An industry that only burns money but doesn’t make money, because the epidemic ushered in a turning point


Welcome to the wechat subscription number of “create a story”: sinacchangshiji
Editor / Luo Guangyan and Shi Ming
Source: one (ID: yitiaotv)
This year’s new epidemic,
Probably the most “high-tech” anti epidemic battle in history:
For the first time, scientists used gene detection to locate the virus as soon as possible,
5g and big data can accurately track the flow of virus at any time,
The algorithm helps doctors to see CT films,
The 10 minute diagnostic time changes to 2 seconds,
Robots deliver medicine, rice and kill,
Online diagnosis and treatment allows patients to avoid face-to-face with doctors
The Internet plus trend has been blowing for years.
Finally, the whole medical industry.
Now the domestic epidemic has gradually subsided,
Many people wonder, what are these “black technologies”?
Will they completely change our lives after the outbreak?
Or is it just a flash in the pan?
I interviewed five people fighting in the front line of “medical AI”,
There are programmers, engineers, chief physicians, community doctors
They witnessed the battle of “Ai anti epidemic”,
And remind us,
In the end, smart medical services serve people:
“Technology should bring people closer
Rather than bringing people closer to the machine. ”
AI campaign: Four Stories
It takes 5-15 minutes for the doctor to read the film with naked eyes,
Our AI only takes two seconds
Researcher, Tianyan laboratory, Tencent, Weidong
I joined Tencent Tianyan laboratory in September 2018, mainly focusing on the algorithm development of medical AI image analysis. Simply put, it is to let the computer look at thousands of images and tell it which organs in the images are normal, which are problematic and where the problem is. After enough learning, it can make a similar judgment according to the information in the image.
This technology has been developing rapidly since 2017, which is used to diagnose cardiovascular diseases, eye diseases, pneumonia and so on.
After the outbreak of novel coronavirus pneumonia, our laboratory has been paying attention to the development of the epidemic, but it has been suffering from useless weapons. Until February 5th, novel coronavirus pneumonia was diagnosed in Hubei by the CT results which were also included in the diagnostic criteria. The suspected cases could be diagnosed by CT.
Tencent AI diagnosis output chart
Novel coronavirus pneumonia CT images were immediately isolated from our homes and heard a news. We immediately organized a team to start developing the AI diagnostic model of the new crown pneumonia imaging. Because of the lack of face-to-face communication and the tight time, we are still under a lot of pressure.
At the beginning, the biggest problem is the lack of data. AI is like a student. To make it accurately identify diseases, it needs to see a lot of lung images to learn to judge. We obtained some novel coronavirus pneumonia data from the cooperative hospitals all over the country, but it was not enough.
So we decided to let AI learn the data of other pneumonia first and look at thousands of cases. After it has seen the data of pneumonia, it has accumulated a whole knowledge of pneumonia, which is the lung image with pneumonia and which is not.
Novel coronavirus pneumonia is a new way to adjust the algorithm. We will adjust the algorithm to give AI a hint, tell it that the new crown pneumonia is different from the pneumonia image it has seen. In this way, novel coronavirus pneumonia can be captured in the image and the patient can be diagnosed.
Because of the emergency situation, we are in a hurry this time. We have debugged the algorithm in about 10 days and deployed it in Wuhan rihai square cabin hospital on February 21.
After putting it into use, the feedback from the medical staff is very ideal, because reading the film is not so tired. At that time, the pressure for doctors to read and analyze CT images was very great. After the change of clinical diagnosis standard, the number of newly diagnosed cases increased nearly ten times every day. Each patient had a chest CT, which would produce 300 images. Each time the doctors read the images with naked eyes, it would take 5-15 minutes.
For the patients who have been diagnosed in the treatment period, CT should be done every 5 days. At that time, there were more than 60000 confirmed patients in Hubei Province.
AI can tell if novel coronavirus pneumonia is the fastest in two seconds. If AI judges the novel coronavirus pneumonia, it will also mark lesions in CT images, tell doctors the basis of their judgement, and the process of marking will take up to one minute.
AI judgment can not be directly used as the basis of medical diagnosis, but it can provide reference for doctors and greatly improve the efficiency of diagnosis.
