Multi-modal AI for comprehensive breast cancer prognostication

Area of Science:

• Oncology
• Digital Pathology
• Artificial Intelligence

Background:

• Breast cancer treatment selection relies on risk assessment using molecular subtypes and clinicopathological data.
• Current risk assessment methods lack the precision for optimal clinical decision-making.

Purpose of the Study:

• To develop and evaluate an AI test that integrates digital pathology with clinical data for improved breast cancer risk stratification.
• To compare the AI test's prognostic performance against the standard-of-care 21-gene assay.

Main Methods:

• Utilized data from 8161 breast cancer patients.
• Developed an AI test combining digital pathology images and clinical information.
• Evaluated the AI test's ability to predict disease-free interval and compared its discrimination to the 21-gene assay.

Main Results:

• The AI test demonstrated robust prediction of disease-free interval (C-index: 0.71).
• The AI test showed numerically higher discrimination (C-index: 0.67) compared to the 21-gene assay (C-index: 0.61).
• The AI test performed well across molecular subtypes, including triple-negative breast cancer (C-index: 0.71).

Conclusions:

• AI-based digital pathology tests offer a promising tool for enhanced risk stratification in breast cancer.
• The AI test has potential implications for improving clinical decision-making across all major breast cancer subtypes.
• The AI test provides robust prognostic performance, particularly in subtypes lacking current guideline-recommended assays.