TCGA-CRC-DX: Non-Normalized Tumor Tissue Image Tiles for Colorectal Cancer

Image tiles of tumor tissue from 604 colorectal cancer histological whole slide images sourced from The Cancer Genome Atlas (TCGA). The dataset was created by Jakob Nikolas Kather at RWTH Aachen University, with tumor tissue manually outlined and cut into non-overlapping 512x512 pixel tiles at 0.5 µm/px resolution. No color normalization or further preprocessing was applied, and tiles are compressed as JPEG files.

Use Cases

• Training machine learning models for tumor detection based on manually outlined tumor tissue regions.
• Developing image analysis algorithms for histopathology using non-normalized, high-resolution (0.5 µm/px) image tiles.
• Conducting correlative studies linking histological image features with genetic data available from cBioPortal.
• Benchmarking computer vision methods on a standardized set of colorectal cancer whole slide image tiles.

Strengths

• Contains 604 whole slide images, providing a substantial sample size for analysis.
• Tiles are high-resolution (0.5 µm/px) and non-overlapping, preserving spatial integrity.
• Data is sourced from the authoritative TCGA database and linked to available genetic data.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment for large-scale training.
• No color normalization was applied, which may introduce variability for model training.

Provenance

Source

The Cancer Genome Atlas (TCGA) database.

Collection Method

Tumor tissue was manually outlined and cut into tiles using QuPath v0.1.2 software.