TCGA-UniformTumor-8K: 25,495 Expert-Annotated Pan-Cancer Histopathology ROIs

TCGA-UniformTumor-8K contains 25,495 high-resolution regions of interest (ROIs) extracted from 9,662 H&E-stained diagnostic histopathology slides. Created by MahmoodLab using TCGA data and updated in 2025, the dataset features manual annotations from two expert pathologists for pan-cancer subtyping tasks.

Use Cases

• Pan-cancer subtyping using the 25,495 expert-annotated ROIs
• Training high-resolution feature extractors on 8,192 x 8,192 pixel histopathology images
• Benchmarking tumor detection algorithms against pathologist-verified ground truth

Strengths

• 25,495 high-resolution ROIs
• Manual annotations by two expert pathologists
• Derived from 9,662 diagnostic whole slide images

Limitations

• Limited to H&E-stained FFPE diagnostic slides
• Excludes slides with poor staining or focus, potentially introducing selection bias

Provenance

Source

The Cancer Genome Atlas (TCGA) and MahmoodLab

Collection Method

Manual annotation and extraction from whole slide images by expert pathologists

Freshness

Last updated November 2025.

Geography

United States