As soon as the medical artificial intelligence wind has started, the pace of supervision has gradually kept up.
According to the new version of the "Classification Catalogue of Medical Devices" (hereinafter referred to as the "New "Classification Catalog") issued by the State Food and Drug Administration in September 2017, it will be implemented on August 1, 2018, and there is a definition of diagnostic software , This means that medical AI imaging companies have "certified work" requirements, can no longer stay in the "free trial" stage of hospitals, and put forward new requirements for promoting the standardized development of the industry.
In April this year, the US FDA approved the first medical device IDx-DR that uses artificial intelligence to detect retinopathy in diabetic patients; on May 24, Imagen's OsteoDetect software was also approved by the FDA, which uses machine learning technology to analyze two-dimensional X-ray image, by identifying the front and back and side X-ray images of the patient’s wrist to determine whether the patient has a fracture.
"The domestic AI is developing much faster in medical applications, but there is indeed no product that has passed the CFDA regulatory approval. There is also a working group within the CFDA that is considering how to supervise and approve. There are no relevant rules yet." August 20, Boston Zhang Mingdong, Chief Medical Officer and Vice President of Regulatory Affairs of Science Greater China, and former medical officer of the U.S. Food and Drug Administration (FDA) Device Center, told the 21st Century Business Herald, "China is also learning how the United States regulates and considers the actual situation in China. Due to the speed of the start of traditional medical devices, the approval requirements between China and the United States are quite different, but for artificial intelligence devices, everyone’s start time is about the same, and the regulatory thinking should be very close. There are also international medical device regulatory alliances that are coordinating , I believe that the regulatory differences between China and the United States in this regard will not be too great, and the coordination will be higher."
Regulatory start
According to the new "Classification Catalog", the risk level of diagnostic software is determined based on the risk level, maturity, and publicity of the algorithm used, and not only based on the processing objects (such as images of diseases such as cancer and malignant tumors).
If the diagnostic software provides diagnostic suggestions through its algorithms, it only has auxiliary diagnostic functions and does not directly give diagnostic conclusions, the relevant products in the diagnostic catalog shall be managed in accordance with the second category of medical devices. If the diagnostic software automatically recognizes the diseased location through its algorithm and provides clear diagnostic prompts, its risk level is relatively high, and related products in the diagnostic catalog are managed in accordance with the third category of medical devices.
Ge Liang, the founder of Diannei Technology, told the 21st Century Business Herald that no company has obtained the third-class certificate, and another 11 companies have worked with relevant departments to formulate the certification standards for the third-class medical devices. The first one is expected to be approved in 2019. Companies with three types of certificates will be generated from these 11 companies. Diannei Technology, Tuma Shenwei, Tuxiang Technology, and Shenrui Medical have obtained Class II certificates.
"Due to the similar diseases studied, the high degree of product homogeneity, and the high degree of business concentration, within two years, about 80% of the hot pseudo-medical AI companies will disappear or shut down the AI ​​department. AI medical is changing from a feverish period to a calm In the future, this will allow companies with clearer business models to emerge.†Ge Liang believes that after the implementation of the new “Categoryâ€, “companies with free trials can continue to drive without a license, but the free era is about to pass and will only burn Companies with money cannot support it, the industry will enter an era of benign competition, and the public will gradually accept the notion that the use of artificial intelligence requires payment."
However, from the perspective of domestic supervision, it is not clear how to review and approve a medical AI product path, and no product has been officially approved.
Taking the experience of the FDA as an example, this year following the approval of the AI ​​product IDx-DR for diabetic retinopathy detection, the FDA approved an imaging AI product OsteoDetect software. Imagen submitted a study of 1,000 X-ray images to the FDA to evaluate the independent performance of OsteoDetect, an image analysis algorithm for wrist fracture detection, and its accuracy in identifying fractures, and compare the results of the algorithm with three Compare the judgment results of a professional orthopedic surgeon. At the same time, another retrospective study of 200 patients was submitted.
Wang Biquan, chief marketing officer of Baiyang Intelligent Technology, told the 21st Century Business Herald, “It is a good thing to have policies to regulate market behavior and prove that the product is one step closer to the listing and marketization. The image is now undergoing approval, especially looking forward to the country’s approval system, The pricing package can be improved as soon as possible."
Difficulties in dynamic evaluation
Due to the rapid iteration of AI technology, there is a huge gap with supervision.
"The difficulty of supervision and approval is related to the characteristics of artificial intelligence." Zhang Mingdong believes that the difficulty lies in that it cannot be like traditional devices. "Medical devices themselves are updated quickly and AI devices are updated faster. Therefore, the supervision of such medical devices cannot continue traditional supervision. The thinking needs to be adjusted. But the same concept is based on the basis of risk, that is, regardless of the type of medical device, the risk must be considered."
FDA Commissioner Scott Gottlieb said at the 2018 Health Datapalooza Conference in Washington that under the premise of protecting patients, the FDA is expanding the opportunities for digital medical tools, and is actively developing new regulatory frameworks and using new methods to review artificial intelligence.
The FDA expects that more and more applications will be submitted for review based on AI tools in the next few years, with medical imaging equipment being the first to bear the brunt. The FDA's attitude towards AI will also focus on its methods of processing real-world data, including structured and unstructured data from pathology slides, electronic medical records, wearable devices, and insurance claims data.
For this reason, the FDA has developed and adopted the Pre-Cert program for AI, which allows companies to make small changes to their equipment without having to submit an audit application every time. Moreover, the FDA will ensure that other aspects of the regulatory framework (such as new software verification tools) are sufficiently flexible to keep up with the unique attributes of this rapidly evolving field. The FDA's supervision of AI must ensure that these new technologies can meet their safety and effectiveness standards.
AI products also need to pass clinical trials to obtain approval, but the current domestic and foreign standards are different.
"Traditional devices are based on the supervision of the product itself, while the supervision of AI products is divided into two aspects, one is the risk of the product itself, and the other is the risk control of the manufacturer, which is the manufacturer." Zhang Mingdong explained that the FDA also has a pre-certification. To certify the manufacturer.
Airdoc founder and CEO Zhang Dalei pointed out, “From knowing it to being accepted, the corresponding payment system is complete, and medical insurance is involved, it takes a long time. However, the CFDA is paying more and more attention to supervision. In our submission and approval process, the CFDA It is also very active and professional, with many ideas, and a special artificial intelligence working group. Unlike the FDA’s new accelerated approval channel for artificial intelligence medical devices, the CFDA is currently more like to follow the approval of new drugs or new medical devices, and is more cautious. ."
Not all AI products require clinical trials. They must be determined based on the risks and uses of AI products. Zhang Mingdong explained that this is why the CFDA is needed to guide the market. “Domestic AI companies have different stages of development, and the key to supervision lies in products. The clinical medical value and role of the product, and then the detailed clinical evaluation process is based on the role. Domestic AI products are still in the initial stage of clinical evaluation, and many products are used directly without clinical evaluation. I am a little anxious. From this perspective From a point of view, the development of medical AI is still at a relatively early stage."
Dongguan Guancheng Precision Plastic Manufacturing Co., Ltd. , https://www.dpowerchargers.com