Production | Tiger Sniff Technology Group

Author | 石 晗 旭

Cover | Visual China

Spring is here.

The number of new and suspected cases of new coronary pneumonia in China has been declining, and the number of cures has been rising. The official closing of 16 square cabin hospitals also brought Wuhan one step closer to getting rid of the worst-hit areas.

As a member of the frontline medical care army, Xu Liangzhou, director of the imaging department of Wuhan Traditional Chinese Medicine Hospital, felt this trend firsthand-

Wuhan Traditional Chinese Medicine Hospital as the designated hospital, the number of chest CT consultations per day has now dropped to 1/3 of the peak of the epidemic, and most of them are re-examinations of patients.

During the outbreak, Xu Liangzhou and the 4 doctors in the department needed to see more than 400 chest CT films and write more than 400 reports every day. The workload of chest CT was almost 8 times as usual. For each CT chest radiograph, there are hundreds of frames of cross-sections that need to be screened.

Writing a report until the early hours of the morning is, for them, the norm in this Spring Festival.

Because CT imaging has a great effect on the diagnosis and treatment of new coronary pneumonia, other designated hospitals are also facing the overload operation of imaging department. In the imaging department of Jinyintan Hospital and Wuhan Central Hospital, a single-day CT admission can even reach 1,000 people.

“I told myself not to cry, but to use my eyes to watch the film.” This is what Fan Yanqing, director of the radiology department at Jinyintan Hospital, told himself at all times. For more than a month, she took 21 people in the team to watch nearly 2 million chest radiographs.

This gives medical AI companies that haven’t spoken for a long time a chance to correct their names.

According to the incomplete statistics of Tiger Sniff, more than 20 companies have successively completed the development of AI + CT new coronary pneumonia auxiliary diagnosis systems, such as Alibaba Hospital, Neusoft Medical, Shukun Technology, Huiyihuiying, Inferred Technology .

Although these products have different forms of expression, their common effect is to improve the efficiency of the imaging doctor: The hours that ordinary doctors need to read movies can be compressed by AI doctors to minutes, 20 seconds, or even 2 ~ 3 seconds.

In the case of a surge in workload, designated hospitals across the country have used at least one AI + CT auxiliary diagnostic system, and the utilization rate is extremely high.

However, once the needs brought by the epidemic are out of the way, will these systems, like AI images of lung nodules, be covered in dust in the hospital and once again be reduced to the tasteless and unfortunate situation of eating?

Frontline Imaging Division at Full Load

On January 20th, the twenty-sixth lunar month of the lunar calendar, four days before the grandest festival in China. On television, Zhong Nanshan is telling everyone with certainty that it is now certain that human-to-human transmission of new coronary pneumonia exists.

Next, the whole city of Wuhan suddenly lost its voice, swept away from the past, and shrank silently into the continuous rainy days.

The most crowded places in the city have changed from Optical Valley Subway Station, Jianghan Road Pedestrian Street, etc. into outpatient clinics of various hospitals. By convention, doctors recommend that those with severe cough go to the imaging department for a chest CT.

What Xu Liangzhou didn’t expect was that so many people would suddenly come.

He is in day shift with the two doctors in the department. Before and after New Year’s Eve in previous years, after the colleagues of the night shift took over at 5 pm, they could leave work on time. However, these days, the day shift doctors can only get off work after 8 o’clock, and the night shift staff basically work overnight.

In the past ten hours, the CT machine continuously scanned the chests of nearly 400 patients, which was 8 times the usual amount. This means that they average everyoneLook at nearly 100 chest radiographs and send out nearly 100 reports to the clinician for diagnosis.

Although unlike nucleic acid testing, which can be used as the gold standard, CT can be the first pass to screen patients for pneumonia and what type of pneumonia.

In the following days, in order to cope with the new coronary pneumonia screening, Wuhan Traditional Chinese Medicine Hospital stopped shooting magnetic resonance imaging and radiography (X-ray), and received CT patients at full load.

A chest CT will generate several hundred frames of images. Doctors need to draw conclusions by observing the volume and density of lung lesions. This kind of quantitative analysis often takes more than 2 hours for the doctor.

