The research on the safety and effectiveness evaluation of AI medical devices is being carried out in an orderly manner.

Artificial intelligence (AI) medical devices refer to medical devices that utilize AI technology, commonly applied in medical image processing, decision support, and other areas, typically appearing as standalone software or in the form of software components within medical device products. The scope of AI technology is broad, with representative technologies currently including deep learning, neural networks, and other new algorithms that train models using a data-driven approach.

The National Medical Products Administration's Medical Device Technical Review Center (hereinafter referred to as the Device Review Center) established an AI working group in 2017 to study the regulatory science of AI medical devices. Members of the working group come from relevant review departments of the Device Review Center, the innovative medical device review department, and the medical device registration and testing institutions of the national drug regulatory department, covering all aspects of pre-market regulation of AI medical devices. After its establishment, the working group conducted extensive industry research. First, it reviewed literature to understand and track foreign regulatory policies and product review and approval situations for AI medical devices; second, it engaged in exchanges with relevant enterprises, research institutions, and medical institutions to comprehensively understand the characteristics of AI technology, product features, industry development status, and clinical usage risks; third, it collected information on AI medical device manufacturers to gain in-depth insights into the development of the AI medical device industry, laying a solid foundation for subsequent regulatory science research on AI medical devices.

Currently, the AI medical device industry is thriving, and issues regarding the safety and effectiveness evaluation of products have become a global regulatory challenge. There is an urgent need to establish requirements and systems for product testing methods, cybersecurity testing methods, evaluation databases, and regulatory technologies. To this end, the Medical Device Registration Department of the National Medical Products Administration is leading a joint research effort with institutions such as the National Institute for Food and Drug Control, the China Academy of Information and Communications Technology, the Chinese Society of Biomedical Engineering, the Shanghai Shenkang Hospital Development Center, Peking Union Medical College Hospital, Zhejiang University, and Sichuan University, striving for breakthroughs in the safety and effectiveness evaluation of AI medical devices and producing a batch of new standards, methods, and tools.

"Research on the Safety and Effectiveness Evaluation of AI Medical Devices" is one of the first key research projects in drug regulatory science in China. The main tasks of the first phase include: formulating general review points; developing review points for three rapidly developing products, namely, diabetic retinopathy auxiliary diagnostic products, CT lung nodule auxiliary diagnostic products, and pathological image AI analysis software; and determining the conditions under which various in-progress databases can serve as evaluation databases to complete database quality research reports.

In terms of formulating general review points, the "Review Points for Deep Learning Assisted Decision-Making Medical Device Software" was published in July 2019. As the number of product applications increases and more review experience is accumulated, the review points are planned to be timely transformed into guiding principles.

Currently, other tasks are in the research and drafting stages, with completion expected next year. The review points for the aforementioned three products are intended to be refined and adjusted based on the characteristics of each product, building on the general review points. At present, there are two products in the diabetic retinopathy auxiliary diagnostic field that have entered the innovation channel and were approved for market launch in August this year; one product in the lung nodule auxiliary diagnostic field has also entered the innovation channel. Peking Union Medical College Hospital has established a diabetic retinopathy image database, and the Device Review Center has participated in some discussion activities, planning to use it as a case study to compile a database quality research report.

In addition, during the COVID-19 pandemic prevention and control, AI software for pneumonia based on lung CT has played a certain role. In February of this year, the Device Review Center quickly initiated related research and published the "Review Points for Pneumonia CT Imaging Triage and Assessment Software (Trial)" in March.

New AI medical device products continue to emerge and gradually find practical applications. The working group will continue to pay attention and incorporate more relevant content into future regulatory science research on AI medical devices, providing strong support for the healthy development of the AI medical device industry.