Annotated necrosis and mitotic figures on 230 rat liver slides of the Open TG-GATEs dataset

<pre># Annotated necrosis and mitotic figures on 230 rat liver slides of the Open TG-GATEs dataset This dataset contains polygonal annotations of 230 whole slide images of the Open TG-GATEs dataset. The annotations were created by Bayer pathologists in a project collaboration between Bayer Pharmaceuticals and Aignostics GmbH. ## Dataset Files This dataset contains only text files which are referring to whole slide images of the Open TG-GATEs dataset. The images are not distributed with this dataset. Image URLs and their respective `md5` checksums are stored in `slides.json`. The annotations are available in the raw json format used in the collaboration as well as in converted [`geojson`](https://geojson.org/) format for easy import into other tools. | file | description | |:-----------------------------------|:-------------------------------------| | slides.json | URLs/md5-checksums of each slide | | annotations/raw/\*.svs.json | original format tissue annotations | | annotations/geojson/\*.svs.geojson | tissue annotations in geojson format | ## Annotated Lesions / Classes The dataset was annotated by 6 expert pathologists with the goal of labelling "Necrosis" and "Mitotic Figures" on rat liver whole slide images. Additional areas in the slides were labelled to improve model training. An overview of the annotated classes is given in the table below: | label | count | area (px^2) | |:---------------------------------------------------|------:|-------------:| | Necrosis (area) | 1576 | 386769004.0 | | Necrosis (single cell) | 1497 | 1624837.5 | | Mitosis | 5205 | 9322482.0 | | Single cell necrosis/apoptosis | 1531 | 1446001.0 | | Central vein | 1125 | 27947588.0 | | Connective tissue | 63 | 8478992.5 | | Dark hepatocytes | 763 | 2516812.5 | | Extramedullary hematopoiesis | 73 | 189508.0 | | Hemorrhage | 142 | 2231979.0 | | Inflammatory infiltrate | 1382 | 9859337.5 | | Large vessels | 224 | 597148962.0 | | Neoplasia | 176 | 478842.5 | | Oval cell/bile duct proliferation | 47 | 1621104.5 | | Parenchyma | 3249 | 673395978.0 | | Portal tract | 1067 | 40819943.0 | | Other lesions | 5 | 9251.0 | | Staining, contamination, and out-of-focus artifact | 46 | 4484697.0 | The pathologists labeled the following number of lesions: | pathologist | annotations | |:--------------|------------:| | pathologist-0 | 15660 | | pathologist-1 | 1749 | | pathologist-2 | 660 | | pathologist-3 | 79 | | pathologist-4 | 22 | | pathologist-5 | 1 | ## Reference The 230 slides of the Open TG-GATEs Pathological Image Database which the annotations refer to are part of the `Toxicogenomics Project and Toxicogenomics Informatics Project under CC Attribution-Share Alike 2.1 Japan` (`CC-BY-SA-2.1-JP`) and are available for download via [10.18908/lsdba.nbdc00954-02-000](http://doi.org/10.18908/lsdba.nbdc00954-02-000). ## License All `*.json` and `*.geojson` files in this distribution are licensed under the same `CC Attribution-Share Alike 2.1 Japan` (`CC-BY-SA-2.1-JP`) license as the Open TG-GATEs Pathological Image Database. The license text is included in the file `LICENSE` in the distribution archive. ## Acknowledgements We thank Dr. Maximilian Alber[^5], Olivér Atanaszov[^5] and Dr. Sharon Ruane[^5] for providing support, the annotation platform, the model training infrastructure and for the great collaboration. We also thank Dr. Yoshinobu Igarashi[^6] for providing insight and advice regarding the dataset and its licensing. [^5]: Aignostics GmbH [^6]: National Institutes of Biomedical Innovation, Health and Nutrition - Toxicogenomics Informatics Project </pre># Open TG-GATEs数据集230张大鼠肝脏玻片的坏死与有丝分裂象标注数据集 本数据集包含Open TG-GATEs数据集中230张全玻片图像的多边形标注。该标注由拜耳制药（Bayer Pharmaceuticals）与Aignostics GmbH合作项目中的拜耳病理学家完成。 ## 数据集文件 本数据集仅包含指向Open TG-GATEs数据集中全玻片图像的文本文件，原始图像并未随本数据集一同分发。每张玻片的图像URL及其对应的`md5`校验和存储于`slides.json`中。标注文件提供了合作项目中使用的原始JSON格式，以及可便捷导入其他工具的转换后[GeoJSON](https://geojson.org/)格式。 | 文件名 | 描述 | |:-----------------------------------|:-------------------------------------| | slides.json | 各玻片的URL与MD5校验和 | | annotations/raw/*.svs.json | 原始格式的组织标注 | | annotations/geojson/*.svs.geojson | GeoJSON格式的组织标注 | ## 标注病变/类别 本数据集由6名资深病理学家完成标注，目标为在大鼠肝脏全玻片图像上标记「坏死（Necrosis）」与「有丝分裂象（Mitotic Figures）」。此外还对玻片内其他区域进行了标注，以优化模型训练效果。标注类别的详细信息如下表所示： | 标注类别 | 数量 | 面积（像素²） | |:---------------------------------------------------|------:|-------------:| | 坏死（区域型） | 1576 | 386769004.0 | | 坏死（单细胞型） | 1497 | 1624837.5 | | 有丝分裂 | 5205 | 9322482.0 | | 单细胞坏死/凋亡 | 1531 | 1446001.0 | | 中央静脉 | 1125 | 27947588.0 | | 结缔组织 | 63 | 8478992.5 | | 深色肝细胞 | 763 | 2516812.5 | | 髓外造血 | 73 | 189508.0 | | 出血 | 142 | 2231979.0 | | 炎性浸润 | 1382 | 9859337.5 | | 大血管 | 224 | 597148962.0 | | 肿瘤性病变 | 176 | 478842.5 | | 卵圆细胞/胆管增生 | 47 | 1621104.5 | | 实质组织 | 3249 | 673395978.0 | | 门管区 | 1067 | 40819943.0 | | 其他病变 | 5 | 9251.0 | | 染色异常、污染与失焦伪影 | 46 | 4484697.0 | 病理学家标注的病变数量如下表所示： | 病理学家 | 标注总数 | |:--------------|------------:| | pathologist-0 | 15660 | | pathologist-1 | 1749 | | pathologist-2 | 660 | | pathologist-3 | 79 | | pathologist-4 | 22 | | pathologist-5 | 1 | ## 参考文献 本标注所依托的230张Open TG-GATEs病理图像数据库玻片，属于`日本CC署名-相同方式共享2.1协议`（`CC-BY-SA-2.1-JP`）下的**毒理基因组学项目与毒理基因组学信息学项目**，可通过[10.18908/lsdba.nbdc00954-02-000](http://doi.org/10.18908/lsdba.nbdc00954-02-000)下载获取。 ## 许可协议 本分发包内所有`*.json`与`*.geojson`文件，均采用与Open TG-GATEs病理图像数据库一致的`日本CC署名-相同方式共享2.1协议`（`CC-BY-SA-2.1-JP`）许可。许可文本包含于分发归档文件中的`LICENSE`文件内。 ## 致谢 感谢马克西米利安·阿尔伯（Maximilian Alber）博士[^5]、奥利弗·阿塔纳佐夫（Olivér Atanaszov）[^5]以及沙龙·鲁安（Sharon Ruane）博士[^5]提供的技术支持、标注平台、模型训练基础设施与卓越合作。同时感谢五十岚义信（Yoshinobu Igarashi）博士[^6]为数据集及其许可事宜提供的专业见解与建议。 [^5]: Aignostics GmbH [^6]: 日本生物医学创新、健康与营养研究所-毒理基因组学信息学项目