The hottest medical AI encounters three developmen

Medical AI encounters three development difficulties

original title: Medical AI encounters three development difficulties

through voice communication, robots can help patients with guidance; After reading the image data, the machine can issue a diagnostic report... With the progress of science and technology, artificial intelligence (AI) medicine has gradually changed from cutting-edge technology to practical application

however, the economic information daily learned that in the process of rapid development, China's medical artificial intelligence is facing three major development difficulties: technical problems need to be broken through, the entry threshold needs to be clarified by the regulators, and the business model also needs to be established. In this regard, experts suggest that the state should take the lead in database construction, break data barriers, realize medical data sharing, formulate standards as soon as possible, and promote the accumulation of high-end talents, so as to achieve a comprehensive breakthrough

the first is to alleviate the tension of medical human resources. Xia Huimin believes that in the context of the current lack of high-quality medical human resources in China, with the help of medical AI, on the one hand, patients at the grass-roots level and in remote areas can obtain the services of medical institutions and medical personnel in developed areas through remote AI medicine, so as to improve the efficiency of the use of medical human resources; On the other hand, with the help of artificial intelligence, the medical big data analysis of patients can optimize the medical service structure and process of the hospital

the second is to restructure the medical service mode, change "treatment" to "prevention", and change passive medical treatment to health services anytime and anywhere. Experts believe that AI can efficiently and accurately integrate medical test data, so that patients can have their own electronic health records and form health big data. Through the monitoring of intelligent mobile terminals and wearable devices, medical institutions and their medical staff can actively find individuals and people with abnormal health conditions, and give health risk tips, health improvement or medical measures in advance. Medical AI can also summarize the laws of disease prevention, diagnosis, treatment and rehabilitation from the perspective of both groups and individuals through the analysis, sorting and induction of intelligent tools

thirdly, it is to help the country formulate more scientific medical and health policies. Li Lanjuan, an academician of the Chinese Academy of engineering and vice president of the Chinese Medical Association, believes that AI simulates medical processes, medical diagnosis, medical advice and treatment plans through massive data, which will be a big change in medical health. Big data intelligent diagnosis and treatment technology is changing with each passing day, which will promote the formulation of public health policies more scientifically

insiders believe that China is expected to rely on these advantages to achieve "corner overtaking" in the field of medical AI. At the same time, as AI gradually changes from cutting-edge technology to practical applications, it may bring major changes to the current medical pattern

technology needs to be broken through

the interview of economic information daily found that at present, China's artificial intelligence medicine is still facing technical problems. It is understood that massive big data and computing power are essential elements for the development of artificial intelligence, especially in medical data sharing. At present, China urgently needs to make up for its shortcomings

"data island" phenomenon is not unified with data standards, making it difficult to share medical data. The accuracy of artificial intelligence needs to learn a lot of data. Experts believe that 1. Select the non cutting rectangular sample: the non cutting rectangular sample is the synthesis of tear initiation and tear expansion. Our country has great advantages in the number of hospital cases, but because the medical data are not shared, there is a "island" phenomenon, which is not conducive to the development of artificial intelligence technology

Hou Haotian, general manager of Zhiye Internet (Xiamen) Health Technology Co., Ltd., which customized software for more than 20000 medical institutions across the country, told that when hospitals use AI products, they need to connect with Internet companies. However, there are some problems in this docking process: the hospital system was relatively closed before, the electronic systems of different hospitals were constructed by different enterprises, and there are barriers between the systems of enterprises

Hou Haotian said that it is difficult for AI enterprises to integrate and study the data fed back by different customers' hospitals, which limits the feedback training of AI machines. How to open the hospital information reasonably and legally is still facing challenges

in addition, there is a lack of data entry, and the standard weight reduction value is relatively large. Yu Shihui, chief scientific officer of Guangzhou Jinyu Medical Laboratory Group Co., Ltd., introduced that taking the pathological Artificial Intelligence Aided Diagnosis institution as an example, the data source of enterprise training models is usually public data sets or scanned image data obtained by enterprises in cooperation with individual hospitals. Yu Shihui said, for example, artificial intelligence needs more than 10000 positive samples for the research and learning of membranous nephropathy. The professional team of a famous medical university in Guangdong has accumulated more than 2000 samples for many years. Although Jinyu medicine has more than 20000 samples, it is necessary to re label each sample if it wants to cooperate, so that the machine can deeply learn under the same disease classification standard. There are many kidney disease classification systems in our country, and the inconsistent standards lead to a large number of high-quality data that cannot serve the development of medical artificial intelligence

experts said that taking CT screening of pulmonary nodules as an example, the diagnostic accuracy of medical AI products in the industry for pulmonary nodules, sugar disease examination and other scenes is generally high, but enterprises usually have their own databases when training their own models, and their respective algorithms are trained according to their own data, and then use their own data to verify the accuracy

second, there is usually an invisible "hidden layer" between the input data of AI and its output answers, which is called "black box". The consequence of the existence of "black box" is that it is difficult to judge whether AI is wrong. "If doctors can see how computers think and draw conclusions, they can make humans believe in computers and feel more at ease about them." Said Zhang Kang, a professor at Guangzhou Women's and children's medical center

the path remains to be clarified

from the regulatory perspective, artificial intelligence has just been applied to the field of health care, some regulatory policies have yet to be clarified, talent accumulation is still insufficient, and a sustainable business model also needs to be established

first, the access policy is unclear. "Drugs and devices have very detailed regulations at the national regulatory level, but medical AI products are new products, and detailed standards are still being formulated." Liu Shiyuan, director of imaging medicine and nuclear medicine at Shanghai Changzheng Hospital, said that at present, nine medical AI products have been applied to the State Drug Administration for three types of devices, but none of them has been approved, and the standards and specifications used are still under discussion

according to the classification provisions in the classification catalogue of medical devices issued by the State Drug Administration in 2017, if the diagnostic software provides diagnostic suggestions through algorithms and only the auxiliary diagnostic function does not directly give diagnostic conclusions, it shall be applied for certification according to class II medical devices; If the lesion location is automatically identified and clear diagnostic tips are provided, clinical verification management must be carried out according to class III medical devices. According to insiders, at present, some state-owned enterprises have applied for class II certificates, but the products applying for class III instruments have not been certified

second, there is a large talent gap. According to industry statistics, at present, there are less than 50000 employees in China's artificial intelligence industry, and less than 2000 technicians trained through colleges and universities every year. Among the practitioners in the artificial intelligence industry, nearly 50% of them have more than 10 years of work experience in the United States, and less than 25% in China

not only that, Xia Huimin believes that the most important thing for AI to go from laboratory to clinic and better serve clinic is to find the pain points and urgent problems in medical treatment. At present, many medical AI teams are dominated by Algorithm Engineers, and there is a relative shortage of compound talents who understand both medicine and computer in China

third, a sustainable business model needs to be established. Yinzhiguo, general manager of Kingdee medical software use standard: piece Technology Co., Ltd., said that medical AI products are expected to make hospitals pay in the form of selling software to establish a sustainable business model, but at present, it is unrealistic to charge consumers directly. How to build a business model to form a business closed loop is still being explored in the industry. (Xiao Sisi, Hu Linguo)