The medical AI is so hot, and the supervision has followed suit step by step.
On November 19th, the Medical Device Technology Evaluation Center of the State Food and Drug Administration issued the Notice on Soliciting Information on Production Enterprises of Artificial Intelligence Medical Devices.
According to the above notice, in recent years, artificial intelligence technology has developed rapidly, and products combined with medical devices have begun to appear. For this reason, the evaluation center has started to carry out relevant research.
The term “artificial intelligence medical device” as mentioned in the notice refers to the medical device that adopts “new generation artificial intelligence technology” in the aspects of product workflow optimization, data processing and auxiliary diagnosis, among which “new generation artificial intelligence technology” refers to deep learning. The neural network is a technology that uses a data-driven training algorithm.
Production enterprises that wish to have the above-mentioned product development, production or registration declarations actively participate in the cooperation. Since there is no relevant product declaration registration at present, in order to facilitate contact and further do related work, the relevant production enterprise information is now collected for domestic and foreign enterprises.
The following companies are not within the scope of this information request: the product does not use the above-mentioned new generation of artificial intelligence technology; the new generation of artificial intelligence technology is used for product development or life cycle management, not the product itself; it is not a medical device.
Due to the rapid iteration of AI technology, there is a huge gap with regulation. In April of this year, the US FDA approved the first medical device IDx-DR that uses artificial intelligence to detect retinopathy of diabetic patients. On May 24, Imagen's OsteoDetect software was also approved by the FDA. The software uses machine learning technology to analyze two-dimensional. The X-ray image determines whether the patient has a fracture by recognizing the front and back and side X-ray images of the patient's wrist.
From the domestic regulatory level, how to review and approve a medical AI product path is not clear, and no products have been officially approved.
Previously, according to the new version of the "Medical Devices Classification Catalogue" issued by the State Food and Drug Administration in September 2017 (hereinafter referred to as "the new "Category"), the implementation of the diagnostic function software appeared on August 1, 2018. The definition of this means that the medical AI imaging company has the requirement of “holding the certificate” and can no longer stay in the “free trial” stage of the hospital, which puts forward new requirements for promoting the development of industrial standardization.
According to the new "Category", the risk level of the diagnostic function software is judged according to the degree of risk, maturity and openness of the algorithm, and is based not only on the target of the treatment (such as cancer, malignant tumors, etc.).
If the diagnostic software provides diagnostic advice through its algorithm, it only has the auxiliary diagnostic function and does not directly give the diagnosis conclusion. The relevant products in the diagnostic catalogue are managed according to the second type of medical device. If the diagnostic software automatically identifies the lesion through its algorithm and provides a clear diagnostic prompt, the risk level is relatively high, and the relevant products are managed according to the third type of medical device in the diagnostic catalog.
“Domestic AI has developed much faster in medical applications, but there is currently no product approved by CFDA. There is also a working group within CFDA that is considering how to supervise and approve. There are no relevant rules.” Boston Scientific Greater China Chief Zhang Mingdong, medical officer and vice president of regulatory affairs, and former medical director of the US Food and Drug Administration (FDA) equipment center, told the 21st Century Business Herald that "China is also learning how the US regulates and considers the actual situation in China. Traditional medical devices Due to the slow start-up, the difference in approval requirements between China and the United States is relatively large. However, in artificial intelligence devices, everyone starts at a similar time, and the regulatory ideas should be very close. There are also medical device regulatory alliances in the world to coordinate, I believe The regulatory differences in the United States will not be too great and the synergy will be higher."
Taking FDA's experience as an example, this year, after the approval of the AI product IDx-DR for diabetic retinopathy detection, the FDA approved an image AI product OsteoDetect software. Imagen submitted a study of 1,000 X-ray images to the FDA to evaluate the independent performance of OsteoDetect's image analysis algorithm for detecting wrist fractures and its accuracy in identifying fractures. The results of a professional orthopaedic surgeon are compared. A further retrospective study of 200 patients was also submitted.
“The difficulty of supervision and approval is related to the characteristics of artificial intelligence.” Zhang Mingdong believes that the difficulty lies in the fact that it cannot be like traditional equipment. “The medical device itself is updated fast, and the AI device is updated faster. Therefore, the supervision of such medical devices cannot continue the traditional supervision. The idea needs to be adjusted. But the same idea is based on risk, that is, regardless of the type of medical device, risk should be considered."
FDA Director Scott Gottlieb's 2018 Health Datapalooza conference in Washington said that while protecting patients, the FDA is expanding opportunities for digital healthcare tools and is actively developing new regulatory frameworks to review artificial intelligence in new ways.
The FDA expects more and more applications to 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 toward AI will also focus on its approach to real-world data, including structured and unstructured data from pathology slides, electronic medical records, wearables and insurance claims data.
To this end, 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 application for review each time. Moreover, the FDA will ensure that other aspects of the regulatory framework, such as new software verification tools, are flexible enough to keep up with the unique attributes of this rapidly evolving field. The FDA's regulation of AI ensures that these new technologies meet their safety and efficacy standards. The medical AI is so hot, and the supervision has followed suit step by step.
