In the investor Marco's impression, when he looked at the project in the medical AI imaging field in his early years, many companies were valued at less than 100 million yuan at that time, but now some companies in the industry have been valued at a crazy level of 1.5 billion US dollars after the latest round of financing.

In the twinkling of an eye, the growth cycle of the industry is no more than 3-5 years, and now it is only 5 years since the establishment of head enterprises such as Healthcare and Keya Healthcare, which have been established for a long time. The enthusiasm of investors and the rise in valuation undoubtedly come from optimistic capital expectations-the listing tide is just around the corner. At present, the "four Little Dragons of Medical AI"-Keya Medical, speculative Medical, Airdoc and Jukun Technology have submitted prospectuses to the Hong Kong Stock Exchange, of which Airdoc has passed the hearing and may be listed in the near future.

Behind the listing tide, after years of pioneering, players finally saw "real gold and silver" for the first time, and the revenue of many companies this year is expected to reach 100 million yuan.However, the tide of listing is just around the corner, and more fierce commercial competition has come. If revenue only doubles, it will inevitably fall behind in the competition.

Commercial war begins with a certificate.

The business wars and listings of the AI medical industry begin with a certificate."on the day we got the license, the phone was blown up, and more than 20 organizations called and wanted to vote, so we had four rounds a year. "Keya insider Liu Dong (a pseudonym) said.

On January 15, 2020, the State Drug Administration approved the first medical AI category III device product, which is a non-invasive CT-FFR product developed by Keya Medical. Since then, the significant benefits brought by product approval, coupled with the booming health and medical demand brought about by the epidemic, have made the industry pick up quickly and become a tuyere again.

Sequoia Capital China, IDG Capital and other established venture capital institutions have stepped in one after another, and even Shen Nanpeng personally came forward to dominate the investment in a certain project. At this time, one of the core indicators of the competition of the target enterprise is whether it has got the "III certificate", which is the admission ticket of the industry. Because for the medical AI industry, a long-term bottleneck is the product review and approval.

In July 2019, the Medical device Technical Evaluation Center of the State Drug Administration issued the key points for the Evaluation of Medical device Software for Deep Learning decision-making, which clarified the key points for the registration and approval of medical device software for deep learning decision-making. that is, data quality control, algorithm generalization ability and clinical use risk. Among them, the risk of clinical use of the software mainly includes: false negative, that is, missed diagnosis, which may lead to delays in follow-up diagnosis and treatment activities, especially for diseases with rapid development of the disease; false positive, that is, misdiagnosis, may lead to unnecessary follow-up diagnosis and treatment activities.

总In terms of regulatory review and approval, one of the directions is that the landed product must be able to withstand the consideration of hospitals, doctors, and patients.From Keya Medical's CT-FFR products, as of the end of August, NMPA has approved 16 III certificates in more than a year, covering brain, chest, lungs, bones and other diseases. Among them, Keya Medical has 1 model, presumed Medical has 2 models, Airdoc has 1 model, and Dikun Technology has 1 model.

The third type of AI medical devices approved by the State Drug Administration in the past, the examination and approval was blocked, which meant that the early investment was separated from the later cash return; products without the certification of three types of devices could only be used by the hospital free of charge and could not enter the hospital's charging catalogue, which directly meant that it was difficult to get through the commercial closed loop. With the ice-breaking of product review and approval, the curtain of commercialization of medical AI is also opened.

Before this, each player can only generate a small amount of scientific research income in the scientific research cooperation with the hospital. In 2019, according to the data of the prospectus, it is assumed that the income of Medical Care, Keya Medical and Sikun Science and Technology is only 6.6 million yuan, 1.2 million yuan and 800000 yuan respectively, which is nothing compared with the R & D investment of nearly 100 million yuan. Since 2020, the income of each family has ushered in an explosive growth.

The income of Medical AI Enterprises according to the prospectus, it is assumed that Healthcare recorded an income of 22.1 million yuan in the first quarter of this year, while Jukun Technology recorded an income of 52.6 million yuan in the first half of this year. This year, it is assumed that medical care can almost catch up with the whole of last year in just one quarter, and the revenue data of Jukun Technology in the first half of this year has more than doubled that of the whole of last year.

In addition, 36Kr Holdings learned from a number of authoritative sources: Airdoc's income in July this year has exceeded that of the whole of last year, and Keya Medical has made tens of millions of yuan in the first half of this year. For the whole of 2021, it is assumed that the internal income forecasts of Healthcare, Sikun Technology and Airdoc this year will all be 100 million yuan.In the long run, the whole medical AI market is still in its infancy.

