The history of medical artificial intelligence can be traced back to the 1950s.
From the birth of the prototype of computer-aided diagnosis in 1959, to the 1980s, the medical expert system (Medical Expert System) made a breakthrough, and now, “Clinical Decision Support System” (Clinical Decision Support System, referred to as CDSS) has become a Hospitals at home and abroad are equipped with standard products, and the auxiliary role of artificial intelligence in the medical field is becoming more and more prominent.
Taking the clinical decision support system as an example, this system can simulate the thinking and diagnostic reasoning of doctors by “learning” the medical knowledge of experts and doctors, and provide diagnosis and treatment plans, thereby improving the efficiency of doctors’ diagnosis and reducing the problems caused by improper medication or improper operation. Medical malpractice, but there are still shortcomings: the decision support system is separated from the Electronic medical record system, and the data synchronization relies on manual labor, which takes a long time, which leads to the lag in the auxiliary diagnosis and treatment decision-making suggestions provided, and the ability to provide personalized medication guidance is also very weak.
Although the intervention of medical artificial intelligence has brought a new turn of events, the progress of personalized medicine in clinical decision support systems is still slow. On the one hand, the lack of diagnosis and treatment data in the current hospital outpatient medical records hinders machine deep learning; on the other hand, even if the data is complete, the diagnosis and treatment levels of different doctors in the same hospital are uneven, which also allows the clinical decision support system to personalize medicines. becomes difficult.
In June this year, Ping An Smart City? Smart Healthcare (hereinafter referred to as “Ping An Smart Healthcare”) proposes a solution that effectively combines clinical guidelines and clinical practice cases, and applies it to its intelligent assistant diagnosis and treatment system AskBob, which can assist doctors in personalized diagnosis and treatment for 1,500 diseases. Medication recommendation.
In November, Ping An Smart Healthcare and Sanofi, a well-known pharmaceutical company, successively released a diabetes AI assistant and a diabetes clinical decision support system. In addition, Ping An Smart Healthcare also cooperates with SingHealth, one of the largest public healthcare groups in Singapore, to provide doctors with personalized treatment plans for patients with type 2 diabetes.
Note: In November this year, the diabetes clinical decision support system jointly built by Ping An Smart Healthcare and Sanofi was exhibited at the 2nd China International Import Expo.
Through in-depth cooperation with authoritative institutions in different professional fields, AskBob, the intelligent auxiliary diagnosis and treatment system of Ping An Smart Healthcare, has done deep enough and comprehensive enough in the field of type 2 diabetes. It allows doctors to directly view the specific content of each drug recommendation, including drug category, generic name, trade name, knowledge evidence, etc., which further improves the personalized treatment of type 2 diabetes.
Why Ping An Smart Healthcare? Against the background of many players in the field of medical artificial intelligence, what is the story of Ping An Smart Healthcare?
Talking about Ping An Smart Healthcare, it is most appropriate to call it a “breakthrough” in the field of smart healthcare.
When Ping An Smart Healthcare entered the game, the domestic and foreign medical artificial intelligence track has been hot for a year. In February 2016, Google announced the establishment of the DeepMind Health department to help the British National Health System in the construction of auxiliary decision-making and further explore medical artificial intelligence; then IBM, Facebook, Amazon and other giants spontaneously gathered to research and promote artificial intelligence. Foreign medical artificial intelligence has officially entered a golden period of development.
Looking back at home again, during this period, AI + medical vertical start-ups have been moving, and Ping An, Ali, Tencent, and Baidu are not far behind, and the “AI medical war” has entered a white-hot stage.
According to the “2018 Blue Book on the Development of the World Artificial Intelligence Industry”, as of the first half of 2018, a total of 4,998 artificial intelligence companies have been monitored worldwide. Among them, the number of China reached 1040. Among the fields penetrated by artificial intelligence in China, the medical and health field accounts for the largest share, reaching 22%.
When the players were fiercely competing, Ping An Smart Medical revealed their achievements: 730 days, building the first AI real-time influenza prediction model in China, and helping government departments in many cities across the country to make influenza predictions; smart medical imaging technology deployed in more than 3,000 hospitals , the number of intelligent imaging service calls is close to 100 million; the diagnosis and treatment recommendation model of the intelligent auxiliary diagnosis and treatment system AskBob has covered 1,500 diseases and has been successfully implemented in more than 14,000 primary medical institutions.
