RationAI/PanNuke

PanNuke


Dataset Description

• —Homepage: PanNuke Dataset for Nuclei Instance Segmentation and Classification
• —Leaderboard: Panoptic Segmentation

Description

PanNuke is a semi-automatically generated dataset for nuclei instance segmentation and classification, providing comprehensive nuclei annotations across 19 tissue types and 5 distinct cell categories. The dataset includes a total of 189,744 labeled nuclei, each accompanied by an instance segmentation mask, and contains 7,901 images, each sized 256×256 pixels. The images were captured at x40 magnification with a resolution of 0.25 µm/pixel. The dataset is highly imbalanced, with the "Dead" nuclei category being particularly underrepresented.

Please note that the dataset was created by extracting patches from whole-slide images (WSIs). As a result, some nuclei located at the edges of patches may be cropped, with fewer than 10 visible pixels in certain cases.

Dataset Structure

The dataset is organized into three folds: fold1, fold2, and fold3, consistent with the original dataset structure. Each fold contains data in a tabular format with the following four columns:

• —`image`: The RGB tile of the sample.
• —`instances`: A list of nuclei instances. Each instance represents exactly one nucleus and is in binary format (1 - nucleus, 0 - background)
• —`categories`: An integer class label for each nucleus, corresponding to one of the following categories:
• —Neoplastic
• —Inflammatory
• —Connective
• —Dead
• —Epithelial
• —`tissue`: The integer tissue type from which the sample originates, belonging to one of these categories:
• —Adrenal Gland
• —Bile Duct
• —Bladder
• —Breast
• —Cervix
• —Colon
• —Esophagus
• —Head & Neck
• —Kidney
• —Liver
• —Lung
• —Ovarian
• —Pancreatic
• —Prostate
• —Skin
• —Stomach
• —Testis
• —Thyroid
• —Uterus

Citation

@inproceedings{gamper2019pannuke,
  title={PanNuke: an open pan-cancer histology dataset for nuclei instance segmentation and classification},
  author={Gamper, Jevgenij and Koohbanani, Navid Alemi and Benes, Ksenija and Khuram, Ali and Rajpoot, Nasir},
  booktitle={European Congress on Digital Pathology},
  pages={11--19},
  year={2019},
  organization={Springer}
}

@article{gamper2020pannuke,
  title={PanNuke Dataset Extension, Insights and Baselines},
  author={Gamper, Jevgenij and Koohbanani, Navid Alemi and Graham, Simon and Jahanifar, Mostafa and Khurram, Syed Ali and Azam, Ayesha and Hewitt, Katherine and Rajpoot, Nasir},
  journal={arXiv preprint arXiv:2003.10778},
  year={2020}
}