Outbreaks of new coronavirus pneumonia are often compared with SARS in 2003. Compared with 17 years ago, in this difficult epidemic battle, increasingly mature AI technology has become an important force.

AI is often mentioned, and it is always associated with keywords such as "loss", "not landing" and "bubble", but under the epidemic, AI has become closely related to our lives.

At airports and subway stations, you have received temperature measurement from an infrared thermal imaging system; when returning to your residence from a foreign country, you may have received a phone call from an intelligent robot to fill out information.

When entering the office building, AI can recognize whether you are wearing a mask; in the hospital, doctors rely on the AI-assisted diagnostic system to quickly determine whether the patient is likely to develop new coronavirus pneumonia.

The epidemic has accelerated the integration of AI and urban life. On the one hand, AI can improve efficiency, and on the other hand, AI can replace labor in some scenarios, which can reduce the risk of infection.

So, in which scenarios are applications welcoming new opportunities? The epidemic will eventually pass. Will the demand for AI in various fields continue? Rancaijing explains in a story.

AI-assisted diagnosis, find the lesion in 2-3 seconds

The most stressful place in this epidemic situation is hospitals in various places, especially hospitals in Wuhan.

The Central Hospital of Wuhan, where Dr. Li Wenliang worked previously, is only one stop away from the South China Seafood Market, and is located in the core area of the epidemic.

After the hospital was determined to be a designated patient for fever patients, during peak hours, the hospital had an average of more than 1,000 CT examinations per day, and front-line radiologists could only read the films overnight.

A chest CT has more than 300 cross-sectional films, and the doctor's naked eye alone reads the images one by one, which is quite a workload.

In 2018, Wuhan Central Hospital began to use Shukun's CT artificial intelligence-assisted diagnostic products, which are mainly used in the field of coronary heart disease.

So after the epidemic broke out, Shukun Technology immediately organized R & D personnel to develop the new coronavirus pneumonitis AI assisted diagnosis product during the Spring Festival and went live in Wuhan Central Hospital in early February.

This product can pick out the lesion image from 300-layer CT images in 2-3 seconds, and generate a 3D lung lobe model, mark the area where the lesion is located, and then guide the doctor to make a reasonable triage of the patient to improve the detection efficiency of new coronavirus pneumonia .

If the patient does CT again, the product can automatically compare the changes of the lesions before and after the patient, which is convenient for doctors to judge the condition.

Wang Xiang, director of the imaging department of Wuhan Central Hospital, has publicly stated that this AI system can improve doctors' efficiency by 50%.

Efficiency has improved. How can accuracy be guaranteed? The relevant person in charge of Shukun Technology told Rancaijing that it is not possible to misjudge, but also to suppress false positives, that is, due to environmental, operational, experimental methods, or the patient's own factors, people who do not have positive symptoms have detected positive results. Dimensions are the difficulty of existing products.

"To control these two dimensions, we need to continuously adjust the algorithms of the AI system with years of medical clinical experience and case judgment ability." The person in charge said that the accumulation of doctors' experience and data is the AI-assisted diagnosis. The key to the system going further.

It is also thanks to the accumulation of data from Shukun Technology in the field of coronary heart disease diagnosis, this time it was able to quickly iterate new coronavirus pneumonia related diagnosis products.

According to reports, the AI system has been used in dozens of hospitals across the country, such as Wuhan University People's Hospital, Wuhan University Zhongnan Hospital, Wuhan Traditional Chinese Hospital, and donated to Thunder Mountain Hospital.

During the epidemic, Yitu Technology and Shangtang Technology also upgraded the new coronavirus pneumonia based on existing smart diagnosis and treatment platforms.

"This epidemic gave me the biggest impression. For ordinary people, good medical resources are really very limited. The epidemic has accelerated the verification of the effectiveness of AI assistance, and there may be more primary medical institutions to choose AI products. , So that AI products with diagnostic capabilities in large hospitals can be quickly copied to the grassroots. "The person in charge told Rancaijing.

