FAIR-Path is the world’s first fairness-oriented pathology AI framework, integrating contrastive learning and weakly supervised multi-instance learning technologies to mitigate population bias at the model level. Validated across 20 cancer types and multiple international medical centers, it eliminated 88.5% of population bias (91.1% in external validation) while maintaining high diagnostic performance. This research is featured as the cover article in Cell Reports Medicine in Dec. 2025 issue.

This technology developed a fairness-centered pathology AI framework that separates disease features from demographic bias through fairness-oriented contrastive learning. Unlike traditional reweighting or oversampling methods, it suppresses bias at the model level. Across multiple cancers and cross-center validations, it reduced 88.5% of diagnostic unfairness while maintaining high accuracy, demonstrating internationally leading medical AI technology.

FAIR-Path allows AI to make equally reliable diagnoses for each patient from a single pathological slice. It can be seamlessly integrated into hospital PACS, digital pathology, and AI platforms, assisting healthcare institutions in establishing consistent verification processes across diverse populations. This enhances the fairness, credibility, and regulatory compliance of AI clinical deployments, showcasing strong potential for intelligent healthcare implementation.