The hottest Shanghai army technology helps the fro

  • Detail

With the help of "Shanghai army" science and technology, the front-line "anti epidemic" of artificial intelligence has entered a critical stage in the prevention and control of the COVID-19. In the process of seizing time with the epidemic, Shanghai's hard core scientific and technological forces are also racing against time, providing a strong guarantee for the improvement of the requirements, especially the dynamic accuracy

with the release of the sixth edition of the diagnosis and treatment plan of covid-19, CT images have become an important basis for decision-making, and the resulting pressure on CT film reading is also rising sharply. As the fastest sub field of AI healthcare, AI imaging has become a key force in epidemic prevention and control. Shanghai AI enterprises have launched ct+ai products to help front-line medical staff fight the epidemic

sentinel before the anti epidemic

with the rapid spread of the epidemic, the chest CT volume of key prevention and control units soared, and the average daily CT volume of some key hospitals once exceeded 1000 cases, several times higher than usual. How to improve the efficiency of film reading has become a major pain point

according to Dr. Ling of the Union Hospital Affiliated to Tongji Medical College of Huazhong University of science and technology, his hospital has 800 Xinguan beds. If he does CT three times a week, it is 2400 person times. Considering other examinations often done in clinic, the radiology department receives an average of 400 people a day, and there are hundreds of patients in each class in three shifts. While the workload is heavy, it is also necessary to ensure the accuracy of diagnosis of this new disease, This makes many front-line doctors in a state of high pressure and fatigue

the lung images of many mild patients are not typical. Some patients with basic lung diseases cover up each other, which increases the difficulty of diagnosis. Generally speaking, a group of patients with COVID-19 will produce more than 300 CT images. Even for senior experts, the time of image reading + diagnosis will take at least 10-15 minutes, most of which are repetitive mechanical labor, which is also time-consuming. Dr. Ling said

in addition to relieving the pressure of front-line doctors to read films, CT images also play the role of sentinel before the anti epidemic

Duanqi, head of research and development of lung AI intelligent analysis product of sensecare of Shangtang technology, told China business news that COVID-19 is highly infectious, and a large number of suspected patients need nucleic acid testing to be diagnosed. However, in reality, the lack of kits and testing ability is still a major problem, and the kit testing shows a high false negative in the early development of the disease, and clinical tests often need to be repeated many times to give real results

the first-line doctors found in clinical practice that in the early stage of COVID-19, the CT film showed a blurred ground glass shadow in which the two sides at the edge began to negotiate, and this early CT manifestation was often earlier than the appearance of clinical symptoms. However, this early ground glass shadow may be ignored in CT film reading, and it is not easy to quantify. Duanqi said

let AI images participate in the quantitative analysis of COVID-19 focus and efficacy evaluation, further improve the efficiency and accuracy of rating, and become a clear and specific demand of front-line doctors. In this regard, Shanghai AI companies have set up special teams and quickly completed the system launch and deployment

on January 28, ETO medical launched the first version of the covid-19 intelligent evaluation system, and within 24 hours, the system was deployed to four hospitals in Hubei with the most urgent epidemic prevention and control tasks. With this system, the time for quantitative evaluation of COVID-19 was reduced from a few hours to 2-3 seconds, greatly improving the efficiency of diagnosis and treatment. According to Shi Lei, vice president of ETO medical, the system has been applied online in dozens of medical institutions in more than a dozen provinces and cities across the country

with the cooperation of cooperative hospitals and medical institutions, Shangtang technology also collected and labeled thousands of pneumonia data in a short time, and quickly developed and completed a high-precision focus segmentation model of COVID-19. Based on the technical architecture design of the sensecare intelligent diagnosis and treatment platform, the application module upgrade combined with the intelligent analysis function for covid-19 was developed and completed in a short time, rushed to multiple hospitals, and synchronously opened the cloud version to rush to Hubei frontline in a remote way

race against time for research and development

it is not easy to carry out product research and development and landing in such a short time, and the personnel problem is the primary challenge

in the case of inconvenient transportation and personnel vacation, it is a certain challenge to complete the system online and deployment with quality and quantity. Shi Lei told China business. ETO Technology launched the project on New Year's Eve. At that time, many R & D personnel were scattered in many places in China, and some of them were in epidemic areas, even overseas. After the project was approved, more than 100 colleagues from the product, R & D and other relevant departments devoted themselves to their work. Each department set daily sprint goals, and the communication and feedback often lasted until the second half of the night

another challenge comes from the problem of sample size. Medical image is essentially an image recognition problem. AI reading requires a lot of annotation data to support machine training. However, in the early stage of the COVID-19, due to the small sample size of confirmed cases, medical institutions lack high-quality clinical diagnostic data. Even if there are a large number of suspected CT data of COVID-19 in the follow-up, they still face the problem of lack of effective labeling

