Understanding AI risks from its characteristics and NMPA regulation perspectives

Area of Science:

• * Medical technology and regulatory science.
• * Artificial intelligence in healthcare.
• * Clinical translation of AI innovations.

Background:

• * Rapid advancements in AI are transforming medical research and healthcare delivery.
• * A significant gap exists between AI innovation and the approval of AI medical devices (AIMDs).
• * Conventional regulatory approaches struggle with the unique characteristics and risks of AI.

Purpose of the Study:

• * To explore the role of regulatory frameworks in bridging the translational gap for AIMDs.
• * To interpret China's National Medical Products Administration (NMPA) full-lifecycle supervision model for AI.
• * To propose a comprehensive framework for evaluating AIMDs.

Main Methods:

• * Systematic mapping of AI characteristics to regulatory control measures.
• * Analysis of the patient-centered AI ecosystem (academia, industry, regulators).
• * Development of a structured checklist for AIMD assessment.

Main Results:

• * Identified intrinsic AI characteristics complicating regulatory evaluation and posing safety risks.
• * Established a point-to-point correspondence between AI traits and regulatory controls.
• * Proposed an evaluation framework extending beyond performance to include development processes and non-functional attributes (safety, usability, explainability).

Conclusions:

• * Effective AIMD evaluation requires a lifecycle approach encompassing development and non-functional aspects.
• * The proposed framework enhances regulatory clarity and promotes safe deployment of AIMDs.
• * This approach fosters patient trust and optimizes outcomes in AI-powered healthcare.