Alibaba’s AI medical “sword forging” four years: 170 hospitals launched, 57 cities


Editor’s note: This article comes from the WeChat public account “Machine Power” (ID: almosthuman2017), author Fan Xue, 36 krypton released with permission.The sudden outbreak is a big test for all walks of life, especially the AI ​​medical industry. From the outbreak consultation, drug research and development, viral gene analysis to clinical diagnosis and other links, it is urgent to enter the battle.Connect the dots into a line and apply “Flower Everywhere”.Behind this is the four-year dormancy of the medical AI team of the Dharma College. From CT imaging of the lung to medical AI, it has made good progress in liver nodules, cardiovascular and orthopedics.In the current epidemic, it is not just the Dharma Institute’s medical AI that “reverses the sword.” Both domestic Internet companies and medical-related companies are on the road of assistance, allowing the value of medical AI to be verified in the fight against epidemic, thereby speeding up the implementation of related products to a certain extent.To solve the current situation of revenue difficulties of medical AI companies in the past.As Hua Xiansheng, a senior researcher at the Bodhidharma Institute, said that AI assists, human doctors improve their efficiency, and this vision is just around the corner.Written by | Fan Xue Editor | In April and now, the epidemic is gradually spreading all over the world. After the storm of the domestic epidemic, the Ali and Bodhidharma Institute will always become key words that cannot be ignored whenever the anti-epidemic process is resumed.Indeed, whether it is a genetic testing platform that cooperates with the Zhejiang Center for Disease Control to reduce the genetic analysis time of suspected cases to half an hour; or the CT imaging system first applied in Zhengzhou “Xiaotangshan”; or it is used in 57 cities in ChinaThe intelligent epidemic robot, Alibaba’s anti-epidemic coverage covers epidemic consultation, drug research and development, viral gene analysis, clinical diagnosis and other links, the intensity is not small.The influence is still spreading. According to Ali statistics, as of March 31, the CT image AI of Dharma Institute has been implemented in nearly 170 hospitals in 16 provinces and cities including Zhejiang, Henan, Hubei, Shanghai, Guangdong, Jiangsu, and Anhui.340,000 clinical cases have been diagnosed.As the epidemic spreads globally, on March 31, the AI ​​of Bodhidharma Institute went out to sea with Alibaba Cloud-JBC, a well-known medical technology institution in Japan, officially launched Alibaba Cloud’s new coronary pneumonia AI diagnostic technology and began to provide this service to Japanese hospitals to help doctorsQuickly screen new coronary pneumonia by CT imaging.One Damo Academy’s “Battle of Swords” is retrograde. “The system interprets CT images of suspected patients with new crowns within 20 seconds and quantifies the severity of the disease. The accuracy of the analysis results reaches 96%.” Data from Hua Xiansheng’s mouth,The auxiliary function of this system is displayed intuitively.Hua Xiansheng is the vice president of Alibaba Group and a senior researcher of Dharma College. He graduated from the Department of Mathematics of Peking University in 2001. After working for 14 years at Microsoft, Wang Jian, a “father of Alibaba Cloud”, cut corners to open Ali’s vision development and landingExploration.On February 4, the Dharma Institute received the task to interpret the CT images of the suspected new crown with the help of AI technology. Within 4 days, the Dharma Institute algorithm team had constructed a recognition model. As soon as the model came out, the team was smallTest on the range dataset.However, products based on only a small range of data are not convincing enough. At that time, the epidemic was still in its infancy, and there was not enough clinical data in the world, and there was no standard data set.In this case, the team changed its mind and retrieved the original stock data-the “Intelligent Diagnosis of CT Symptoms of the Lungs” data in the Tianchi Contest.The cooperation between the hospital, Wanliyun, Changyuanjia and Guper Hospital quickly broke through the limitation of insufficient training data.It is understood that the Tianchi Big Data Competition is hosted by Alibaba and is a high-end algorithm competition for scientific researchers worldwide.By opening up massive data and distributed computing resources, participants have the opportunity to use their designed algorithms to solve social or business problems.Finally, based on the CT image sample data of more than 5,000 cases, the Dharma Institute learned and trained the lesion texture of the sample, processed retrospective data through NLP natural