Deep water area of medical AI: some big problems still need to be solved

What is the prediction accuracy of your product in clinical use? You just said how much accuracy has been achieved and how to ensure the safety? What is the current generation of your products? What is the profit model? .. In addition to 5G, artificial intelligence is still a topic of great concern this year Recently, the author watched a medical artificial intelligence competition at a close distance There are more than a dozen medical projects on the scene After each road show, the judges and teachers will ask questions These are the most frequently asked questions Most of these judges and teachers are doctors and investors in the medical field Behind their sharp questions, they are also some of the most frequently faced problems in the application of artificial intelligence in the medical field: accuracy, security and business model So back to the core and commonplace question: trust For example, is the sensitivity, specificity and accuracy of artificial intelligence for disease recognition worthy of doctors' trust? If doctors use artificial intelligence to assist decision-making, should patients trust doctors? With the development of artificial intelligence technology, voice interaction, computer vision and cognitive computing, in-depth learning and other technologies are gradually mature Various applications in the field of artificial intelligence medical treatment have become possible, such as voice recording medical records, medical image intelligent recognition, auxiliary diagnosis and treatment, intelligent analysis of medical robots, personal health big data and AI medicine research and development But when it comes to practical applications, three years ago Geoffrey Hinton, the pioneer of artificial intelligence, swore, "we should stop training radiologists now Within five years, in-depth learning will be better than radiologists " Today, hundreds of emerging companies around the world are exploring the application of deep learning in radiology However, what is the number of radiologists replaced by AI? Radiologists are still in short supply all over the world Today's AI + medical, in addition to trust and commercialization, also faces challenges in policy, supervision and other aspects Recently, Tencent group legal affairs department released the "medical AI industry supervision and legal research" (hereinafter referred to as the "report") to discuss and analyze the legal and regulatory themes in the field of medical AI According to the report, medical treatment The legal and regulatory issues faced by AI include: whether the medical AI products need to be registered as medical devices when entering the market; how to use the health medical data and personal information legally in the process of research and development and operation of medical AI products; how to confirm the product liability and medical damage liability of medical AI products in the process of use; how to How to protect the intellectual property rights of medical AI products, how to face and avoid the medical ethical problems brought by medical AI products, etc 1 How to protect privacy? No matter in the development stage or application stage, medical AI will inevitably collect and use personal health information For example, some health management AI products will collect and use user health information, some cancer CT images or MRI photos will be used for medical AI model training How to realize the open sharing of data and explore the reasonable boundary of big data mining under the premise of ensuring the security of personal information and privacy has become a key problem for the development of medical AI "The current law stipulates that the collection and use of personal information should meet the principles of openness, transparency and minimum sufficiency, but the complexity of medical AI's use of information and the unknowability of black box test make the above principles difficult to meet Moreover, the review obligation boundary faced by the developers of medical AI to obtain data from the third party is not clearly defined by law "In the face of these difficulties, the report points out that the current international compliance practice is" de labeling " "Deidentification" refers to the process of making personal information unable to identify the main body of personal information without the aid of additional information through the technical processing of personal information The personal information processed by "deidentification" is called "deidentification information" The use of "de identified information" helps to reduce the risk of data security, and more conducive to the protection of personal information security However, according to the report, China's de labeling standards for health care data have not yet been unified in terms of specific implementation It is difficult for developers to judge the extent to which information is processed Some experts said that they hope relevant departments will issue the standards for demarcation of medical and health information as soon as possible, and appropriately relax the restrictions on the use of demarcated information, so as to promote the development of compliance of medical AI technology 2 Ethical issues of medical AI The future medical AI products may have human emotions, personalities and habits, participate in medical and health-related activities to varying degrees, and even replace doctors to engage in independent diagnosis and treatment At this time, whether and how to protect the ethical status and legal rights of medical AI will be the issues to be considered in the future The EU's discussion on the rights of artificial intelligent robots is at the forefront of the world, and it has been considering whether to give the legal