in response to this problem, Shangtang science and technology urgently organized several experienced doctors in non core epidemic areas to manually label the data of hundreds of cases of covid-19 in one and a half days. Based on the preliminary model generated by this batch of data, using the self-developed semi-automatic labeling tool, we quickly completed the semi-automatic labeling of thousands of cases of pneumonia data, which were reviewed and modified by expert doctors in batches. Yitu technology gave full play to its reserves in small sample data, migration learning and other technologies, and quickly completed case collection, labeling, algorithm research and development, product design and other work

in the process of transporting the server to the hospital site for deployment, engineering colleagues also experienced a difficult test. Duanqi told China business. During the epidemic, the traffic was blocked, and the closure of highways in various regions changed greatly, which caused great difficulties to the logistics and transportation of servers. In the process of deploying the products to Qingdao West Coast people's Hospital, Shangtang technology also encountered heavy snow. The arrival time of the server was 3 days later than the original plan, and the delivery was even more distant

in order to enable the front-line to use the products as soon as possible, the technical engineer team from Shanghai rushed to Qingdao Shunfeng logistics distribution center in the evening after urgent discussion. One truck confirmed that it was close to midnight to find the server from tens of thousands of packages. Then the team directly pulled up the server and went to the hospital for on-site deployment. The deployment was completed at 3 a.m. the next day, He returned to the hospital at 8 a.m. to train doctors on the use process of the product

in late February, the resumption of work and production across the country gradually ushered in the return peak of the flow of people, resulting in a surge in demand for epidemic screening, screening of health personnel, suspected patients and physical examination. In addition, the epidemic trend in surrounding countries, especially Japan and South Korea, is not optimistic. Due to the special geographical location of Shanghai and Qingdao, there are many people and flights, which also poses greater challenges to the prevention and control work. Besides early screening, how to play the role of AI in treatment and healing has become a new goal

the road to industrialization in the future

at present, the landing and application of AI images has made rapid progress, which can assist front-line doctors in the early screening and quantitative analysis of pneumonia, but nucleic acid as the gold standard for diagnosis still plays an irreplaceable role. Duanqi emphasized

there are still some imperfections in the diagnosis of COVID-19 only by CT images. For example, there are still some early patients. Although nucleic acid detection is positive, there are no obvious signs in CT. Therefore, image AI must be comprehensively analyzed in combination with medical history, laboratory testing, etiology and so on

in fact, medical imaging is the fastest application field of AI in medical treatment. As early as 2017, Tongling City issued the "opinions of Tongling Municipal People's Government on establishing a working mechanism to promote the construction of copper based new material industry convergence and development base", which attracted a large number of companies and obtained about 4billion yuan of high-value financing. However, the industry has been facing problems such as difficult access to high-quality data, high cost of data annotation, unclear business model and so on. Many products have fallen into dust in hospitals and are facing the dilemma of c-round death

a sudden epidemic has put the accumulation of AI technology in the medical field into the most rigorous testing ground, greatly accelerated the landing of AI scenarios, and also provided new ideas and directions for the subsequent industrialization

According to Duan Qi, AI has also shown great potential in patient screening, screening of potential chemical small molecule drugs and biological macromolecular drugs, and epidemiological analysis. At first, many epidemiological studies have used AI to predict the spread rate, infection and death toll of the virus. Although the prediction results were affected by the available epidemiology at that time, it provided a good reference model and reference for further research in the future

among them, small molecule drug screening is an important direction. Based on the relationship between diseases, proteins, viruses and genes, AI is used to establish a model to screen out potentially usable drugs from listed drugs, including drugs against Ebola and AIDS. At present, many screened drugs have been used in clinic and achieved preliminary results

at present, our intelligent health team not only uses AI to screen the above chemical small molecule drugs, but also devotes itself to the research and development of biological macromolecular drugs. The team used AI to screen specific antibodies of cured patients to help doctors diagnose and treat. Duanqi said

in addition, he believes that the control of the epidemic is largely due to the strong measures taken by various regions to fight the epidemic, and the government hospital community has built a front line of strict prevention and control. Therefore, in addition to providing AI based intelligent assistance for front-line doctors, providing disease management and monitoring for cities and regions is also the key development direction in the future. For example, the joint regional health committee actively empowers the medical consortium to help it achieve hierarchical diagnosis and treatment

the epidemic gives different links of China's medical system an opportunity to work together. What it produces is not only clinical experience, but also a large amount of big data. With the application of AI and other innovative technologies, it can continue to optimize itself. Jinan Shijin microcomputer controlled steel wire rope special experimental machine, the managing partner of Qiming venture capital, has a perfect function of protective measures: hydraulic and electrical, some Liang yingyu said

in her view, China is expected to use first-hand data to develop a logistics cooperation model that is not limited to pharmaceutical development and virus decoding, but also includes civil and institutional cooperation, and then combine it into a new health and epidemic prevention industry to realize the dream of modernizing China's medical system and improving medical level in combination with science and technology

editor in charge: penghaibin

Copyright © 2011 JIN SHI