In the field of medical AI in China, many enterprise giants are active. Tencent is a very unique “play” among them.

Huawei and Alibaba adhere to the concept of ecosystem, strive to build PAAS, hand over the breakthrough of specific diseases to lower enterprises, and strive to make their own cloud platform serve more projects; Baidu, which has strong AI scientific research strength, is still starting in the field of medical AI. Baidu world conference announced an AI fundus screening machine; IFLYTEK focuses on NLP. At present, it is accomplished in the detection of pulmonary nodules and the construction of smart hospital.

In contrast, Tencent obviously pays more attention to the practice of medical AI. The layout of diseases is dense and solid, and the promotion speed is fast.

From November 16 to November 18, Tencent medical AI Department launched three AI projects in Guiyang, Hangzhou and Shenzhen.

In Guiyang, Tencent chose to cooperate with Wang Ningli expert team of Tongren Hospital to start the AI glaucoma early screening project. In Hangzhou, Tencent participated in the construction of digestive endoscopy artificial intelligence committee to standardize the auxiliary diagnosis function of AI in the use of doctors. In Shenzhen, Tencent launched a national key R & D program to develop a new service model solution for clinical assistant decision support based on artificial intelligence.

Careful analysis shows that glaucoma and digestive endoscopy are strongly related to Tencent’s existing AI products, while CDSs is a new part, which can fill the gap in Tencent’s overall AI layout. In addition, at the launch meeting of this national key R & D plan project, Fan Wei, director of Tencent medical AI laboratory, also revealed Tencent’s research on psoriasis, otolithiasis, cardiovascular and cerebrovascular diseases and other diseases, which enriched the whole medical layout.

Why is Tencent’s medical road so delicate? To personally develop the CDSS system, and with the strength of the group team, go deep into the research of each subdivided disease of medicine? The logic behind this deployment is thought-provoking.

This paper consists of four parts:

What new actions does Tencent take after participating in national projects?

What kind of development trend does Tencent’s layout imply?

What new scenes did Tencent participate in under the cooperation? “

Down to earth, looking up at the starry sky, Tencent’s inclusive Ai Road.

From these four aspects, arterial network tries to clarify the new layout of Tencent in the field of medical AI.

Participating in the national special plan, Tencent launched AI + CDSS research

The key project of “research and development of digital diagnosis and treatment equipment” is one of the first six pilot projects launched by the national key research and development plan in 2017. The AI assisted diagnosis project of “special research and development of digital diagnosis and treatment equipment” led by Tencent this time aims to study the clinical assisted decision support technology and its service mode solution based on artificial intelligence.

Throughout Tencent’s medical layout, Tencent, which focuses on maternal and child health care and tumor treatment, has almost covered the whole medical process: the intelligent guidance system provides intelligent guidance and triage for patients through multiple ports; Subsequently, Tencent took charge of screening for common diseases; Wechat medical insurance payment and electronic health card in smart hospital provide patients with the functions of medical insurance payment and electronic medical record storage.

However, the above layout only lacks the clinical process. The lack of this system not only makes it difficult for Tencent to obtain the data of clinical patients, but also limits the diagnostic role of AI. Fan Wei replied in the interview: “Tencent does not have to do this system, but in order to deeply solve cardiovascular diseases, we must develop this system to assist.”

According to the statement of the mission statement of the national key R & D plan project, this project intends to solve the scientific problems of information extraction, semantic analysis, knowledge discovery and other scientific problems of artificial intelligence assisted clinical decision support system (aiacdss) traditional Chinese medicine health big data (including hospital information system, electronic case, health file, main complaint, medical record and other multi-source and multi-mode data) The five technological development directions imply the future development trend of Tencent medical.