Some doctors worry about the accuracy of AI when they first use it. But according to our research, our AI model has a higher diagnostic accuracy than younger doctors. And with more and more case data, we will continue to train and adjust the algorithm, so that its accuracy continues to rise.
In Hubei, novel coronavirus pneumonia novel coronavirus pneumonia was found in the patient. CT image of the patient showed that he had a large lung lesion and looked like a new crown pneumonia. But a senior doctor didn’t think so.
Our algorithm also judged novel coronavirus pneumonia. Finally, the patient was diagnosed with pulmonary edema. So our model has a strong ability of identification.
Looking at CT results with AI, it should be said that it is the most mature part of the current development of intelligent medicine. Because this part has a very objective measurement index, the more objective and less human factors participate in the work, the algorithm can perform better than human.
Before, the disinfection robot couldn’t be sold,
This outbreak caught fire on CCTV
Marketing Manager of Yinchao titanium rice robot technology
2018 is the year of smart medical station, and I joined in TIMI robot that year. Our main types of robots are: disinfection robot, material distribution robot and ward service robot.
We have many hospital customers in Wuhan, and each customer has a sales person to follow up.
In the middle of January, the salesmen found that the hospital had only bought a disinfection robot, which was specially used in the operating room. As a result, other departments often borrowed the robot, the infection department borrowed it for disinfection, and the ICU borrowed it for disinfection. We immediately realized that the epidemic was serious.
Titanium rice robot works in the front line of new crown

On January 20, TIMI had already deployed a team of engineers to mobilize robots to send directly to Wuhan. But at this time point, it’s Spring Festival and epidemic situation. Logistics and transportation have become the most headache for us. We’ll contact the logistics company. They won’t take your order at all.
In this case, we organize our own vehicles and hands, plan the transportation routes of goods and engineers, and escort the robots to the past. When we sent the first 10 disinfection robots to Wuhan Union Medical College Hospital, it was 2 a.m. on January 21.
Disinfection robot is working
After being sent to the hospital, the rest of the work is very simple. Our engineers will start to scan the map, which is to let the robot recognize the surrounding environment. Like when we human beings go to a strange place, we should first get familiar with the environment. I need to have a look around and know that the table is here and the bench is there. Next time I walk, I will avoid it.
The same is true for robots. They will walk through all the spaces and learn by themselves, knowing that this is a table, which needs to be taken care of when disinfecting. After scanning the map, engineers can leave and let the robot disinfect itself.
Some people may think, what is special about this kind of robot? My sweeping robot can do it. But in hospitals, the requirements for safety and cleanliness are much higher than those of hotels, restaurants, or factories.
For example, disinfect the operating room with a robot. Our cell phone went into the operating room, even if it was full of signals, the phone could not be called out. In the robot, if the traditional GPS navigation and the traditional obstacle avoidance algorithm are used, it is easy to collide, so there will be a big problem. Therefore, medical robots will have unique algorithms to ensure their stability and obstacle avoidance ability.
Material distribution robot in the shelter hospital
Before the outbreak, hospitals preferred to buy our delivery robots. Because they thought before disinfection with artificial disinfection or ultraviolet disinfection on the line, why spend a lot of money to buy disinfection robot? But this time, they found that neither of the two methods of disinfection could be thorough, and that manual disinfection would face the risk of exposure.
At the beginning of design, our disinfection robot is designed according to the design concept of human-machine separation. The robot itself can disinfect the place that needs to be disinfected before people enter.
At that time, the epidemic situation in Wuhan was very serious, and hospital disinfection was very frequent. Originally, a disinfection robot only used to disinfect the operating room on the 4th floor. Now it is not only necessary to disinfect the whole floor, but also to disinfect the building once, and sometimes to move it to the next building. In addition to disinfecting hospitals, medical staff offices and dormitories are also disinfected.
At that time, one of their family members drew our disinfection robot “big white” when drawing, because he thought our robot was protecting their safety. We are all moved after seeing it and feel that our work is meaningful.
