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 parquet 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… See the full description on the dataset page: https://huggingface.co/datasets/medarc/TCGA-12K-parquet-shuffled.
