Pathology Benchmark · MahmoodLab
TCGA-UniformTumor-8K
A pan-cancer pathology benchmark of 25,495 high-resolution image regions for evaluating slide encoders.
What TCGA-UniformTumor-8K is
TCGA-UniformTumor-8K is a computational pathology benchmark developed by MahmoodLab for region-level pan-cancer subtyping. It provides 25,495 standardised regions of interest extracted from whole-slide images within The Cancer Genome Atlas (TCGA).
The dataset addresses the need for standardised targets when testing high-resolution slide encoders. By providing large uniform tumour tiles measuring 8,192 × 8,192 pixels, it allows researchers in medicine and biology to assess how well computational vision models represent tumour morphology across diverse cancer types.
How it was made
The dataset is curated by tiling 25,495 regions of interest from TCGA whole-slide images at a fixed resolution of 8,192 × 8,192 pixels, focusing on uniform tumour regions suited for evaluation pipelines.
Key facts
- Dataset size
- 25,495 regions of interest
- Image dimensions
- 8,192 × 8,192 pixels
- Source data
- The Cancer Genome Atlas (TCGA) whole-slide images
- Primary application
- Slide-encoder evaluation and pan-cancer subtyping
What you can use it for
- Slide encoder benchmarking
- Assessing the feature representation performance of high-resolution digital pathology slide encoders on standardised large-area image tiles.
- Pan-cancer subtyping
- Evaluating region-level classification algorithms across distinct tumour types derived from TCGA cohorts.
- High-resolution feature extraction
- Testing patch aggregation and embedding techniques across uniform 8,192 × 8,192 pixel pathology inputs.
Explore the data
This dataset is gated
Its authors require their terms to be accepted before the data can be read. Accept them on the original dataset page. The platform does not accept terms on your behalf, so the rows are not shown here.Strengths and limits
Good at
- Contains 25,495 high-resolution regions of interest for large-scale evaluation.
- Provides standardised 8,192 × 8,192 pixel tiles specifically prepared for testing modern slide-encoder architectures.
- Covers diverse tumour types through its pan-cancer TCGA foundation.
Watch out for
- Data consists of cropped regions of interest rather than entire whole-slide images.
- Derives solely from TCGA cohorts, inheriting any histological sampling boundaries present in the original repository.
Questions
Can I look inside TCGA-UniformTumor-8K without downloading it?
Not here: its authors gate it behind terms each user must accept on the original page. Its description and statistics are still shown.
What licence is TCGA-UniformTumor-8K under?
Its authors declare cc-by-nc-nd-4.0. Read the terms on the original page before reusing the data.
What are the dimensions of the images in this dataset?
Each region of interest measures 8,192 × 8,192 pixels.
What is the primary scientific purpose of TCGA-UniformTumor-8K?
It serves as a benchmark for region-level pan-cancer subtyping and for evaluating high-resolution slide encoders.
How many image regions are provided?
The dataset contains 25,495 tiled regions of interest.
Written from TCGA-UniformTumor-8K's own dataset card. Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.