The three plans are: “for acute coronary syndrome, stroke, skin diseases and other acute and chronic diseases corresponding to chest pain, headache, pruritus and other symptoms, develop accurate diagnosis and treatment decision support technology oriented to clinical path and technical specifications, and build a multidisciplinary consultation intelligent system with mixed diagnosis and treatment decisions of general and specialized departments”; Build and update the clinical knowledge base based on the natural knowledge base and multi-mode medical case diagnosis and treatment technology, and support the development and application of multi-mode clinical knowledge base and multi-mode clinical information processing system based on the clinical knowledge base and evolution technology; “Focus on the establishment of medical and health data cloud platform, establish medical and health files on the premise of ensuring data security and personal privacy, provide consultation services for patients, provide decision support for doctors, provide knowledge services and data support for researchers, and provide information support for the formulation of national medical policies”.

At the conference, Fan Wei disassembled the whole plan into five topics, which were jointly solved by Tencent, cooperative hospitals and cooperative enterprises. The five topics cover the decision-making system and knowledge base before, during and at the back end, as shown in the figure below:

At the meeting, Dr. Fan Wei only talked about the application of CDSS in multiple diseases, and did not mention the future development direction of medical informatization. However, it is undeniable that AI + CDSs is at the forefront, and many informatization software are transforming from traditional expert system to CDSS system driven by AI.

Compared with the traditional CDSS system, AI can not only associate the system with the authoritative knowledge base, but also assist doctors in clinical decision-making. In the ICU environment, it will further expand the application of death prediction, ventilator early warning and so on. Secondly, in the direction of scientific research, Tencent will jointly generate some high-value clinical data beyond routine examination in the process of cooperation with top-level third-class hospitals, and make full use of it to explore its value on the premise of complying with medical data use norms and data security.

In fact, Tencent is not doing nothing in the direction of medical informatization. Since 2014, Tencent has invested in a number of informatization enterprises. When the CDSS system is formed, Tencent must have a certain informatization level. It is necessary to go further and strengthen his, PACS and other systems with AI.

From module development to in-depth subdivision, from independent research to joint attack

Chen wanwan, vice president of Tencent, said: “the training cycle of medical talents is long, the cost is high, and there is a shortage of high-quality doctor resources. The public welfare of China’s medical industry makes it impossible to use price and market for supply management, and technology can alleviate the problem of supply and demand in a relatively short time”. It can be seen that solving the insufficient supply of medical resources is the fundamental motivation for AI to penetrate into medical treatment.

Looking back on Guiyang and Hangzhou, Tencent focused on glaucoma and digestive endoscopy, which is also to solve the problem of insufficient medical resources. Glaucoma early screening is an extension of shadow seeking sugar net screening products, while digestive endoscopy is an extension of shadow seeking gastrointestinal cancer. Nowadays, the relevant auxiliary screening products have matured. How to expand their functions and make them adapt to a richer environment is a major development direction of medical AI laboratory.

At the same time, these two-step plans reflect a major development trend of AI images, that is, from the image analysis of the whole process to gradually focus on early screening, accelerate the release of AI products to the grass-roots level, and solve the problem of high incidence of domestic diseases from the upstream.

For a long time, image AI has focused on DR, MRT and other images, and the feeding data often includes all stages of patients’ illness. For class III hospitals, such AI products do reduce the burden of doctors, but the more far-reaching value of AI lies in improving the service level of grass-roots doctors and popularizing early screening of diseases.

Take gastrointestinal tumors as an example. At present, the incidence rate of digestive tract tumors accounts for 43.5% of the incidence rate of cancer in China. If such tumors can be found early, the cure rate is as high as 95%. If AI technology can enter the screening link of digestive tract diseases, the mortality of digestive tract malignant tumors will be significantly reduced.