He believes that the AI-assisted diagnostic system has scientific significance in addition to diagnosis for this pneumonia. In the next stage, we should do some clinical research on the disease to build a better prevention and control system. In the future, the application of AI technology in the field of auxiliary diagnosis will become a trend.

Non-contact temperature measurement, even without mask

During the rework tide, crowds of people in public places such as stations, airports, office buildings, etc., how to monitor abnormal body temperature has become a major problem.

Usually everybody enters and exits the community, and there will be doormen one by one using a temperature gun to measure, but in a place with a lot of people, this approach is not realistic, so AI companies have begun to show their magical powers and use infrared thermometers instead.

The infrared thermometer uses infrared thermal imaging cameras to perform non-contact long-range temperature measurement on people within a range of more than 3 meters, which can measure 100 people at the same time.

If someone in the crowd exceeds 37.3 ℃, the thermometer can quickly identify and prompt the staff to perform a secondary measurement review.

How can we label the temperature of each person in the temperature group?

Xie Yinan, vice president of Mengshi told Rancaijing, Mengshi has been doing human detection in AI recognition. This time they combined ordinary optical cameras and infrared temperature-sensing cameras into one. The modified hardware can solve this problem.

In addition, in winter, people have very little exposed skin and wear masks. How to find an effective temperature measurement site is another challenge.

The method of defiance is to scan the mask first, and then quickly position the forehead through the mask.

"Now everyone is rushing to launch infrared temperature measurement. The market demand is large, and the response speed has become an issue. In addition, the accuracy of temperature measurement will be affected by many factors, especially in stations with a large number of people, Where the temperature difference between indoor and outdoor is large and the environment is complicated, there will inevitably be temperature fluctuations. "Said Wang Yaheng, general manager of Xi'an Youqun Information Technology Co., Ltd.

In order to ensure the accuracy of temperature measurement, a key piece of hardware called blackbody is currently in short supply.

"In the past, this product was generally produced on a small scale, and the market demand was not so large. The sudden outbreak of the epidemic put forward extremely high requirements for the response speed of blackbody manufacturers. AI companies face this strange supply chain, How to quickly achieve large-scale mass production has also become a top priority. "Wang Yaheng said.

In this regard, contempt has introduced a new algorithm, even if there is no black body, it can ensure that the error of human body temperature measurement is ± 0.3 ° C, which provides new possibilities for the industry.

There is also a special case in which everyone is currently wearing a mask and hat. For resuming employees, wearing a mask can easily achieve face punching and temperature measurement, which has become a new need.

For the scene of office access control, Shangtang Technology has developed the SenseThunder-E Vulcan Thermal Camera + SensePass Pro non-passing temperature measurement solution, which automatically intercepts abnormal body temperature and those without masks.

Some start-up companies also spotted the opportunity to enter the market. The Pomegranate Group information technology epidemic prevention and temperature prediction system targeted the relevant departments of the Shangrao Municipal Government in Jiangxi and produced an "face recognition + temperature monitoring + cloud analysis" integrated system.

"Now all walks of life have a need to increase the epidemic prevention function. For example, schools will increase the number of gates containing temperature measurement plates, and they must be rebuilt before school starts." Wang Yaheng told Rancaijing that during the epidemic period, some companies had product manufacturing capabilities, production capabilities, Companies in the sales channel have begun to actively seek cooperation with AI companies to add new functional points such as temperature measurement, which is a good thing for the industry.

New demands mean new opportunities. "This epidemic tells the AI industry that it is necessary to calm down to practice internal skills, such as whether your AI algorithm can be connected to other mature IoT devices such as cameras, switch sensors, access control, gates, etc. Change.

Product first, I think this is the first thing AI companies need to learn during this epidemic. Xie Yinan said.

AI tracking, restore the trajectory of confirmed patients

When you enter or leave the community or office building, you must fill in your personal information on the system every day. This is also inseparable from the intelligent service of AI.