According to the Frost Sullivan report, the market size of China's AI medical devices is about 300 million yuan in 2020, while the market size is expected to be about 500 million yuan in 2021, will reach 20.4 billion yuan in 2025, and the annual compound growth rate of the industry from 2020 to 2025 will be 134%. Globally, the market size of AI medical devices is expected to reach $17.7 billion in 2025, with a compound annual growth rate of 119% from 2020 to 2025. From then on, commercialization has become the absolute core proposition of the current medical AI device industry.

At present, all players are rubbing their hands and commercializing in an all-round way, trying to win the first place in the current "commercialization war".With the submission of various enterprises to be listed, enterprises not only need to have a considerable financial statements to win the trust of investors, but also need to establish the ability of self-hematopoiesis to bring the development of enterprises to a benign track.

This reflects the organizational structure of the enterprise, that is to vigorously expand the sales team and establish their own sales network. In the actual business, medical AI enterprises mainly choose to cooperate with channel dealers, because in the current medical system, there are differences in different regions, even different hospitals in the same region, and there is a certain basis of customer relationship between channel dealers and local hospitals, so the cost performance is much higher than that of self-built through its own sales team to manage local channel dealers. According to the prospectus, it is speculated that Tech currently has 75 sales staff, 242JK sales team and 87 Airdoc sales staff.

In addition, 36Kr Holdings learned that the expansion of the sales force of Keya Medical is even more intense. After Wang Liwei became chairman of Keya Medical, the sales force of Keya Medical increased to hundreds of people in two months.

According to people familiar with the matter, the large-scale expansion of the sales force is essentially to open up channels for products to enter the hospital. "like the idea of doing the Internet, the company will first hold the channel in hand, and then a series of products will be able to enter through this channel. From the point of view of customer expense, the price-to-performance ratio of 5 products is always higher than that of 1 product. "

At present, there are three pricing models for medical AI products entering the hospital: pay a fixed fee when entering the hospital, which is the current common pricing model; subscription fee model, in which the hospital pays the subscription fee on a regular basis to obtain the right to use the software; according to the actual use, charge on a case-by-case basis. Outside the core scene of the hospital, each player also wades out other ways of commercialization according to his own product pipeline.

Medical AI volume

In the early entrepreneurial process, all the enterprises cut in from a core product, such as the lung nodule product of presumptive medical treatment and the CT-FFR product of Keya Medical, but at present, each enterprise has moved from single-site to multi-site and multi-disease, enriching the layout of the product line. According to the prospectus, it is assumed that there are more than 10 product lines in Healthcare, Keya Medical and so on.

Technically, for various head enterprises, with the accumulation of technology and data in the past, the marginal cost of developing other pipelines is constantly being pulled down. In the view of Liu Yu, a veteran in the industry, "there is basically no technical threshold for expanding product pipelines, and algorithm modules are all ready-made, that is, to do some new labeling and three-dimensional reconstruction. You want to rebuild a" heart ", rebuild a" lung ", rebuild a" liver ". In addition to serving the CT-FFR of the Department of Cardiology and the artificial Intelligence diagnosis system of hemorrhagic Stroke in the Department of Neurology, Keya Medical has obtained the certification of FDA in the United States and is guided by clinical application. The products cover the whole process of screening, diagnosis, treatment and follow-up. There are 11 products in total, such as intelligent heart, intelligent brain, intelligent imaging and so on. Jukun not only does cardio-cerebrovascular and pulmonary nodules, but also begins the research and development of abdominal liver examination to try to find lesions such as liver cirrhosis and liver cysts. Presumably, on the basis of the leading chest and lung products, medical treatment has also horizontally expanded the field of cardio-cerebrovascular diseases. Cancer and infection, trauma, major diseases, and in the vertical into the clinical field of chest and lung surgery.

The impact of the layout of product pipelines on commercialization is that in addition to differences in core products, various enterprises have fallen into homogenized competition on many products, which inevitably causes some people in the industry to sigh, "Medical AI is badly rolled up." 36Kr Holdings learned through multiple interviews that at present, the commercialization of medical AI mainly includes in-hospital and out-of-hospital scenarios and the following five directions: 2H (Hospital), 2G (Government), 2B (Business), 2P (Pharma & Medtech).Each company competes for business opportunities on the subdivision track based on its own core products (mainly those that have been approved with the III certificate).