Even in the evening of the 30th year of the New Year, there were more than 200 patients on and off the CT machine, and the two night shift doctors did not have time to sleep at all. After writing these new film reports, it was already bright.

On January 27, Wuhan Traditional Chinese Medicine Hospital was listed as the new crown treatment designated hospital. The more patients admitted to the hospital, the busier the imaging doctor is—not only do they have to film suspected cases, they also need to review the inpatients every three or five days and compare and analyze the treatment effect.

The situation of Jinyintan Hospital and Wuhan Central Hospital being designated hospitals is also very serious. The peak number of consultations once reached 1,000.

Even when two shifts are in operation, writing a report until two or three in the morning is normal. The imaging doctor was exhausted.

AI admission to watch movies

In this special period, medical AI images that have been gathered in the field of lung nodules but have not performed well on the ground have suddenly been placed with high hopes. Especially after the National Health Commission of China included CT in the diagnostic criteria for new coronary pneumonia on February 5.

Yitu Medical’s “Chest CT New Coronavirus Pneumonia Intelligent Evaluation System” is the first AI image-assisted diagnostic system in China to be used in the field.

On the 30th of this year, the Shanghai Public Health Clinical Center found Yitu and put forward specific requirements for AI images: participating in quantitative analysis of the lesions and evaluating the efficacy, improving efficiency and accuracy.

From the urgent call for a team of 100 people scattered throughout the country that has entered the holiday mode, the product was launched at the Shanghai Public Health Clinical Center on January 28, and it took 4 days to medically according to the plan. After completing the preliminary clinical verification, the system has successively entered the People’s Hospital of Wuhan University, the Union Hospital of Tongji Medical College of Huazhong University of Science and Technology, and the First People’s Hospital of Jingzhou.Above 90%, more excellent can achieve 96%.

This is reminiscent of the influx of first-generation medical AI images into the diagnosis of lung nodules. According to relevant media statistics, of the more than 140 medical AI companies, nearly 120 are doing medical imaging business, and about 100 of them are assisted in diagnosis of pulmonary nodules.

Especially around 2016, the enthusiasm of entrepreneurship and investment made it an air outlet. From the perspective of the development of AI, image recognition is more mature; while a lot of the work of imaging doctors is dull and boring. The preliminary screening of medical images by AI can help them free up time and focus on solving more difficult cases.

But just two years later, some imaging doctors ridiculed, “Chinese lung nodules are not enough.” It is extremely common to have several companies’ systems deployed in a radiology department, but the actual utilization is not as expected.

First, the sensitivity and specificity of AI cannot be balanced. If the sensitivity is improved, the eyes of the AI ​​will be too sharp, and many results that are not nodules will be misreported. The high false alarm rate will make doctors have to carefully review, and the efficiency improvement is naturally limited.

For medical AI companies, the commercialization of pulmonary nodule diagnosis is not very feasible.

The outbreak of neocrown pneumonia gave AI + CT another good landing scene. Not only in Wuhan, when the country’s epidemic was at its worst, the number of admissions to imaging departments in designated hospitals doubled.

However, as the epidemic is brought under control, the prospects for AI + CT look bewildered. Unless before the end of the epidemic, the third type of medical device certificate (the AI ​​device used to authenticate and identify the lesion and provide a clear diagnosis) is still empty, it is still empty talk to sell products to hospitals or other medical institutions.

However, most medical AI companies are not worried about this. On the one hand, after the fight against lung nodules that year, families have already embarked on the path of differentiated development and opened up new fields in cardiovascular disease and breast cancer. On the other hand, the prevalence of influenza has made mature AI + chest CT still has a lot of room for development in the field of viral pneumonia.

Moreover, once AI + CT can complete the transition from quantitative analysis to qualitative analysis, it will definitely bring greater help to future epidemic prevention and control and scientific research.

Anyway, the imagination space for medical AI is still large. As for commercialization, it’s better to wait until the product is really clear.

I am Shi Yanxu in this article. I am concerned about medical science and technology. Welcome related industry people to find out from me. WeChat: shx0427, be sure to indicate your identity.