language, and trained the CT image recognition network using CNN convolutional neural network.On February 15th, the first convenient location for this system was Zhengzhou Qiboshan Hospital. As Zhengzhou “Xiaotangshan”, it was a designated treatment hospital for newly diagnosed patients with pneumonia. There were about 300 CT images of a patient.It takes about 5-15 minutes for the doctor to visually analyze the CT image of a case, which puts great pressure on the doctor’s clinical diagnosis.The CT image AI system of Dharma Hospital can directly calculate the proportion of the lesion site, quantify and predict the severity of the disease, greatly improve the diagnosis efficiency, and win valuable time for the treatment of patients.AI automatically recognizes CT chest radiographs of new coronary pneumonia cases. It is worth mentioning that, in addition to the CT image recognition algorithm, Dharma Institute and Alibaba Cloud have developed an auxiliary diagnostic algorithm that can be based on patient basic information, symptoms, and laboratory test results., Epidemiological history, imaging reports and other multi-dimensional information to further assist doctors in formulating scientific treatment plans.As of March 31, this AI system has been deployed in 170 hospitals in 16 provinces and cities including Zhejiang, Henan, Hubei, Shanghai, Guangdong, Jiangsu, and Anhui, and diagnosed more than 340,000 clinical cases.In addition to the CT imaging AI system, during the epidemic, the Dharma Hospital has many things to do, covering multiple links such as epidemic consultation, drug development, viral gene analysis, and clinical diagnosis.On January 27, the intelligent epidemic robot developed by Bodhidharma Institute overnight.After going online, it was put into use in all parts of the country. It quickly landed in 27 provinces, municipalities and autonomous regions across the country. It made 16 million free calls for prevention and control for 57 cities and reached more than 200,000 people with abnormal physical conditions.On February 1, the medical AI algorithm of Dharma College was officially applied to the etiology detection of new coronary pneumonia.Bodhidharma Institute collaborated with Zhejiang Provincial Center for Disease Control and Prevention to reduce the genetic analysis time of suspected cases to half an hour using an algorithm. This technology can avoid missed detection of nucleic acid detection and detect mutant viruses in a timely manner.”Our genetic analysis covers more than 95% of the entire sequence, with an accuracy rate of nearly 100%. From automated database construction to gene sequencing to the final step of genetic analysis (Dharma Institute only provides the last step of the service), the entire process takes only 14 hours”Other genome-wide sequencing methods generally take two to three days,” Hua Xiansheng said.Algorithm experts of Bodhidharma set up genetic testing and analysis parameters. In addition, during the outbreak, Bodhidharma Medical AI team also provided medical professional translation system, epidemic prediction and other systems to provide information support for a wider range and higher level of anti-epidemic.2. Four years of “sword forging” In 2016, Bodhidharma Medical AI was led by Hua Xiansheng, then vice president of Alibaba iDST.At that time, the artificial intelligence boom had just started. The discussion of artificial intelligence in the industry was only limited to technology research and development. Basically, no consideration was given to the landing scenarios. It is even unknown how specific the medical field can drive the landing of technology.However, Hua Xiansheng firmly believed that artificial intelligence entered the field of medical health is an inevitable thing. As an expert who has been cultivating for many years in the field of visual intelligence, the Hua Xiansheng team began to cut into the medical AI from CT images of the lungs and quickly made achievements.In July 2017, the AI ​​of the Bodhidharma Institute broke the world record in the international authoritative lung nodule detection competition LUNA16, winning the championship with an average recall rate of 89.7% (the proportion of nodules successfully found in the sample data).The contest requires players to analyze 888 lung CT samples for lung nodules.The sample contains a total of 1186 lung nodules, and more than 75% are small nodules less than 10 mm.In the end, the average recall rate of pulmonary nodules found by the Bodhidharma in seven different false alarm rates reached 89.7%, exceeding the second place by 0.2%.The recall rate under different false alarms this time “first” seems to have a “milestone” significance, which laid the foundation for the CT automatic diagnosis system of the new coronary pneumonia in Ali Dharma