1）2H

Medical imaging data account for more than 80% of all clinical data, which is the cornerstone of clinical diagnosis, disease treatment and health management. In the actual scene of the hospital, the analysis of medical images is complex and time-consuming, because it requires doctors to combine multi-layer image data with fine decision-making process. Due to the uneven distribution of medical resources in China, the analysis of medical images by doctors with insufficient experience and equipment in many areas is not accurate and inefficient. The development of artificial intelligence technology is expected to bring solutions.

In the early days of the industry, medical AI enterprises mainly wanted to solve these actual pain points in hospital clinical scenes, such as presuming that medical treatment adhered to the R & D idea of "coming from clinic to going to clinic", and polished algorithms and products with hospitals in the name of scientific research cooperation. With the maturity of the product, the product naturally wants to enter the actual clinical scene of the hospital for use by the imaging department, radiology department and other departments. At present, the hospital scene is the most mature scene for the commercialization of medical AI, and of course, it is also the most competitive scene at present.

In the industry, there are generally three forms for products to enter hospitals: the first is for enterprises to cooperate with doctors through cooperative distributors to guide patients to use medical AI products through the authority of doctors; the second is to package medical AI products into the large-scale plan of hospital informatization, put together orders with other informatization plans, and enter the hospital through the bidding and purchasing process of the hospital. The third is that the product enters the hospital's charging catalogue through the process of taking the price code and so on. The first form and the second form are extremely dependent on customer relationship, because from the point of view of the hospital, from the point of view of product performance, each family takes three types of certificates, and the numerical differences in specificity and sensitivity are very small.

Under such circumstances, what the product enters the hospital to compete is the customer relationship and the product price.Some people in the industry revealed to 36Kr Holdings that enterprises with similar product pipelines, such as Jukun Technology and Shenrui Medical, are drastically involved in the sales link.According to the prospectus, Jukun's products have entered 1200 domestic hospitals, and Keya has established a cooperative relationship with more than 1000 domestic hospitals. This data includes free use of hospitals, and future payment conversion is the focus. The presumed products have covered nearly 20 countries around the world. The third form is the one that can really make medical AI products mature in commercialization. CT-FFR (Deep vein score), the core product of Keya Healthcare, is exploring in this direction.

2）2C

Although Koya's products are also admitted to the hospital, for the hospital, it is an additional fee item, which in essence is paid by the patients. Koya's deep vein score can perform a non-invasive physiological evaluation of the coronary artery only based on the patient's computed tomography angiography (CTA) images to accurately diagnose whether the patient has myocardial ischemia. In the past, under the path of diagnosis and treatment of patients with coronary heart disease, to judge the myocardial blood supply, patients had to spend thousands of yuan on radiography, and then spent more than 10,000 yuan on guide wire.Koya's products can significantly save medical costs, and patients only need to spend more than 3000 yuan to avoid unnecessary invasive coronary angiography (ICA) tests or percutaneous coronary intervention (PCI) surgery.If the product wants to enter the hospital charging catalogue, it needs to go through the link of taking the price.

"for every price we take, we take it according to the province, and we approve it by province by province. Then, the window period of each province is actually different. You don't have it all the time. Basically, each province has only two window periods a year. After the province got the price code, he had to announce it for almost a month before issuing it layer by layer. Then the next step is to sign a contract with the hospital through the dealer before you can finally see the price code in the hospital. Liu Dong revealed that it will take almost nine months for the whole process to come down.

According to Liu Dong, Keya is making a comprehensive price declaration for deep vein score products. "prices have come out in seven provinces, realizing the charge to patients." Prices in another 14 provinces are progressing smoothly. The specific pricing of products varies from province to province, and the marketing model of Keya Medical is also based on the actual situation of provinces and hospitals.

3）2G

It is assumed that the approved III products of pulmonary nodules (InferRead CT Lung) and pneumonia (InferRead CT Pneumonia) have found a good way of commercialization of To G at present. Because for the major disease of cancer, if we can detect and intervene in the development of the disease at an early stage, it will help to achieve better treatment results and significantly reduce the overall medical costs.

For example, when cancer is first diagnosed at an earlier stage, the cost of treatment for cancer is generally lower. In China, the cost of lifetime direct treatment for patients with lung cancer diagnosed for the first time in the early stage is estimated at US $9200, while the cost of lifetime direct treatment for patients with lung cancer at the advanced stage of the first diagnosis has almost doubled (US $17800).

