BrcaSeg

BrcaSeg performs pixel-level classification of epithelial and stromal tissues in whole-slide hematoxylin and eosin (H&E) breast histopathology images to quantify tissue composition and enable correlation with The Cancer Genome Atlas (TCGA) breast cancer gene expression data.

Key Features:

• Convolutional Neural Network (CNN)-based classification: Uses a CNN to classify regions of interest in whole-slide H&E images.
• High classification accuracy: Achieves 91.02% classification accuracy distinguishing epithelial from stromal tissues.
• Pixel-level mapping: Generates detailed pixel-level maps delineating epithelial and stromal regions across histopathological images.
• Training on breast cancer tissue microarrays: Trained using well-annotated breast cancer tissue microarrays.
• Validation on TCGA images: Validated with images sourced from The Cancer Genome Atlas (TCGA) Program.
• Integration with genomic data: Pairs tissue maps with TCGA gene expression data to support correlation analyses and Gene Ontology (GO) enrichment.

Scientific Applications:

• Tissue Ratio Estimation: Estimates epithelial and stromal ratios within tumor microenvironments from H&E images.
• Correlation Analysis: Models relationships between tissue ratios and gene expression profiles from TCGA breast cancer data.
• Gene Ontology Enrichment Analyses: Identifies genes correlated with tissue ratios and performs GO enrichment to reveal common and subtype-specific biological processes across breast cancer subtypes.

Methodology:

BrcaSeg employs a convolutional neural network (CNN) to classify regions of interest in whole-slide H&E images; the CNN was trained on well-annotated breast cancer tissue microarrays and validated using images from The Cancer Genome Atlas (TCGA), producing pixel-level epithelial/stromal maps.