Characterization of Breast Cancer Intra-Tumor Heterogeneity Using Artificial Intelligence

Area of Science:

• Oncology
• Computational Pathology
• Medical Imaging

Background:

• Intra-tumor heterogeneity (ITH) is a key feature of breast cancer (BC), impacting disease progression, prognosis, and treatment efficacy.
• Accurate characterization of ITH in BC remains a significant challenge due to its inherent complexity.

Purpose of the Study:

• To utilize deep learning (DL) techniques for comprehensive evaluation of ITH in early-stage luminal breast cancer.
• To elucidate the impact of ITH on tumor behavior and patient outcomes using advanced computational methods.

Main Methods:

• A large cohort of 2561 early-stage luminal BC cases was analyzed using whole slide images (WSIs).
• Morphological features from tumor and stromal components were annotated, and a DL model was developed to quantify heterogeneity.
• An overall heterogeneity score was generated and correlated with clinicopathological features and patient outcomes.

Main Results:

• 162 morphological features were quantified, showing significant inter-feature correlations.
• High ITH was significantly associated with larger tumor size, poor differentiation, high proliferation, no special type (NST) tumors, and low estrogen receptor (ER) expression.
• A high overall heterogeneity score independently predicted poor patient outcomes and was linked to aggressive tumor behavior.

Conclusions:

• Deep learning models offer a powerful approach to accurately decipher the complexity of intra-tumor heterogeneity in breast cancer.
• DL-derived heterogeneity metrics provide valuable supplementary information for predicting patient outcomes in early-stage luminal BC.