PUMA: melanoma H&E images with nucleus instances, ten nucleus classes and six tissue classes

Provider
Mark Schuiveling et al. (PUMA challenge); curated by SPORA
License
CC0-1.0 (source)
Size
9 files
Format
PNG, Parquet (in sampled files)
Stored at
Hugging Face (https://huggingface.co/datasets/csmp-hub/puma)
Access
Free and public. No account with the data provider is needed.
DOI
10.5281/zenodo.15050523
Category
Health & Biomedical
Source
https://doi.org/10.1093/gigascience/giaf011

About this dataset

PUMA (Panoptic segmentation of nUclei and tissue in MelAnoma) is the public training set of the PUMA challenge: 206 H&E regions of interest from melanoma, half from primary tumors and half from metastases, each 1024x1024 pixels. Nucleus instances carry ten classes, made with model assistance and corrected by hand, and tissue regions in six classes were hand-drawn and reviewed by dermatopathologists. puma.parquet holds one row per image with RGB image and masks as PNG bytes; the original GeoJSON nucleus and tissue annotation archives are included alongside. Patient IDs are not available, so image names alone cannot group images by patient, and the challenge's private test images are not included. About 430 MB (measured 2026-11-17). Images and masks are stored as PNG or TIFF bytes inside Parquet; read them with polars or pandas plus Pillow, or with Hugging Face datasets. Use for panoptic nucleus segmentation, TIL quantification and tissue segmentation in skin cancer. Curated and packaged on Hugging Face by the SPORA project (csmp-hub), which keeps the original license and citation files. The authors' Zenodo record 15050523 is released under CC0 1.0; cite Schuiveling et al., GigaScience 2025 (doi 10.1093/gigascience/giaf011).

How to load this dataset in Python

from datasets import load_dataset
ds = load_dataset("csmp-hub/puma")
print(ds)

Data structure

File types in sampled files: PNG, Parquet

Files

Example file: puma.parquet

Analyze this dataset without downloading it

On Scigantic, PUMA: melanoma H&E images with nucleus instances, ten nucleus classes and six tissue classes opens in a hosted JupyterLab notebook with the data already mounted, so you can read files where they are stored instead of copying them first. Each notebook includes an AI assistant that can see the mounted files and write and run analysis code. Open the dataset page to launch a notebook, or browse the full catalog.

This page is also available as Markdown and as JSON.

Related Health & Biomedical datasets