“Big white” painted by children
Our online diagnosis and treatment,
Served one tenth of the severe patients in Wuhan
Hao Nan, initiator of NCP (new crown) life support network
“NCP (new crown) life support network” is an online diagnosis and treatment project. I started to think about it on the night of January 22. At that time, I received a message that Wuhan hospital was not enough and doctors were not enough. But you can’t let the sick people not go to the hospital. What about the sick people who can’t see the disease?
I used to be a doctor in Peking University Hospital for seven years. On January 23, I launched action initiatives in dozens of wechat groups, including the alumni group of Beijing Medical University. In one week, more than 400 medical workers, including doctors, nurses and medical students, were recruited inland. As well as more than 200 volunteers with social workers and psychological backgrounds, they provide care and companionship services for online patients.
We are very different from other online diagnosis and treatment platforms, where patients call in and ask for help, and then it’s over. We hope to pay equal attention to diagnosis and treatment and care, so we divide wechat group into three levels, which is the same as the real hospital.
Process patient information by category
The first is the triage group. Before you come here, you will be greeted by volunteers, who will ask you to fill in the form and ask you where you are uncomfortable, what symptoms you have and whether they are serious. According to the situation you fill in the form, it is a triage platform to determine which level of wechat group you should go to.
First line is equivalent to ordinary outpatient service. Doctors will answer questions and solve doubts for patients in wechat group. The difference is that these patients don’t leave after asking, and stay in the group. If they have any new questions, they can continue to ask, keep in touch with the doctor, instead of constantly changing people to see them.
If the patient is in a serious condition, we will transfer it to the second line. The second-line groups are all more professional and experienced doctors with rich ward management experience, who will continue to pay attention to the patient’s condition online.
If the patient is seriously ill and in danger of life, we will transfer him to the third tier group. The third line is equivalent to intensive care unit, in this group, there are several doctors around a patient.
Recruit and reserve doctors as patients increase
If such a setting is placed offline, it will need a lot of space, and there will not be many people to serve. Offline, even if patients can enter the hospital, they can’t see the doctor.
This is the biggest advantage of online diagnosis and treatment: it can maximize the use of medical resources. Online doctors can see more than ten wards remotely, and pay attention to the situation of many patients in different groups at the same time.
Some of our core doctors often go to other partners’ wechat consultation group for consultation, while guiding the care and even rescue of critical patients in our second and third line wards, which does not exceed their load. Let’s serve the most people with few resources.
Novel coronavirus pneumonia was consulted most of the time, but there were other diseases, so we were equipped with doctors from various departments.
Contactless distribution process of materials produced by NCP life support
I’ve been a doctor, and I know that doctors tend to get impatient when they see a doctor online. The online diagnosis and treatment is separated from the screen, and the communication effect is not good. Sometimes, even if the professional advice is obtained from the doctor, the patient will still experience bad, or even psychological collapse.

So we have special social workers and psychological volunteers to provide companionship and care, and online one-to-one psychological guidance for some cases. Novel coronavirus pneumonia has been the cause of a case. Her parents died in this outbreak. She also had a new crown pneumonia. We followed her condition and gradually recovered from her illness.
From the time I started this project to the first ten days of February when the shelter hospital was in place, the middle 20 days was the most stressful time for us to work. We have roughly estimated that we have directly or indirectly served nearly one tenth of the severe patients in Wuhan at that time.
Thank you message from the family of the patient who was rescued
With our current level of science and technology, in fact, it can make doctors’ work easier, connect enough social work resources, and make doctor-patient relationship more human to human, rather than just human to machine communication.
The common people are very simple. There are only three requirements for intelligent medical treatment:
Helpful, good operation and supporting services
Pan Xiangdong “top ten family doctors” in Shanghai
I am a general practitioner in charge of Hongqiao street, Changning District. Many people will think that you just take the medicine prescribed in the community hospital? In fact, outpatient service is just one of my tasks. I also need to be responsible for the basic public health management within my jurisdiction, such as tracking the situation of patients with chronic diseases such as hypertension and diabetes, preventing and controlling infectious diseases, etc.
