AI/ML Policy
This page explains NMPA's regulatory approach to artificial intelligence and machine learning in medical devices, including classification, clinical validation, algorithm transparency, and post-market monitoring requirements.
Regulatory basis
NMPA's digital health and SaMD framework is based on:
- MDSAR 2021 general device framework applied to software
- NMPA standalone software classification guidance (独立软件分类界定指导原则)
- NMPA SaMD registration technical review guidance (人工智能医疗器械注册审查指导原则, 2022)
- IEC 62304 as adopted in China's YY/T 0664 standard
Content on this site is written with AI assistance and is intended as a navigation aid only. Always verify against official NMPA sources before making regulatory decisions. Not affiliated with NMPA or any Chinese Government body. Not legal or regulatory advice.
Add overview of NMPA's AI/ML policy: (1) Standalone software (AI algorithms without hardware) classified under Medical Device definition if intended for diagnosis/treatment; (2) Must follow IEC 62304 and YY/T 0664 (Chinese equivalent) for software lifecycle and documentation; (3) SaMD registration requires algorithm transparency, training data documentation, performance validation on Chinese population data, and cybersecurity risk assessment; (4) Clinical evaluation must include real-world performance data; (5) Post-market surveillance mandatory with quarterly adverse event review for algorithm-related failures; (6) NMPA guidance (2022) emphasizes traceability and explainability for machine learning models used in clinical decision-making.