across both subsets, including 12,000 cell annotations and 7,800 soma annotations. Glioma C6-spec annotations include Type A cells (spheroid/spindle), Type B cells (flat/spread) and soma (nucleus body). Glioma C6-gen subset annotations consist only of general cell instances, without differentiation into types or nuclei. Article reference: "Glioma C6: A Novel Dataset for Training and Benchmarking Cell Segmentation," Malashin et al., 2025. Authors: Roman Malashin¹ ², Svetlana Pashkevich³, Daniil Ilyukhin¹, Arseniy Volkov³, Valeria Yachnaya¹ ², Andrey Denisov³, Maria Mikhalkova¹ Affiliation(s): ¹ Pavlov Institute of Physiology, Russian Academy of Science ² Saint-Petersburg State University of Aerospace Instrumentation, Russia ³ Institute of Physiology, NAS of Belarus Acknowledgments: The collection and annotation of this dataset were carried out with financial support from the St. Petersburg Science Foundation and the Belarusian Republican Foundation for Fundamental Research (BRFFR, SCST) under the grant “Detection of tumor cells in nervous tissue using deep learning methods” (contract No. M24SPbG010).本数据集为用于细胞分割任务的、通过相差显微镜（phase-contrast microscopy）拍摄的C6胶质瘤细胞训练与测试图像。 示例展示了C6胶质瘤细胞的相差显微图像与人工标注的分割掩码。 数据类型：带COCO格式对应标注的相差显微图像。 显微成像类型：细胞接种后以24小时或72小时为间隔拍摄的二维相差显微图像。 显微镜设备：采用BestScope BS-2092型相差模式显微镜，搭载10×与20×物镜。 细胞类型：C6胶质瘤细胞（大鼠神经胶质肿瘤细胞，ATCC CCL-107）。 图像分辨率：2592 × 1944 像素²。 文件格式：.tif（8位）。 文件命名规则：文件名包含培养时长（24h或72h）与显微镜物镜倍率（10×或20×），例如spec_24h_10x_17.tif。 数据集子集： Glioma C6-spec：在严格控制的成像条件下拍摄的45张图像，划分为训练集（30张）、验证集（4张）与测试集（11张）。 Glioma C6-gen：在多变的成像与接种条件下拍摄的30张图像，用于测试模型泛化能力。 标注信息： 两个子集中共包含超20000个标注对象，其中12000个为细胞标注，7800个为胞体标注。 Glioma C6-spec的标注包含A型细胞（球形/梭形）、B型细胞（扁平/铺展型）与胞体（细胞核团）。 Glioma C6-gen子集的标注仅包含通用细胞实例，未进行细胞类型或细胞核的区分。 文献引用： "Glioma C6: A Novel Dataset for Training and Benchmarking Cell Segmentation", Malashin等人，2025年。 作者： Roman Malashin¹ ², Svetlana Pashkevich³, Daniil Ilyukhin¹, Arseniy Volkov³, Valeria Yachnaya¹ ², Andrey Denisov³, Maria Mikhalkova¹ 所属机构： ¹ 俄罗斯科学院巴甫洛夫生理学研究所 ² 俄罗斯圣彼得堡国立航空仪器制造大学 ³ 白俄罗斯国家科学院生理学研究所 致谢： 本数据集的采集与标注工作获得圣彼得堡科学基金会与白俄罗斯共和国基础研究基金（BRFFR, SCST）资助，项目主题为“基于深度学习方法检测神经组织中的肿瘤细胞”，合同号为M24SPbG010。