Hospital, and opened the wider research scope of Dharma University..In December 2018, the AI ​​of Dharma College stood out from nearly a hundred teams and won two firsts in the global LiTS (Liver Tumor Segmentation Challenge, CT image segmentation challenge of liver tumor lesion area); the heart coronary centerline extraction in 2019In the Rotterdam competition, Bodhidharma AI won the first place in the automatic extraction competition, and related papers were received in advance by the international top medical imaging conference MICCAI 2019; in 2019, Bodhidharma AI won the EMNLP 2019 microbial community information extraction competition (BBTask) “Relationship and entity joint extraction” task champion.It is precisely because of the “accumulation” of the basic research of the Bodhidharma in the past four years that it brought the application of technology from point to line to surface during the epidemic, allowing Ali to see the deep integration of bottom-level innovation and industrial soil, constantly changing.may.Alibaba’s healthy upgrade will occur in the context of the full-scale outbreak of business applications brought by these technical blessings.The sudden emergence of the three assisting doctors to improve the effectiveness of the epidemic will undoubtedly carry out a “big test” for all walks of life, especially the AI ​​medical industry. However, if you change the issue from another angle, this is actually an opportunity.It seems that the epidemic has greatly enhanced the awareness of consumers and hospitals on Internet medical treatment and imaging AI systems.Originally, medical AI products were not smooth in the actual implementation in hospitals. “Everyone in the medical AI industry has very little income. In the past, it was based on research and development, such as scientific research cooperation to earn income. Now products can also be placed in hospitals through tenders.”An employee of Ying Ying previously revealed to the heart of the machine.Tendering is a way for medical AI enterprise products to enter the hospital. The hospital releases the demand and bids for the same category of companies, but the process is not smooth. Whether the product or scientific research results are quickly landed, it is difficult to reach the market quickly. Therefore, channel resources and hospital contacts areIt seems very important, they can help the product “into” the hospital to a certain extent.However, even if it enters the hospital, how much it can be accepted, and the degree of application is unknown. According to the Beijing News “Finding Chinese Makers”, in the field of imaging where AI products are piled up, a top three hospital may install more than 10 AI companies at the same time.There are only one or two products that doctors really use, because AI cannot find the diagnosis after finding nodules, and the current products are mostly focused on lung nodule search, and homogenization is serious. However, these AI products that require deep learning are fewWith the doctor’s error correction and supplementation, the model iteration will also be slower.This point Ma Li was very touched. “I have communicated with many senior doctors before. Some doctors are more open to information technology, but some doctors are skeptical about AI.” “But after the outbreak, according to the doctors we have contacted,It was found that their awareness of AI medical treatment has been significantly improved, “Ma Li added.In addition to ordinary people and hospitals, the epidemic situation is also promoting the introduction of Internet medical related policies. On March 2, 2020, the National Medical Insurance Bureau and the National Health and Health Commission issued the “About the promotion of new Internet pneumonia epidemic prevention and control during the” Internet “medical insurance services”Guidelines”, it is clear that Internet medical institutions can issue electronic prescriptions online for insured persons, and adopt multiple ways to flexibly dispense medicines, and insured persons can enjoy medical insurance payment treatment; on March 5, “Internet medical” was included in the central government for the first timeMedical insurance documents.With the successive introduction of the “Internet medical insurance” policy, the vitality of Internet medical care has been further stimulated, and the patient’s concept of medical treatment is also constantly changing.The value of medical AI is verified in the fight against epidemics, and it will have a profound impact on the medical industry and the public.In the next few years, the digitalization and intelligence of the entire medical industry will be greatly improved, which will also become the normal state of the medical system: AI assists, human doctors improve efficiency, Hua Xiansheng added..

read more