Professor Wang Ningli told “If we put this AI at the grass-roots level, it means that we have done one thing and sent all doctors in large hospitals to the grass-roots level. In the past, patients were treated in large hospitals or doctors in large hospitals were sent to the grass-roots level. Now information flow can replace personnel flow to complete this work, which saves a lot of labor costs. According to estimates, AI at this stage can save 30% of doctor resources, and this number will continue to increase with the development of AI technology. ”

That said, today’s medical AI is still a long way from doctors. Polo of Tencent health believes that medical treatment is a very complex problem, and complex problems should embrace more medical experts. Another topic of the Hangzhou conference, namely “strengthening academic linkage and tapping the potential value of medical AI”, is to solve the problem from this perspective.

Artificial intelligence data learning sample is a big problem. Chang Jia, general manager of Tencent smart medical products center, pointed out that there are hundreds of millions of data to learn, whether it is go or face recognition, but in the medical field, the requirements for data are higher. “If the same medical image data is marked by multiple doctors, the consistency is relatively low; the judgment of the same doctor looking at the same picture at different times is also inconsistent.”

Gastrointestinal cancer contains many kinds of cancer, and different doctors are good at different scenes. To accurately judge the status of gastrointestinal tract during the operation of digestive endoscopy requires the cooperation of many doctors. Medical resources do not allow such resource allocation, and AI has such potential.

This requires finding more and better “teachers” for AI. The participation and help of doctors and medical experts can train AI and provide better AI training standards. “Machines can’t learn from human beings, but they can gather the best of human beings. With the cooperation of top hospitals and experts, AI is expected to break through the bottleneck faced by doctors in some fields,” Chang Jia said Chen wanwan also pointed out that “the development prospect of medical AI completely depends on the open attitude and supporting policies of the medical community to cultivate valuable artificial intelligence.”

What other medical scenarios have Tencent laid out in cooperation with the hospital?

It is undeniable that AI at this stage is still in the primary stage, and the links of AI into medical treatment are not wide enough, deep enough and many enough. Therefore, Tencent chose more hospitals to jointly develop diseases. These projects often have the support of experts, data and national policies. At present, Tencent has extended the scene to the following areas.

1. Psoriasis

Psoriasis is also called psoriasis. This skin disease can not be completely cured. Almost every patient has a risk of recurrence after rehabilitation, and leads to complications such as metabolic syndrome, cardiovascular and cerebrovascular diseases, diabetes and so on. Generally speaking, the correct treatment can effectively alleviate the disease, but the cost is as high as 30000-50000 a year.

Can AI be applied to this scenario? The answer is yes. Dr. Wu Xian, an expert researcher of Tencent medical AI laboratory, put forward a complete set of solutions.

The first part is the offline self inspection system before diagnosis. Because dermatologists have different specialties, some are good at psoriasis, some are good at rosette sores, etc., it is very important to transfer patients to the correct doctor. The system can assist patient dialogue, upload pictures, offline screening and online referral. At the same time, it can also accurately classify and greatly shorten the consultation time. In only 2-3 minutes, the system can complete the patient’s main complaint and let the doctor make a quick judgment.

The core of the second part is the psoriasis diagnosis and prediction system, which can diagnose the subtypes of psoriasis and predict the corresponding complications and recurrence. If the doctor knows in advance that the patient is joint psoriasis through the system, biological agents can be used for treatment in advance; For other kinds of psoriasis, it can also make personalized treatment plans in advance, improve diagnostic efficiency and reduce treatment cost.

In the third part, after the diagnosis, Tencent tried to simulate the PASI (psoriasis area and severity index) score and improve the PASI score through consistency. The reconstructed PASI scoring system will be put into wechat applet to facilitate patient self-test and realize continuous prognosis tracking of patients.

2. Otolithiasis

Under normal circumstances, the otolith is attached to the otolith membrane. When some pathogenic factors cause the otolith to fall off, these falling otoliths will swim in the liquid called endolymph in the inner ear, resulting in strong vertigo, which is otolith.

