Manual Tumor Annotations in TCGA Diagnostic Whole Slide Images

Manual annotations of invasive tumor tissue on diagnostic whole slide images from The Cancer Genome Atlas. Annotations were created by trained observers using QuPath software to enrich regions for molecular prediction studies, excluding non-tumor areas. The project merges COAD and READ tumor types into a single CRC category.

Use Cases

• Training segmentation models to identify invasive tumor regions based on manual annotations.
• Developing molecular prediction models using enriched tumor tissue regions.
• Comparing annotation strategies for efficiency in excluding non-tumor tissue and artifacts.
• Studying tumor morphology across major solid tumor types from TCGA.

Strengths

• Annotations were performed by trained observers using QuPath v0.1.2, indicating a standardized process.
• Focuses on enriching for invasive tumor tissue specifically for molecular prediction studies.
• Covers major solid tumor types from the TCGA resource.

Limitations

• The aim was not to create perfect pixel-level annotations, which may limit precision for some tasks.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count and dataset size are unknown, which may limit suitability assessment.

Provenance

Source

The Cancer Genome Atlas (TCGA) diagnostic whole slide images.

Collection Method

Manual annotations by trained observers using QuPath v0.1.2, converted to CSV format.