On November 19th, the Medical Device Technology Evaluation Center of the State Food and Drug Administration issued the Notice on Soliciting Information on Production Enterprises of Artificial Intelligence Medical Devices.
According to the above notice, in recent years, artificial intelligence technology has developed rapidly, and products combined with medical devices have begun to appear. For this reason, the evaluation center has started to carry out relevant research.
The term “artificial intelligence medical device” as mentioned in the notice refers to the medical device that adopts “new generation artificial intelligence technology” in the aspects of product workflow optimization, data processing and auxiliary diagnosis, among which “new generation artificial intelligence technology” refers to deep learning. The neural network is a technology that uses a data-driven training algorithm.
Production enterprises that wish to have the above-mentioned product development, production or registration declarations actively participate in the cooperation. Since there is no relevant product declaration registration at present, in order to facilitate contact and further do related work, the relevant production enterprise information is now collected for domestic and foreign enterprises.
The following companies are not within the scope of this information request: the product does not use the above-mentioned new generation of artificial intelligence technology; the new generation of artificial intelligence technology is used for product development or life cycle management, not the product itself; it is not a medical device.
Due to the rapid iteration of AI technology, there is a huge gap with regulation. In April of this year, the US FDA approved the first medical device IDx-DR that uses artificial intelligence to detect retinopathy of diabetic patients. On May 24, Imagen's OsteoDetect software was also approved by the FDA. The software uses machine learning technology to analyze two-dimensional. The X-ray image determines whether the patient has a fracture by recognizing the front and back and side X-ray images of the patient's wrist.
From the domestic regulatory level, how to review and approve a medical AI product path is not clear, and no products have been officially approved.
Previously, according to the new version of the "Medical Devices Classification Catalogue" issued by the State Food and Drug Administration in September 2017 (hereinafter referred to as "the new "Category"), the implementation of the diagnostic function software appeared on August 1, 2018. The definition of this means that the medical AI imaging company has the requirement of “holding the certificate” and can no longer stay in the “free trial” stage of the hospital, which puts forward new requirements for promoting the development of industrial standardization.
According to the new "Category", the risk level of the diagnostic function software is judged according to the degree of risk, maturity and openness of the algorithm, and is based not only on the target of the treatment (such as cancer, malignant tumors, etc.).
If the diagnostic software provides diagnostic advice through its algorithm, it only has the auxiliary diagnostic function and does not directly give the diagnosis conclusion. The relevant products in the diagnostic catalogue are managed according to the second type of medical device. If the diagnostic software automatically identifies the lesion through its algorithm and provides a clear diagnostic prompt, the risk level is relatively high, and the relevant products are managed according to the third type of medical device in the diagnostic catalog.
“Domestic AI has developed much faster in medical applications, but there is currently no product approved by CFDA. There is also a working group within CFDA that is considering how to supervise and approve. There are no relevant rules.” Boston Scientific Greater China Chief Zhang Mingdong, medical officer and vice president of regulatory affairs, and former medical director of the US Food and Drug Administration (FDA) equipment center, told the 21st Century Business Herald that "China is also learning how the US regulates and considers the actual situation in China. Traditional medical devices Due to the slow start-up, the difference in approval requirements between China and the United States is relatively large. However, in artificial intelligence devices, everyone starts at a similar time, and the regulatory ideas should be very close. There are also medical device regulatory alliances in the world to coordinate, I believe The regulatory differences in the United States will not be too great and the synergy will be higher."
Taking FDA's experience as an example, this year, after the approval of the AI product IDx-DR for diabetic retinopathy detection, the FDA approved an image AI product OsteoDetect software. Imagen submitted a study of 1,000 X-ray images to the FDA to evaluate the independent performance of OsteoDetect's image analysis algorithm for detecting wrist fractures and its accuracy in identifying fractures. The results of a professional orthopaedic surgeon are compared. A further retrospective study of 200 patients was also submitted.
“The difficulty of supervision and approval is related to the characteristics of artificial intelligence.” Zhang Mingdong believes that the difficulty lies in the fact that it cannot be like traditional equipment. “The medical device itself is updated fast, and the AI device is updated faster. Therefore, the supervision of such medical devices cannot continue the traditional supervision. The idea needs to be adjusted. But the same idea is based on risk, that is, regardless of the type of medical device, risk should be considered."
FDA Director Scott Gottlieb's 2018 Health Datapalooza conference in Washington said that while protecting patients, the FDA is expanding opportunities for digital healthcare tools and is actively developing new regulatory frameworks to review artificial intelligence in new ways.
The FDA expects more and more applications to 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 toward AI will also focus on its approach to real-world data, including structured and unstructured data from pathology slides, electronic medical records, wearables and insurance claims data.
To this end, 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 application for review each time. Moreover, the FDA will ensure that other aspects of the regulatory framework, such as new software verification tools, are flexible enough to keep up with the unique attributes of this rapidly evolving field. The FDA's regulation of AI ensures that these new technologies meet their safety and efficacy standards.