Otolith was studied by Tencent medical AI laboratory in cooperation with Professor Li Huawei of Otolaryngology Department of Shenzhen Second People’s hospital. According to Dr. Fan Wei, Tencent medical AI laboratory takes the black part in the middle of the eye as the key point for detection through the deep learning network to improve the diagnostic accuracy. At present, this AI project is still in the research and development stage.

3. Cerebral palsy, scoliosis

What cerebral palsy and scoliosis have in common is that these two diseases will cause movement disorders, which mostly occur in children and seriously affect the appearance and gait of children. Although these two diseases cannot be cured, the earlier they are found and corrected, the more likely they are to exercise normally.

Tencent’s work in this field is mainly to assist doctors in Shenzhen Hospital of the University of Hong Kong to transfer patient and instrument data to Tencent cloud. In this way, doctors can check faster, transfer the children with problems to the hospital for treatment in time, and intervene in patients with scoliosis and cerebral palsy as soon as possible.

Compared with other AI scenarios, the research and development of this AI is more in line with the original intention of seeking film to pay attention to women and children, and can better show Tencent’s sense of social responsibility and social value as an industry leader.

4. Parkinson’s disease

Different from cerebral palsy, Parkinson’s disease is a neurological disease. Compared with cerebral palsy, which is related to motor control, the research of neurological diseases is more complex. Fortunately, there are still breakthroughs in science and technology. In 2017, researchers at the University of North Carolina developed a set of deep learning algorithms that can predict autism in infants. This prediction method has 81% accuracy and 88% sensitivity. Compared with 50% accuracy of behavior questionnaire, the reliability is greatly improved.

Parkinson’s research has also made progress. The intelligent evaluation system of Parkinson’s motor function developed by Tencent medical AI laboratory in cooperation with Professor Wang Jian of the Department of Neurology of Huashan Hospital is trying to evaluate the condition of Parkinson’s patients through video analysis.

Through computer vision, researchers automatically mark 21 nodes on the patient’s hand to capture the movement of the patient’s hand. When walking and performing specified actions, AI can quantitatively analyze the movement of patients. The special feature of this technology is that the measurement of patients is carried out without wearing any sensors, which can shorten the examination time of patients over 30 minutes to 3 minutes. At the same time, patients who are inconvenient to travel can see a doctor without going to the hospital. Compared with the existing auxiliary diagnostic products, patients can intuitively feel the changes of AI to their treatment methods.

Professor Wang Jian said: “We did a pre experiment. The expert group scored more than 1000 videos from nearly 200 patients and handed them to the machine for learning. After learning, the machine will score the experimental group and match the results with the expert scores. The preliminary coincidence of the experimental results reached 81.3%. This score still has a lot of room for improvement, but the success rate of the first stage has exceeded 80%, which is also a small success. ”

5. Electrocardiogram

ECG is a potential scenario for AI application in the future. Recently, Lepu medical artificial intelligence ECG product was approved by FDA, which also indicates the potential of this scenario.

According to Dr. Du Nan, an expert researcher of Tencent medical AI laboratory, the application of AI in ECG mainly includes the following three points:

1. Clinical monitoring. In emergency and nursing homes, doctors need to monitor the patient’s ECG for 24 hours. AI is competent for this monitoring task and sends the data of patients with abnormal ECG to doctors.

2. Diagnosis and treatment assistance. More and more family patients will collect data through the family electrocardiograph, upload it to the cloud platform, and then the remote experts will mark the data. However, in the long-time mechanical work, the experts may be tired, and it is difficult to avoid misjudgment and missed judgment. Artificial intelligence can help doctors reduce missed judgment and misjudgment.

3. Risk prediction. For postoperative and post hospital patients, doctors hope to observe the changes of patients’ ECG through home equipment. If AI detects that the patient has abnormal ECG at home, it will automatically send a notice to ask the patient to return to the hospital for further consultation, and inform the family to pay attention to the patient’s health.