On January 28, we were instructed to start work ahead of time to prevent and control the epidemic. At that time, the epidemic situation in Hubei had broken out. Shanghai organized medical teams to support Hubei since the eve of the lunar new year. Our community should prevent backyard fires, ensure the normal development of daily work, and meet the health needs of chronic disease patients in the community.
There are nearly 2000 diabetics in the street area of our center. We follow up these patients every month, monitor their blood sugar and other values, and adjust the medication according to their condition.
Blood glucose measurement
After the novel coronavirus pneumonia came, many old people were afraid to go out because they were afraid of infection. In the past, he might come to me once a week. I would like to knock wooden fish every time. I would keep telling him the precautions for diabetes and always knock him well. The blood sugar is basically stable.
Now it’s all out of order. The old man is trapped at home and can’t exercise. He may not pay much attention to his diet. Our doctors can’t monitor it remotely. Their blood sugar level is under control.
I’ve also been exposed to some portable glucose continuous monitoring devices before. The sensor of this device is like a sticker. After it is pasted on the arm, it will not be affected in 14 days whether you take a bath or swim or exercise. You can take a scanner to monitor your blood sugar in real time anytime and anywhere.
I felt for once that the biggest impact of this portable device is that it will change your lifestyle. With this device, I can think of my blood sugar anytime and anywhere. During that time, every time I finished a glass of milk or a loaf of bread, I would use the instrument to measure my blood sugar value. I find that my blood sugar fluctuates obviously at night, and I will really start to pay attention to my dinner diet.
Portable sticker to monitor blood glucose value at any time
But I later found that many patients did not want or would not like to use these intelligent portable devices. On the one hand, most of these patients are old people. They are conservative. They feel that these devices are still unfamiliar and complex to operate, and they need to pay for them. They will refuse to use or abandon them in the middle if they don’t feel the intuitive benefits.
On the other hand, according to the law, the data of these portable devices can only be used as a diagnostic reference for community doctors, not as a diagnostic standard.
I personally think that if we want to popularize the intelligent medical equipment into the community, we need to meet three conditions: first, it should be helpful to me, second, it should be easy to operate, and third, it should have supporting medical services.
If such portable devices for monitoring blood sugar can be popularized from top to bottom in the future, the management of family doctors for patients with chronic diseases will be more convenient and efficient.
Now, we follow up every patient manually offline to confirm whether they have any problems. It will cost a lot of manpower and material resources, and the coverage and effect may be limited. With the development of aging population, we need to pay attention to and manage more and more patients with chronic diseases. If we completely rely on the existing human resources to cover them one by one, it is a little inadequate.
In the future, if the smart device is used, the relevant data of patients can be uploaded to the smart device data center remotely. After big data screening, abnormalities are found and pushed to the medical staff in real time. I think this can greatly improve work efficiency and management coverage, and patients in need can get medical help in a more timely manner.
Internet medicine: the future has come?
Smart medicine is not a “black technology” that comes out suddenly. It has been dormant for decades.
In the 1950s, artificial intelligence was just invented and applied in the field of medical treatment. But until 2006, the algorithm of neural network deep learning was invented, the learning ability of AI began to take a qualitative leap. Ten years later, alpha go beat Li Shishi. The topic of artificial intelligence is hot all over the world, and medical AI becomes the focus again.
2016 is known as the first year of AI medical investment in China. 27 enterprises raised funds in this year, of which 16 enterprises raised more than 10 million yuan. In 2017 and 2018, the domestic AI medical industry announced nearly 30 financing events, with a total financing amount of US $142 million in the field of medical imaging diagnosis. According to KPMG data, this financing scale is equivalent to a quarter of the global investment in the industry in the same period.
However, since 2018, many medical AI enterprises’ products have encountered the problem of realization in practical application, “only burning money but not making money”. By the end of 2019, medical AI has entered a cold winter, waiting for reshuffle.
The outbreak of a novel coronavirus pneumonia in early 2020 was considered to have accelerated the shuffling process.
What humans may miss, AI won’t

In the novel coronavirus pneumonia, the AI+CT imaging diagnostic technology developed by the hospital has averaged less than 20 seconds and the accuracy rate is 96%.
