Visual Intratumor Heterogeneity and Breast Tumor Progression

Area of Science:

• Computational pathology
• Breast cancer research
• Tumor heterogeneity analysis

Background:

• Intratumoral heterogeneity is a known prognostic indicator in cancer.
• Genomic heterogeneity does not always correlate with histologic (visual) heterogeneity.
• Understanding the source and impact of different types of heterogeneity is crucial for accurate prognostication.

Purpose of the Study:

• To develop a predictor for histologic heterogeneity in breast tumors.
• To assess the association between visual heterogeneity and patient outcomes.
• To evaluate the relationship between visual heterogeneity and molecular heterogeneity.

Main Methods:

• A VGG16 image classifier was trained on 5907 core images from 1655 breast tumors (Carolina Breast Cancer Study).
• Patient-specific visual features were extracted and hierarchically clustered to define visual heterogeneity.
• Generalized linear models and Cox models were used to assess associations with clinical features, molecular heterogeneity, and tumor recurrence.

Main Results:

• Low visual heterogeneity was more common in basal-like and ER-negative tumors, and in tumors from younger and Black women.
• Tumors with lower visual heterogeneity exhibited a significantly higher risk of recurrence (HR=1.62).
• Low heterogeneity was associated with single-subclone tumors and TP53 mutations, with consistent findings across epithelial, stromal, and combined image analyses.

Conclusions:

• Histologic heterogeneity provides complementary prognostic information beyond molecular indicators.
• Low visual heterogeneity is a predictor of worse outcomes in breast cancer.
• Integrating multiple sources of heterogeneity is essential for a comprehensive understanding of tumor progression.