Datasets:
File size: 3,485 Bytes
c13030e 8bfedc2 c13030e 8bfedc2 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 | ---
license: cc-by-nc-sa-4.0
task_categories:
- image-segmentation
tags:
- medical
- histopathology
- h-and-e
- nuclei
- instance-segmentation
- pan-cancer
- tcga
- gtex
size_categories:
- 1K<n<10K
---
# PanNuke
Pan-Cancer H&E Nuclei Instance Segmentation and Classification dataset
(Gamper et al., ECDP 2019; arXiv:2003.10778). Mirrored from the
[Warwick TIA Centre release](https://warwick.ac.uk/fac/cross_fac/tia/data/pannuke).
## Composition
- **7,901** RGB patches of 256x256 at 40x magnification (~0.25 um/pixel)
- **19** tissue types pooled from TCGA / GTEx (Breast, Colon, Lung, Kidney,
Prostate, Stomach, Ovarian, Bladder, Esophagus, Pancreatic, Thyroid, Skin,
Cervix, Adrenal_gland, Bile-duct, Liver, HeadNeck, Testis, Uterus)
- **~189,744** annotated nuclei, multi-pathologist QC
- **3 folds** (PanNuke is designed for 3-fold cross-validation):
- `fold1`: 2,656 patches
- `fold2`: 2,523 patches
- `fold3`: 2,722 patches
## Schema
| Field | Type | Description |
|---------------|-------------------------------|-------------------------------------------------------|
| `image` | PIL RGB 256x256 | H&E patch, uint8 |
| `inst_map` | PIL grayscale uint16 256x256 | Global per-pixel instance ID (0 = background) |
| `type_map` | PIL grayscale uint8 256x256 | Semantic class per pixel (see table below) |
| `tissue` | int | Tissue ID 0-18 (alphabetical) |
| `tissue_name` | str | Tissue type name |
| `fold` | int | Source fold (1, 2, or 3) |
| `sample_id` | str | Unique ID, `foldN_NNNN` |
### Semantic class encoding (`type_map`)
| Value | Class |
|-------|--------------------------|
| 0 | Background |
| 1 | Neoplastic |
| 2 | Inflammatory |
| 3 | Connective / Soft tissue |
| 4 | Dead |
| 5 | Epithelial |
### Instance map (`inst_map`)
Each connected nucleus is assigned a unique 16-bit integer ID per patch
(starting from 1). IDs are derived from the original 6-channel mask
arrays released by Warwick: for each foreground class channel, instance
IDs are offset by the running count of nuclei in previous channels so
the result is globally unique within the patch.
## License
CC BY-NC-SA 4.0 (research / non-commercial use). Same license as the
original Warwick release.
## Citation
```bibtex
@article{gamper2020pannuke,
title={PanNuke Dataset Extension, Insights and Baselines},
author={Gamper, Jevgenij and Koohbanani, Navid Alemi and Graham, Simon and Jahanifar, Mostafa and Benet, Ksenija and Khurram, Syed Ali and Azam, Ayesha and Hewitt, Katherine and Rajpoot, Nasir},
journal={arXiv preprint arXiv:2003.10778},
year={2020}
}
@inproceedings{gamper2019pannuke,
title={Pannuke: An open pan-cancer histology dataset for nuclei instance segmentation and classification},
author={Gamper, Jevgenij and Koohbanani, Navid Alemi and Benet, Ksenija and Khuram, Ali and Rajpoot, Nasir},
booktitle={European Congress on Digital Pathology},
pages={11--19},
year={2019},
organization={Springer}
}
```
|