personality of intelligent robot "E-man" with the right ability and responsibility for its behavior According to a European Parliament report in early 2017, "in the long run, create an independent legal status for robots, at least let the most sophisticated automated robots have the identity of an electronic person, make them responsible for possible damage, and may apply the identity of an electronic person to situations where robots make autonomous decisions or interact independently with third parties in other ways." However, the issue of granting artificial intelligence legal personality has aroused great controversy Shortly after the report's release, 156 AI experts from 14 European countries, including computer scientists, law professors and the company's chief executive officer, signed an open letter warning that it was "inappropriate" to grant legal personality to robots from a "legal and ethical perspective" Although limited by the development of technology, it is difficult to give a clear answer to these ethical dilemmas at this stage However, the report believes that with the continuous progress of science and technology, attention should be paid to such issues, and gradually explore and establish ethical norms and codes of conduct for medical professionals to make diagnostic decisions using AI, so as to adapt to the more extensive and in-depth application of AI in the medical field 3 Whether medical data is monopolized or shared is the cornerstone of the development of artificial intelligence The control and monopoly of data may directly affect the development of medical AI industry According to the report, from the perspective of data possession, data monopolization itself does not violate the anti monopoly law Only because monopolized data forms a dominant market position in a relevant commodity market, and enterprises abuse this dominant market position, or because monopolized data causes serious exclusion and restriction of competition, can they violate the anti-monopoly law At present, no special legislation has been made directly for data monopoly related to medical AI, but some countries have introduced policies to encourage the open sharing of data At present, China has made plans for the opening of government data and public enterprise and public institution data; the European Commission has put on the agenda the establishment of technical specifications for safe access and cross-border exchange of health data for research purposes; the United States has established the overall direction of combating data barriers and encouraging data sharing through the 21st century cure act According to the report, legislators can focus on the progress of medical data sharing in various countries, gradually establish a unified standard for medical data sharing, further promote the fair sharing of medical data, ensure the safe and legal use of data by medical AI developers, and promote the fair competition of medical AI developers 4 The cross-border transmission of medical big data must comply with the regulations In the aspect of cross-border transmission, international exchange and cooperation are indispensable elements for the technological progress of medical AI However, the report points out that from the perspective of national interests, health care big data has become an important national asset, so almost all countries have adopted a more cautious attitude towards exit control The measures for the administration of population health information (for Trial Implementation) issued by China on May 5, 2014 stipulates that population health information can only be stored in China, not in any form The national health and Health Commission further stipulated in Article 30 of the national health and medical big data standard, safety and service management measures (for Trial Implementation): "health and medical big data shall be stored on safe and reliable servers in China If it is necessary to provide it to overseas companies due to business needs, safety assessment and audit shall be conducted in accordance with relevant laws and regulations and relevant requirements." According to the report, how to continuously promote the flow of health care big data and the healthy development of related science and technology industries while protecting the interests of citizens and the state is a practical proposition worthy of further exploration in the future Summary: according to IDC's statistics, by 2025, the total value of AI application market will reach US $127 billion, of which the medical industry will account for one fifth of the market size, and AI + medical field is the general trend After decades of development, artificial intelligence has gradually penetrated into all aspects of medical treatment and entered the deep water area At the regulatory and policy level, China health information and health care big data society has recently set up a professional committee, Standards Committee and Working Committee for smart hospitals and artificial intelligence applications, which will formulate the industry standards for smart medical and artificial intelligence applications in China in the future So with the removal of technology, policy, supervision and other difficulties, the long-awaited subversive revolution will come Reference 1 Super frontier of this report: is medical AI desensitized or not? 2 Advancingaiinhealthcare: it's sallabouttrust, https://www.statnews.com/2019/10/23/advancing-ai-health-care-trust/penicillin This article is reprinted by yaozhi.com The copyright belongs to the original author The purpose of reprint is to transmit more information, which does not represent the view of this platform If the content of the work, copyright and other issues are involved, please contact our website message, we will delete the content in the first time.