Talking about the strategy when entering the market, Xie Guotong, chief medical scientist of Ping An Group, said that they chose a “sparsely populated” road at that time – starting from the whole process of disease diagnosis and treatment, launched intelligent disease prediction, intelligent medical image screening, intelligent clinical A full-process solution matrix such as decision support system, intelligent chronic disease management, and intelligent medical quality control.
Note: Xie Guotong, chief medical scientist of Ping An Group.
At that time, most of the peer players only focused on one of the links, or intelligent medical imaging screening, or intelligent clinical decision support system. In fact, this kind of single link is a drop in the bucket for solving the complex diseases of patients. Only a composite model that covers multiple links and is based on one examination site and conforms to multi-organ problems is the development direction of medical artificial intelligence products.
When determining the product strategy, Xie Guotong said frankly that Ping An Smart Healthcare made full use of Ping An Group’s accumulation in capital, technology, scenarios, medical claims and other aspects. Taking technology as an example, it is understood that Ping An will continue to invest 1% of its trillion-dollar revenue in research and development every year. This year, it is expected to have 10 billion in funding.
With its unique positioning and relying on the accumulation of Ping An Group’s own resources, Ping An Smart Healthcare seems to have been pressed the “accelerator button”, and the whole process products are rapidly updated and iterated.
Technology is the foundation of the field of artificial intelligence. Kai-fu Lee mentioned in his book “Artificial Intelligence” that the fundamental reason why computers can increase human cognitive ability and intelligence lies in the advancement of deep learning.
The same is true in the field of medical artificial intelligence. Smart medical products are inseparable from the deep learning of machines, and deep learning needs to label a large amount of data and do training on tens of thousands of images, so that a correct diagnosis can be made. Therefore, data, computing power and algorithms are the three basic elements for building products.
At present, there are many problems in the quality of domestic medical data. The diagnosis and treatment data in the outpatient medical records of many hospitals are incomplete and untrue, such as irregular diagnosis behaviors, and errors in information generation and collection. If there is a problem with the data source, it is impossible to talk about the later disease prediction and diagnosis accuracy.
In this regard, Xie Guotong said that in order to ensure data quality, Ping An Smart Medical chooses to cooperate with top-ranking hospitals at home and abroad for product polishing. For example, in the cardiovascular field, it has cooperated with Fuwai Hospital and Beijing Anzhen Hospital; In terms of ophthalmology, it has also cooperated deeply with Beijing Tongren Hospital and Shanghai Ophthalmology Hospital.
On the basis that the data quality is guaranteed as much as possible, the algorithm computing power has become another core factor to test the success of the rapid polishing of the product. At present, domestic companies including Baidu have already tested the waters in this field, and Ping An Smart Healthcare also has two “housekeeping treasures” – Saifei AI algorithm platform and AskBob text understanding technology.
It is understood that Saifei AI algorithm platform can provide data scientists with a series of tools such as self-developed deep learning framework SFE, intelligent annotation, scarce sample generation, modeling algorithm suite, distributed acceleration, model compression and high-performance inference, so as to improve Model training speed and quality.
“Take intelligent labeling as an example, labeling is the first step in deep learning. Labeling data is actually a very painful thing for doctors, and it consumes a lot of manpower and energy, and the intelligence of our Saifei AI algorithm platform Labeling quickly solves the problem.” Xie Guotong said.
Legend: Recently, the core technology paper of Saifei platform was accepted by AAAI-20, the top international academic conference on artificial intelligence.
It is understood that Ping An Smart Healthcare has cooperated with a number of authoritative national clinical medical research centers to label case images. Taking the kidney pathology image analysis in cooperation with the National Kidney Disease Clinical Research Center as an example, the use of Saifei AI algorithm platform can reduce the number of doctors’ annotations by half, but does not reduce the recognition accuracy of the final AI model.
In addition to the Saifei AI algorithm platform, the AskBob text understanding technology is another core technology of Ping An Smart Healthcare. This technology incorporates Ping An Smart Medical’s unique text understanding technology that integrates language models and knowledge graphs, enabling machines to understand diseases through the entire medical process and better help doctors accurately diagnose diseases. At present, this technology is mainly used in AskBob, an intelligent auxiliary diagnosis and treatment system for products.
In August this year, in the COIN 2019 text understanding competition held by EMNLP, the top international natural language processing conference, the AskBob text understanding technology successfully won the world championship in the overall score of the competition with the support of Saifei AI algorithm platform, and the multiple-choice text understanding ( All the individual champions of the two subtasks (accuracy 90.6%) and cloze text comprehension (accuracy 83.7%).
