Meta-Album PanNuke: 19 Human Tissue Types for Classification

The PanNuke dataset contains 5,530 images of 19 different human tissue types, transformed from a nuclei segmentation task into a tissue classification task. The images are 128x128 pixels and have undergone stain normalization to reduce bias. It was created by researchers including Jevgenij Gamper and adapted for the Meta-Album benchmark by Romain Mussard in March 2022.

Use Cases

• Training image classifiers to identify 19 different human tissue types based on microscopic images.
• Benchmarking few-shot learning algorithms on a standardized medical imaging dataset.
• Developing models for stain-invariant analysis of histopathology slides using the stain-normalized images.
• Studying feature representation for tissue classification tasks derived from segmentation data.

Strengths

• Contains 5,530 images across 19 distinct tissue classes, providing a multi-class classification challenge.
• Images are preprocessed to a consistent 128x128 pixel size and have undergone stain normalization, which may reduce technical bias.
• Clear lineage from the well-cited PanNuke dataset, with specific citations provided for both the original and extended versions.

Limitations

• Dataset was transformed from a segmentation to a classification task; the suitability of features for classification versus the original segmentation purpose is not detailed.
• The description does not specify the train/validation/test split or the distribution of images per class, which is critical for model evaluation.
• Column-level documentation and sample data are unavailable, requiring manual inspection after download to understand data structure.

Provenance

Source

PanNuke: An Open Pan-Cancer Histology Dataset for Nuclei Instance Segmentation and Classification (Gamper et al.).

Collection Method

Semi-automatically generated segmentation and classification of nuclei, later transformed into a tissue classification task for Meta-Album.

Freshness

Created for Meta-Album on 01 March 2022; update frequency is unknown.