It is understood that the "smart outbound platform" of technology companies such as Baidu, HKUST, Alibaba Cloud, and JD.com's intelligent outbound robots have all been applied in the anti-epidemic process.

In addition, Jingdong Mathematics also launched an epidemic inquiry robot, which can ask suspect patients and analyze whether the other party needs isolation treatment.

After the user scans and logs in, he can also query the information of the hot clinics and designated hospitals, and he can buy medicines in the "online pharmacy".

"However, after this epidemic, I also found that the team's ability still needs to be improved. For example, the development of the first version of the inquiring epidemic robot took only three days, and there will be a large number of users to ask questions every day. Some questions It will surpass the original database, which also inspires us to continue to iterate the product and improve the knowledge base. "Said Bo Liefeng, chief scientist of JD.com.

He added, "In addition, when we configure the outbound robot, we actually need the manual participation of community workers. The team also needs to build a more complete knowledge base that can continuously iterate to improve automation. Degree."

Fortunately, after the epidemic has passed, outbound robots can also play a larger role in various scenes where people are dense. For example, after the logistics and distribution personnel arrive, they can use the outbound robot to confirm that the items have been delivered.

In the epidemic situation, how to quickly restore the trajectory of diagnosed patients is an urgent issue.

JD.com provides a situation analysis and early warning system for high-risk groups of the epidemic, helping the government analyze the course of action of residents, providing early warning and investigation to high-risk groups.

The system can analyze the behavioral trajectory of diagnosed patients and help isolate potential close contacts.

"Governments across the country want to know where the people returning from the epidemic area are located and how to perform efficient investigations. It is almost impossible to investigate one by one manually. Spatio-temporal data always changes dynamically, which is related to the data of telecommunications operators, government data and enterprises. The problem of data connection. We provided this system to 18 provinces and cities, and Beijing found 500 close contacts. "Zheng Yu, vice president of Jingdong Group, chief data scientist of Jingdong Digital Technology, and president of Jingdong City, said the patient's trajectory was confirmed. It is the most urgent and tricky demand, and the system often counts twice a day.

Ordinary citizens can report their health status and travel records through the community epidemic intelligent control platform, and use the generated two-dimensional code to enter and exit the community.

"For the community control platform, we have landed in Yulin and Yizhuang. Yizhuang now has more than 40,000 households using it every day, saving more than 40,000 sheets of paper and 40,000 minutes to fill out forms." Zheng Yu Called.

Drones and AGVs will be applied to more fields

Since the outbreak, "no one +" has become a high-frequency keyword.

During the Spring Festival, a video of a village cadre shouting was widely circulated in the circle of friends, but what many people don't know is that the megaphone actually has a drone version.

According to DJI, DJI's "MAVIC2" dual-light version of the industry model can be equipped with megaphones and thermal imagers, integrating megaphones, thermal imaging temperature measurement, and spraying disinfection water at high altitude.

On February 4, DJI Agriculture launched the "Jiangsu Army Epidemic" and established a 10 million war epidemic fund to support the elimination of the epidemic situation in various places. So far, DJI agricultural plant protection drones have killed more than 400 million square meters in thousands of villages across the country.

However, it should be reminded that this type of product requires professional control.

DJI said that in addition to spraying and killing epidemic prevention, DJI has also provided drone solutions in patrol, quarantine, epidemic prevention and material delivery scenarios.

In addition to drones, unmanned vehicles are also used for the inbound and outbound storage and distribution of medical supplies, and for the material selection and production line handling of protective equipment manufacturers.

"We have a customer who is based in Wuhan and is responsible for pharmaceutical distribution in Hubei. During the epidemic, we specifically provided him with technical support to prevent the surge in orders and lead to system instability." Sun Yu, co-founder of Kuaicang Intelligent, told Rancaijing.

But Sun Yu said that the warehouse has a good effect. The last mile of the open environment is indeed difficult. There are still many challenges in the normal operation of unmanned vehicles, mainly due to the complexity of the environment and weather, and the uncertainty of legal risks.

