Histological Images for MSI Classification in Gastrointestinal Cancer, TCGA Cohort

218,578 unique image patches derived from histological images of colorectal and gastric cancer patients in the TCGA cohort. The dataset was created by Jakob Nikolas Kather of RWTH Aachen University and includes preprocessed snap-frozen tissue slides. Images are split into training and test sets for MSS and MSIMUT tumor classifications.

Use Cases

• Train image classification models for MSS vs. MSIMUT status based on histological patches.
• Benchmark computer vision algorithms for cancer subtype detection using preprocessed TCGA images.
• Study the visual features associated with microsatellite instability in gastrointestinal cancer.
• Develop diagnostic support tools for pathologists using standardized, color-normalized image tiles.

Strengths

• Contains 218,578 unique image patches, providing a substantial volume of training data.
• Preprocessing includes tumor detection, resizing to 224x224 px, color normalization, and patient-level randomization.
• Training sets are equilibrated by undersampling, which may help address class imbalance.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Last update date is unknown; freshness unverified.

Provenance

Source

TCGA cohort (original whole slide SVS images from https://portal.gdc.cancer.gov/)

Collection Method

Derived from snap-frozen tissue slides, preprocessed with tumor detection, resizing, color normalization, and patient assignment.