In January, NMPA issued a guideline that aims to standardize the requirements for the registration of artificial intelligence medical. Qualtech summarizes several key points for manufacturers’ reference:
1. The Scope of Product
The guideline is applicable to medical devices employing AI-based clinical decision-making software with deep learning technology (including independent software, software components, hereinafter referred to as “Software”). - [--suggested revision!--] This software uses artificial intelligence (AI) algorithms to analyze series of medical images for indicators associated with a certain medical condition.
Applications utilizing deep learning technology for pre-processing (e.g., image quality improvement, imaging speed improvement, image reconstruction), process optimization (e.g., one-click operation), and conventional post-processing (e.g., image segmentation, data measurement) may take this guideline as a good reference.
2. The Concept of Supervision
The concept of this guideline focuses on the control of software data quality, the ability to generate valid algorithms, and relevant risks to clinical use. Clinical use risks shall consider the direct impact of data quality control and the algorithm validity, and, equally important, the indirect effects of invalid computational resources (such as the operational environments).
(1) The typical design and development process of such software involves user requirement analysis, data collection, algorithm design, verification and validation. While the two main guidance documents “The basic principles of software design and development” and “Appendix of Medical Device Good Manufacturing Practice for SaMD”, comprise good references for the basics of software development, the new guideline also presents some essential requirements.
(2) The risks in clinical use mainly include “the false negatives” and “the false positives”. Imported software shall also consider the impacts caused by the differences of ethnicity, epidemiology, clinical diagnosis and treatment practices. The clinical trial shall meet the requirements enclosed in the "the Regulation of Good Clinical Trial Practice". In establishing device software equivalence, same variety of products or clinical reference standards (e.g., gold standards) shall be selected as non-inferior control group.
(3) In addition to considering the software's built-in network security, the process control of the network and data security shall be factored in during the entire software lifecycle, including pre-marketing development and post-marketing phases. If the software is to be used with cloud computing services or mobile computing as terminal devices, it also needs to comply with the requirements set forth under “the Guidance of the Cyber Security Guidelines for Medical Device”.
3. The Requirement of Submission Dossiers for Registration
(1) Scope of application: The manufacturer shall clarify the intended use, intended operating environment and core functions of the software. Target population, target disease, clinical use, target end-users, environment, and contraindications (if applicable).
(2) Submission information: The manufacturers shall provide the relevant algorithm research data (including, but not limited to: data source compliance proof, algorithm performance influencing factor analysis, and other relevant information), result of different databases (such as test sets, public databases, the evaluation database, the retrospective study, and database for the clinical trials). Should there be any significant deviations in the databases, causes and effects shall be analyzed.
Moreover, it’s required to provide research data on network and data security process control. Sometimes, basic information (such as name, creator, data volume, data distribution) and usage of the public database shall be provided.
So far only US FDA and China NMPA has published guideline for deep learning-based medical device, hence, it would be worthwhile to incorporate both guidelines into the company’s regulatory strategy. The guideline issued by NMPA covers the requirement on most phases of medical device life time, including the design and development, clinical trial and product registration. We recommend the manufacturers to take this guideline as a reference even if the China is not the first priority market.
The first medical device with deep learning which has recently got its approval is a mobile software, where the applicant has chosen to seek NMPA’s support first for carrying out its clinical trial phase. For the manufacturers planning to launch their products in China, we suggest you to take a close watch for NMPA’s issuances related to AI products. Understanding the registration flow for AI devices is good for building up a feasible strategic plan appropriate for the regulatory framework of the product.
Qualtech offers comprehensive regulatory research service for all Asia countries. With our full assistance, your company shall gain confidence in us – taking the shorter registration route and getting approval in a short period of time. Feel free to contact our business department for more details.