TCGA WSI UNI2H Features: Tile-Level Embeddings for Pathology

Tile-level UNI2-h embeddings extracted from The Cancer Genome Atlas (TCGA) whole-slide images by W8Yi as of March 2026. The collection is organized by TCGA project codes and includes H5 feature files alongside visualization overlays for computational pathology research.

Use Cases

• Slide-level cancer classification using H5 tile-level embeddings
• Spatial analysis of tumor microenvironments using 'vis/' overlay verification
• Transfer learning for pathology tasks using pre-extracted UNI2-h features

Strengths

• Uses the UNI2-h foundation model for feature extraction
• Includes 'vis/' overlay images for pipeline verification
• Organized by standard TCGA project identifiers like TCGA-HNSC

Limitations

• Does not include raw whole-slide images
• Tile dimensions and magnification settings are not explicitly defined in the metadata

Provenance

Source

The Cancer Genome Atlas (TCGA)

Collection Method

Feature extraction via UNI2-h model pipeline

Freshness

Last updated March 2026.

Geography

US