Novel coronavirus pneumonia, CT technology and AI are also being screened for products of new crown pneumonia. Wuhan Central Hospital uses the AI assisted diagnosis system of Shukun technology, which can quickly read chest films and identify inflammatory lesions within 2 to 3 seconds. Wang Xiang, director of imaging department of the hospital, said the system “improves doctors’ working efficiency by 50%.”.
In fact, AI medical imaging is recognized as the most mature and most likely to take the lead in commercializing medical AI products. The data of thousands of cases can develop a set of diagnostic products with relatively high accuracy.
In 2019, there are more than 140 domestic enterprises engaged in medical AI, nearly 120 of which are engaged in medical imaging business, and about 100 enterprises are engaged in pulmonary nodule imaging products.
Before the outbreak of the epidemic, the resources of domestic imaging doctors were in a serious shortage: the annual growth rate of the number of outpatient radiologists in China was 30%, but the annual growth rate of the number of imaging doctors was only 4.1%.
According to data from Le Qing think tank, the number of misdiagnosed images in China is about 57 million every year. In addition, patients need to queue up to make an appointment for filming all year round.
The sudden outbreak of the epidemic has brought two great challenges to front-line imaging doctors: one is the sharp increase in the number of reading films, a large number of which are high-intensity repetitive physical labor; the other is the lack of basic hospital equipment, uneven experience and quality of doctors. Both will lead to misdiagnosis and missed diagnosis.
“I always remind myself not to cry because my eyes need to be used to see massive CT and X-ray films,” said fan Yanqing, director of Radiology Department of Wuhan Jinyintan hospital
According to the situation at the time of the most serious outbreak, more than 10000 suspected patients are newly added every day in Hubei Province, with a total of more than millions of CT films. At the same time, there are tens of thousands of confirmed patients’ needs for reexamination. It is estimated that the front-line film reading doctors have a daily workload of 5 million.
Robot analysis X-ray brain tomography
AI can solve the urgent need of doctors.
Pan Yunhe, academician of the Chinese Academy of engineering, said that at present, the AI model used to analyze chest radiographs has achieved a detection rate of more than 98% for lung cancer, far higher than that of artificial.
Shao Yifu Hospital of Zhejiang University used AI to analyze the image of keratitis. The accuracy rate has exceeded 80%, which is more accurate than 96% of the doctors participating in the test.
On March 16, Microsoft Research Institute, National Institutes of health and the White House Office of science and technology jointly released covid-19 Open Access database, and people began to use AI to mine the connection between academic papers.
AI will understand the content of the paper faster than trained scholars, and track some answers to the most critical questions of the current epidemic: how does the new coronavirus spread? What do we know about its origin and evolution? How do we make vaccines and specific drugs?
Michael kracios, chief technology officer of the US federal government, told the media: what human beings may miss, AI will not.
What do we do when AI is misdiagnosed?
Although the accuracy of AI diagnosis is very high, reaching 96% or even 99%, there are still 1% or 4% errors.
A number in statistics is a mountain when it comes to individuals. Who is responsible for this mistake once it happens?
Japan Economic News reported on July 3, 2018 that due to the possibility of AI misdiagnosis, the Ministry of health, labor and welfare of Japan has positioned AI medical equipment as a device to assist doctors in diagnosis, stipulating the final responsibility for diagnosis, which shall be borne by doctors.
Researchers in the British Medical Journal (BMJ) recently warned that “many studies and media claim that AI has reached the same level as, or even better than, human experts in interpreting medical images, but the quality of AI is actually poor and exaggerated, which poses a risk to the safety of patients.”
Researchers at Imperial College London found that in a study of the accuracy of clinical diagnosis of AI, 350 Chinese cataract patients participated in the experiment, some of them were diagnosed by AI and some were diagnosed by experts. The average time of AI diagnosis is faster than that of experts, but the accuracy rate is 87%, and the accuracy rate of experts is 99%.
AI is like a child, needs to grow. Its performance is related to the quality of the “food” it can get, which is data.
The speed of data generation and sharing is increasing rapidly. According to IDC (international data company), the total amount of medical data in the world will reach 40 trillion GB in 2020, 30 times of that in 2010.
