AI for Rare Breast Cancers Needs Specific Safeguards

Researchers at the University of Oxford propose a framework for safely implementing artificial intelligence in the treatment of rare breast cancers. They recognize these cancers, defined as occurring in fewer than 6 per 100,000 people, present unique challenges for AI development due to limited patient data and biological complexity.

The team outlines five key domains for clinical governance: defining AI’s intended use, validating performance with small cohorts, governing synthetic data, ensuring human oversight, and continuously monitoring the AI system. The framework emphasizes adapting general AI principles to address the specific needs of rare cancer research.

This includes rigorous data quality checks, preventing data leakage, and scoring the plausibility of synthetic data. While synthetic data can aid development, the researchers stress it should not replace validation using real-world clinical cases.

The proposed roadmap aims to enable precision oncology for rare breast cancers while prioritizing patient safety and responsible AI implementation. Further work will focus on refining these safeguards and applying them to specific AI applications.