In the transportation sector, he is more optimistic about the product form of unmanned trucks (unmanned container trucks), especially for unmanned trucks in trunk logistics.

This epidemic has made the "Automated Guided Vehicle (AGV) +" composite product more widely used.

The Guangdong Provincial People's Hospital has introduced twin robots "Pingping" and "An'an" to distribute meals and medicines in isolation wards and infection departments, and recover quilts and medical waste.

Later, various types of disinfection robots, inspection robots, temperature measurement robots, etc. appeared, all based on the innovation of AGV, so as to meet the application of various scenarios.

AI algorithm shortens analysis time and helps to develop new drugs and vaccines

During the new crown pneumonia epidemic, companies such as Baidu, Alibaba, Shangtang, etc. adopted open core algorithms and computing capabilities to fight the epidemic with medical research institutions. It is mainly used for viral gene sequencing, new crown pneumonia new drug and vaccine research and development, and protein screening. .

These are realized in the laboratory through computerized comparison and calculation. Using artificial intelligence algorithms, the original hours of genetic analysis of suspected cases can be shortened to half an hour, and the mutation of the virus can be accurately detected to gain valuable time.

According to Xiong Weiming, a partner of Huachuang Capital, the research and development of a drug usually takes about 1 billion US dollars and 10 years. With the help of AI, this time may be greatly shortened. On the other hand, AI Pharma is not a hot market, and it takes time to start a business in this field.

He introduced that in foreign countries, companies such as Schrodinger have been in business for 30 years, and the company's total revenue in 2018 was only 60 million US dollars. In contrast, black diamonds were listed less than six years after their establishment, but currently So far, only two pre-clinical drugs have been made.

So whether AI technology can achieve the results you expect, the entire industry is still waiting to see.

Using artificial intelligence to help new drug development is no longer a new topic. As early as October 1981, the cover of Fortune Magazine already reported that Merck was using computers to help improve the efficiency of new drug development.

But despite this, there is currently no drug successfully developed by AI.

"Drug research and development is a very long process, and the cost is huge. In this process, AI can currently play a role in accelerating the selection of compounds, but after clinical, the effect is actually limited." Li Ming, executive director of the Chinese Academy of Sciences Capital Called.

For pharmaceutical companies, the ultimate value of the company depends on conditions such as the pipeline of new drugs, preclinical conditions and target selection.

Therefore, from the perspective of entrepreneurship, Xiong Weiming believes that it should be advantageous to cut in from the upstream and downstream of the pharmaceutical industry and its surroundings. If startups have both AI capabilities and an understanding of the pharmaceutical industry, there will be great opportunities.

Conclusion

The epidemic will eventually pass, will the demand for AI in various fields gradually fade away?

"The penetration of AI is in stages. At present, most startups are focused on image and voice semantics. Customers in these two areas are concentrated in the security and financial fields. Demand is rigid and the ability to pay is relatively strong. Now AI is also gradually infiltrating. To the industrial sector, but where AI companies should go next, the industry is still watching. "Said Li Ming, executive director of the Chinese Academy of Sciences Capital.

Li Ming believes that overall, under the epidemic, the response speed of various AI companies is very fast, but now there are many products in the upstream supply chain that have no value and coupled with the pressure to resume work, AI companies will be under great pressure .

After the epidemic has passed, some scenes will eventually settle down. For example, the thermometer at the Capital Airport was used during SARS, and the security check for the Beijing subway was used during the Olympic Games. This epidemic will also accumulate usage habits in some scenarios.

"But in retrospect, the new features and products in these scenarios are not the most critical thing in determining the development of AI companies. The epidemic will be a major reshuffle for all industries, companies with backward models, single products, and unreasonable financial conditions. Will be eliminated. This epidemic is a catalyst, but companies can't develop with catalysts alone. The big wave of sand and sand is the test for small and medium-sized AI companies. "Li Ming said.

Special Report: Fighting The New Coronavirus