However, 80% of this data is unstructured data. They are hidden in the case data of the hospital, in a large number of images, without proper quantification, they can not be extracted, recognized and processed by AI.
At the beginning of 2020, vas Narasimhan, CEO of Novartis, explained how data has become a pain point in the development of medical AI: “we have to spend most of our time sorting out data sets before we can run this algorithm.” He believes that we underestimate how little high-quality data is available and how difficult it is to organize and link data.
In China, the data of health care industry is scattered. Hospital is the largest production organization of medical data, but in reality, no hospital is willing to share medical data free of charge. On the other hand, these data also involve patient privacy.
When domestic AI medical startups need patient data to train AI and let AI learn, they can only talk with hospitals one by one, which creates many “grey areas”.
Liu Shiyuan, director of Radiology branch of the Chinese Medical Association, once suggested that if a large data center can be built within a province and the data of all hospitals in the province can be collected and used, the development of medical AI may be promoted more legally and effectively in the future.
Medical treatment is the relationship between people,
It’s not about people and machines
In this AI anti epidemic, the other two bright applications are medical robots and online diagnosis and treatment.

In Shanghai Children’s Medical Center, the robot “Xiaobai” works. In the case of insufficient protection resources, it reduces the frequency of face-to-face communication between doctors and patients, and reduces the risk of cross infection between doctors and patients.
Huoshenshan and leishenshan hospitals also use intelligent delivery robots, which can deliver laboratory tests, medicines and meals to patients according to the needs of the hospital.
It’s no problem that the sweeper robot doesn’t look like a human at all. But in addition to its normal function, the robot used in the hospital always seems to be expected to be able to simulate human’s appearance more, like human’s appearance.
This is related to the patient’s psychology. Researchers from the Military Medical University of China recently conducted a survey of “patients’ cognition and trust in artificial intelligence medical treatment”. They found that the highest acceptance and trust of patients to AI was in the medical logistics sector, followed by the medical auxiliary sector with less contact between doctors and patients.
In the operation and other medical core links, the more work of artificial intelligence intervention, the more important the role, the lower the patient’s acceptance and trust.
When it comes to life and death, we tend to trust our peers rather than the seemingly cold AI.
Wu Xi, deputy chief physician of the Endocrinology Department of Huashan Hospital Affiliated to Fudan University, told us that their department is now using AI to analyze the image of diabetic retinopathy. He believes that AI has the ability to replace human doctors for diagnosis. This is determined by the nature of medicine.
“Different doctors must be able to make the same diagnosis according to some symptoms, unless some symptoms are ignored. Since people and people can make the same diagnosis, aren’t people and AI doing the same? AI will be more sensitive and careful, and the probability of making mistakes will be smaller. ”
In February this year, Nature magazine reported a new kind of blood drawing robot. The total success rate of blood sampling was 87%, and for the subjects with clear vein, the success rate was 97%. This result is better than most human nurses.
Photograph: Miriam D รถ RR
“Our future work may be to cooperate with AI, not to study how to overcome disease, but to figure out what is disease, what is patient and what is health.”
During the epidemic period, the traffic of several online diagnosis and treatment platforms increased explosively: during the epidemic period, the number of visits of Ping an good doctors to the platform was 1.1 billion, the daily average number of new users was 9 times as much as usual, and the number of new users registered increased 10 times.
After novel coronavirus pneumonia was introduced by doctors in spring, the number of patients in Wuhan and its surrounding areas increased by 10 times compared with the previous ones. The online consultation platform of dingxiangyuan has increased the number of daily visits by more than 300%.
In response to these patients on the online platform, they are living people. As of December 31, 2019, the self owned medical team of Ping An Hao doctor is 1409. On Ding Xiangyuan’s app, there were more than 15000 active doctors online during the epidemic.
Hao Nan, founder of NCP life network, said caring is sometimes more important than medical help itself. Edward Trudeau, the founder of the first tuberculosis laboratory in the United States and a physician, left a famous saying that has been revered by the medical community so far: “sometimes cure, often care, always comfort”.
The development of medical AI, perhaps in the future, will bring us closer to the essence of Medicine: promoting the relationship between people rather than between people and machines. Let’s see.