Ultrasound AI and AI analysis software in China have four guidance documents released by the Center for Medical Device Evaluation (CMDE) under the National Medical Products Administration (NMPA) on July 10, 2023:
- Technical review points for imaging ultrasound AI software (process optimization function)
- Performance evaluation points for pathological image AI analysis software
- Clinical evaluation points for pathological image AI analysis software
- Performance evaluation points for hematological flow cytometry AI analysis software
In order to regulate the management of artificial intelligence medical devices and related products, the CMDE has issued these guidance documents as references for applicants to better prepare for China registration applications.
Technical review points for imaging ultrasound AI software (process optimization function)
The guidance document is applicable to the registration of ultrasound imaging artificial intelligence (AI) software with process optimization functionalities:
- The AI software can be integrated into ultrasound devices or exist as standalone software:
- Typically, if the product’s function is integrated into imaging and ultrasound equipment, the software falls under code 06-07 in the current medical device classification catalogue and managed as class II or class III in China.
- If the product exists as a standalone software, it falls under code 21-02, and classified as a class II device under the category of process optimization.
- The scope of the guideline covers AI functionalities such as standard view localization, physiological structure recognition, and automated measurements.
- The guideline aims to reduce repetitive tasks for medical professionals, improve efficiency, and enhance the standardization of ultrasound diagnostics.
- The guidelines do not cover AI functionalities related to assisting in diagnosing suspicious lesions, as they fall under a different category.
Scope of pathological image AI analysis software
Pathological image artificial intelligence analysis software refers to medical devices that use artificial intelligence technologies such as deep learning to perform functions like segmentation and detection on digital pathological images. These digital pathological images include images of cells or tissues under a microscope obtained through data imaging equipment and pathological images from Whole Slide Imaging (WSI) technology. The product is used in medical institutions and/or medical laboratories to assist pathologists in providing information for disease diagnosis, prognosis, treatment, etc. However, it cannot be the sole basis for clinical diagnostic decisions.
The software described in this guideline is intended for assisting decisions and can provide lesion feature recognition, nature of the lesions, and other information to pathologists. It falls under the management category of class III medical devices in the current medical device classification catalogue in China. The classification code is 21-04-02.
Performance evaluation points for pathological image AI analysis software
The applicant must provide research materials on the product’s performance and technical requirements, along with explanations of the research and compilation process, to establish the basis for determining functional and safety indicators related to the software. As part of the software research materials, the applicant should submit a self-testing report according to GB/T 25000.51. Alternatively, they can submit a self-inspection report or an inspection report in place of the self-testing report. The technical requirements for the product should be determined based on its own characteristics.
Clinical evaluation points for pathological image AI analysis software
The clinical evaluation of pathological image artificial intelligence analysis software should comply with the requirements of the “Guiding Principles for Medical Device Clinical Evaluation.” Several factors affect the performance of this software, including core algorithms, algorithm training, and other aspects. Even if two different analysis software use the same core algorithm, they may use different training datasets during the algorithm training process, leading to potential differences in annotations. Therefore, for confirming the clinical performance of such products, clinical trials should be conducted, and direct comparisons between similar products may not be suitable.
Performance evaluation points for hematological flow cytometry AI analysis software
Hematological flow cytometry artificial intelligence analysis software refers to software that assists doctors in analyzing structured data generated by multi-parameter flow cytometry technology/flow cytometers using artificial intelligence algorithms, such as data files or graphic files. It is used for auxiliary in vitro diagnosis of blood diseases, typically for qualitative or quantitative analysis, and can be a standalone software or a software component.
The software uses artificial intelligence algorithms to analyze structured data, and its applications include:
- Basic analysis: assisting flow cytometry personnel in automatic gating and analysis through a single artificial intelligence algorithm for visualizing flow cytometry data.
- Advanced analysis: using multiple artificial intelligence algorithms in combination (e.g., integrated learning algorithms and convolutional neural network algorithms) to achieve automatic gating assistance in blood disease diagnosis (benign/malignant, leukemia subtype classification, lymphoma typing, etc.), and monitoring minimal residual disease after chemotherapy.
The product is used in medical institutions and medical laboratories, and the results are provided as references for practicing physicians. As an auxiliary diagnostic tool for hematological flow cytometry, the diagnostic results suggested by the software cannot be the sole basis for clinical diagnostic decisions.
According to the current medical device classification catalogue in China, the classification code for standalone software is 21-04-02 (Computer-Aided Diagnosis/Analysis Software), and the classification code for software component is 22-01-08 (Flow Cytometry Analyzers)
The guidance document is primarily applicable to auxiliary decision-making artificial intelligence medical devices, which fall under Class III management. Other categories of products may follow their applicable criteria. The determination of artificial intelligence medical software classification can be referenced from the “Guiding Principles for the Classification of Artificial Intelligence Medical Software Products.”
Further information
Read the original CMDE announcement on the four newly released guidance documents for ultrasound AI and AI analysis software in China.
By adhering to the outlined requirements, manufacturers can ensure their products meet the necessary safety and performance standards. Also, Cisema can assist manufacturers in navigating the registration process effectively. Discover our services for medical device registration, renewals and NMPA Legal Agent.