TCGA-BRCA Histopathology Image Tiles from Two Cancer Centers

190 preprocessed histopathological image tiles derived from the TCGA-BRCA breast cancer dataset. The data includes 100 images from Walter Reed National Military Medical Center (A2 cohort) and 90 images from Roswell Park Comprehensive Cancer Center (E2 cohort), processed by Jakob Nikolas Kather at RWTH Aachen University. Images are 256x256 pixel JPEG tiles, color-normalized using the Macenko method, and accompanied by clinical and slide metadata tables.

Use Cases

• Training weakly supervised AI models for feature prediction based on unannotated image tiles.
• Testing model generalization across different tissue source sites based on the separate A2 and E2 cohorts.
• Developing transfer learning workflows for histopathology based on the provided clinical metadata.
• Benchmarking image analysis algorithms on standardized, color-normalized H&E-stained tissue samples.

Strengths

• Includes 190 preprocessed image tiles from two distinct medical centers.
• Provides clinical (CLINI) and slide (SLIDE) metadata tables for patient-level analysis.
• Images are standardized via the Aachen Protocol, including tessellation and color normalization.

Limitations

• Row and column counts for the metadata tables are unknown.
• The total number of unique patients represented in the 190 tiles is unspecified.
• Last update date is unknown; freshness unverified.

Provenance

Source

The Cancer Genome Atlas Breast Invasive Carcinoma (TCGA-BRCA) project.

Collection Method

Whole slide images were tessellated into tiles and color-normalized using the Macenko method as per the Aachen Protocol.

Geography

United States (Walter Reed National Military Medical Center and Roswell Park Comprehensive Cancer Center).