Through the realization of the above functions, on the one hand, ECG AI can liberate the burden of doctors, on the other hand, it can find the patient’s condition in time, which not only improves the diagnostic efficiency, but also saves the patient from the dilemma of sudden condition.

6. Cardiovascular related diseases

Cardiovascular disease is one of the main diseases endangering people’s health in China, and it is also one of the important reasons for Tencent to develop CDSS system.

Acute coronary syndrome (ACS), also known as myocardial infarction, is another scene of cooperation between Tencent and Professor Sun Ningling of Peking University People’s hospital. When patients have atypical symptoms such as fatigue, sore throat and palpitation, it is difficult for patients and doctors to judge the condition of patients, which is prone to false registration and misdiagnosis.

Professor Sun Ningling hopes to solve this problem through intelligent interrogation, that is, to maximize the collection of patient information through intelligent interrogation, so as to more accurately evaluate the nature of the disease and answer whether the patient’s disease is acute or subacute? Is it treatable? Whether interventional therapy is needed and so on.

“We compared and identified some hospital arrhythmia diagnosis data with Tencent AI, and we can see that AI is higher than the diagnosis rate of hospital equipment and manual in terms of positive accuracy and sensitivity. Therefore, I think using AI for early diagnosis of abnormal ECG can greatly improve the clinical management effect of patients.”

Down to earth, looking up at the starry sky, Tencent’s inclusive Ai Road

From the overall layout, Tencent first focused on the AI scene with huge demand, and then continued to conduct in-depth research on relevant diseases along the existing product line. After completing the basic layout, Tencent began to increase cooperation with hospitals and the government, try to open up tracks in some more subdivided fields and carry out some innovative AI application development.

Under this idea, we can divide Tencent’s disease selection into the following three scenarios.

Main application scenarios

At this stage, Tencent’s products have covered six scenarios, including early screening for esophageal cancer, early screening for lung cancer, early screening for sugar net lesions, early screening for breast cancer, early screening for colorectal cancer and early screening for cervical cancer. What these six products have in common is that the number of patients far exceeds the workload of doctors, and a large number of patients gather in class III and class A hospitals and cannot be decentralized to the grass-roots level; At the same time, the high cost of cancer treatment has greatly increased the burden on families and countries. In this scenario, the role of AI includes improving the early screening ability of grass-roots doctors, reducing the burden of class III hospitals and doctors, freeing doctors’ time, reducing medical expenses, etc. There is a clear possibility of commercialization in these scenarios, that is, once the nmpa is approved, the layout enterprise can quickly recover the R & D funds.

Potential application scenarios

Glaucoma, ECG and ACS are potential application scenarios. This kind of AI application is affected by one or more factors such as relatively small market, difficult to obtain data and high technical threshold, but there is still room for commercialization. These scenarios will develop rapidly after the commercialization of the first batch of AI products.

Scientific research application scenario

AI scenes such as cerebral palsy, psoriasis, otolith and Parkinson’s disease, on the one hand, in order to meet the national research requirements for some special diseases, on the other hand, the research itself has pioneering significance. The research on psoriasis meets the third item of the project: “for acute and chronic diseases such as acute coronary syndrome, stroke and skin diseases corresponding to chest pain, headache and pruritus, develop accurate diagnosis and treatment decision support technology for clinical path and technical specifications”, while the intelligent evaluation system of Parkinson’s disease motor function uses visual capture technology to analyze the patient’s movements without wearing sensors, It not only saves a lot of examination time, but also realizes offline consultation, which is of great pioneering significance.

However, no matter which mode, it means that Tencent takes another step towards to B business. At the same time, Tencent will obtain a lot of training data and AI development experience from scientific research cooperation, and Tencent cloud will therefore play a deeper role in AI + medical treatment.

However, the long road of medical research and development is similar to a long-distance race. Participants need to restrain the temptation of sprint and reasonably arrange their physical fitness in order to achieve better results in the whole race.

Leave a Reply

Your email address will not be published.