Dataset Viewer
Auto-converted to Parquet Duplicate
chunks
list
config
dict
updated_at
string
[{"chunk_bytes":268440838,"chunk_size":9370,"dim":null,"filename":"chunk-0-0.bin"},{"chunk_bytes":26(...TRUNCATED)
{"chunk_bytes":268435456,"chunk_size":null,"compression":null,"data_format":["pickle","int","int","i(...TRUNCATED)
1761508812.56646
YAML Metadata Warning: empty or missing yaml metadata in repo card (https://huggingface.co/docs/hub/datasets-cards)

Attribution

This dataset contains 224 x 224 JPEG patches from whole-slide images originally downloaded from The Cancer Genome Atlas (TCGA) that are available in the NCI Genomic Data Commons (GDC) Open Access tier. We mirror and repackage a commonly used ~12k WSI subset in LitData format for ease of training. We exclude patches that did not pass HSV thresholding, following the procedure in Kaiko.AI's Midnight paper. Patches were randomly sampled across magnification levels. There are a total of 24,985,184 patches included.

Primary source

  • The Cancer Genome Atlas Program (TCGA), National Cancer Institute and National Human Genome Research Institute

Use and redistribution

  • These WSIs originate from the GDC Open Access tier. Redistribution of Open Access TCGA content is permitted.
  • Do not attempt to re-identify participants. Follow NIH Genomic Data Sharing policy and GDC Open Access terms.
  • Cite TCGA and the GDC in publications that use this dataset.

License note

  • No additional license from us. This mirror follows TCGA and GDC Open Access terms.
Downloads last month
2,022

Collection including medarc/TCGA-12K-litdata