It is undeniable that technological advantages are an important factor for Ping An Smart Healthcare to “break through”.
No matter how mature the technology is, the final product still needs to land. Players in the medical artificial intelligence industry, after experiencing the financing upsurge, are now gradually beginning to “honestly” dive in and start the product launch.
“Currently, most medical AI companies have implemented their products in hospitals, imaging centers, and physical examination institutions, while Ping An Smart Healthcare has its own unique implementation scenarios: on the one hand, it cooperates with partners in Ping An’s internal medical ecosystem to form its own implementation model; On the other hand, with the help of Ping An Smart City and other forces, it has carried out extensive scientific research cooperation with key tertiary hospitals and scientific research institutes, provided diversified smart medical services for local governments, and empowered grass-roots hospitals through smart medical integrated solutions.” Xie Guotong said .
Xie Guotong’s remarks may explain another major advantage of Ping An Smart Medical in addition to technology – the diversification of landing scenarios.
Ping An’s internal medical ecosystem is divided into patient side, medical service provider side and payment side. With Ping An Good Doctor as the online portal, Ping An’s commercial insurance and medical insurance technology as the payment portal, and Ping An Smart Healthcare’s advanced medical technology as the enabling technology platform, Ping An’s entire medical ecological closed loop has been formed.
On the patient side, starting from Ping An Good Doctor, intelligent health monitoring, precise health management, remote diagnosis and treatment by family doctors, online drug purchase and offline medical service coordination can be realized, improving the service experience of patients in all aspects.
On the medical service provider side, Ping An Smart Healthcare provides end-to-end, full-cycle solutions, starting from disease prediction, screening, auxiliary diagnosis and treatment, chronic disease management, etc., to help patients find and treat early, and help grass-roots doctors to improve work efficiency and efficiency. level of diagnosis and treatment.
On the payment side, Ping An can effectively improve the operational efficiency of medical insurance and commercial insurance through data linkage with the government, medical institutions, etc. with the help of Ping An medical insurance technology, and help medical insurance to quickly enter the digitalization, so that the entire process of medical insurance records can be traced.
In the external scene, the government and medical institutions are the two cores of Ping An Smart Healthcare.
On the government side, Ping An Smart Healthcare’s intelligent disease prediction system has been successfully implemented in Chongqing and Shenzhen, helping government departments to predict influenza and hand, foot and mouth disease. The accuracy of the prediction models is over 90%.
On the medical institution side, Ping An Smart Healthcare aims at the grassroots level, and the number of institutions that have implemented the intelligent auxiliary diagnosis system alone has exceeded 14,000. According to Xie Guotong, the current non-tertiary hospital diagnosis and treatment basically accounts for half of the country’s total diagnosis and treatment times. In addition, the country has been encouraging hierarchical diagnosis and treatment in recent years. In the future, there is a lot of room for imagination of players from all walks of life at the grassroots level.
Through medical institutions, Ping An Smart Healthcare’s products have also penetrated into grassroots residents to help them with disease screening and chronic disease management. For example, the product smart eye screening OCT has been exhibited at events and exhibitions many times, and it has been stationed in the health station of Pengpu Xincun in Shanghai to conduct disease screening for the common people.
On June 6th this year, the National Eye Care Day, Ping An Smart Healthcare teamed up with Beijing Friendship Hospital to use the system for free fundus screening. Among the 31 citizens who were screened, 10 were found to be suspected of retinopathy. It was verified by the on-site ophthalmologist.
Caption: Ping An Smart City Smart Eye Screening OCT product has been stationed in Pengpu Xincun Health Station in Shanghai to conduct disease screening for the common people.
It is worth mentioning that, after Ping An Group became a smart city builder, it externalized its accumulated technological potential and industry experience to different scenarios, while Ping An Smart Healthcare, as the main business segment of Ping An Smart City, has launched its products. Into the “fast lane”. The cooperation between Ping An Smart Healthcare and governments at all levels has been deepened, and the joint research with top academic institutions and tertiary hospitals has also made breakthroughs. It can be said that with this momentum, its future development is worth looking forward to in the industry.
If the information age has made Internet companies such as BAT, then artificial intelligence will inevitably push a group of artificial intelligence companies to a climax. Who doesn’t want to catch up with this high-speed train and enjoy this wave of dividends?
For Ping An Smart Healthcare, its efforts in the past two years have given it a firm foothold in the field of medical artificial intelligence. However, the product itself is not static. How to further break through the technical bottleneck and meet the diversified needs of users in the future will be a problem that needs to be further considered.