The use of medical AI products for large-scale screening produces value expectations that reduce overall public health costs, and the government is expected to be the payer.Recently, the TB AI solution for medical treatment has been included in the United Nations Global Purchasing list of Tuberculosis Medical equipment (referred to as "GDF"). This means that in the future potential purchasers (mainly global government sovereign fund procurement, aid fund procurement and international organizations) will be able to place orders directly for presumed medical TB diagnostic capabilities. It is assumed that the selected products will help health care workers to screen in remote areas (X-ray + AI), which is very valuable in areas where medical conditions are poor.

Not only the United Nations equipment purchase list, it is speculated that medical pneumonia (InferRead CT Pneumonia) products have also been purchased by the European Union government and Japanese government many times during the anti-epidemic period, taking the lead in going out of the way of medical AI enterprises to go to sea. In China, it is presumed that Healthcare has also won a number of government orders.

In addition to speculation, Airdoc has also won orders from the government. For example, in the cooperation project between Airdoc and Qingdao, Airdoc received more than 1 million yuan after the completion of the first batch of 150000 fundus screening, and then it will generate corresponding income according to different stages of the project. At present, the direction of To G is mainly to serve the high-throughput screening of local governments, and the core products of inference and Airdoc are naturally suitable for such scenarios. According to the prospectus, 22291 significant positive users were found in Airdoc retinal screening in 2020.

In the view of Zhao Yang, a partner for Lai Capital, the most important thing for technology is to find a suitable application scenario, and at present To G is the most robust commercial direction of medical AI.In the medical field, the government will certainly make continuous investment in disease screening, epidemic prevention and control, and so on. "if enterprises with solutions can be close to the needs of the government and establish a good cooperative relationship with the government, then every government order will become a strong endorsement for subsequent bidding, and it will be a very good asset. "

4）2B

Airdoc, which has passed the listing hearing on the Hong Kong Stock Exchange, is expected to become the first share of medical AI. Airdoc has been approved III certification product AIFUNDUS can use retinal images and other data to accurately detect and assist the diagnosis of chronic diseases, specifically, the Airdoc-AIFUNDUS workflow includes three main steps: retinal image acquisition, image quality control and image analysis and classification. After the analysis is completed, a report with retinal images, test results, disease progression and referral recommendations will be generated. Doctors can refer to the report when providing diagnosis and medical advice. Doctors can also make reference to the report to decide whether it is necessary to refer to other departments (such as ophthalmology) for follow-up consultation.

The highlight of the company is that the earlier commercialized enterprises in the industry have made about 30 million yuan in revenue in 2019, while several other enterprises that have submitted their forms, such as Jukun, have an annual income of only one million yuan. In addition to hospitals, Airdoc has found the important commercial direction of out-of-hospital To B, cutting into the supply chain to cooperate and sell with various parties in the supply chain, including medical examination centers, community clinics, insurance companies, optometry centers, pharmacies, etc. For example, in the cooperation between Airdoc and Aikang Guobin, it follows the principle of paying per person, and the two parties share according to a certain proportion. In addition to Airdoc, Keya Medical is also exploring the commercialization of this direction and is discussing relevant cooperation with some insurance companies.

5) 2P

This direction is a business opportunity for medical AI products to explore the value chain of medical devices.

The cooperation between GPS and medical AI enterprises Jukun Technology was the first to make real money and silver income in this direction and sold its products to medical device manufacturers. According to people familiar with the matter, in 2020, Sikun Technology reached a deep cooperation with Philips, a giant manufacturer of equipment, and its products were sold in hospitals with Philips CT machines. The price of medical AI products purchased by hospitals is about hundreds of thousands of yuan. In the deep cooperation with Philips, Sikun Technology has made tens of millions of yuan in sales revenue.

In addition, in the whole income structure of Jukun, the commercial income of equipment manufacturers accounts for about 50%.Subsequently, Keya Medical also made efforts in this direction, and also made an income of more than 10 million yuan. At present, medical quasi-intelligence and other enterprises are also following in this direction.In July 2021, GE Healthcare released Edison Magic Box, an one-stop, cross-brand integrated AI application platform, trying to solve the problem of unified port access in the industry and provide one-stop AI services for hospital users.

On the list of cooperation announced at that time, there were seven enterprises: Shenrui Medical, speculation Medical, Keya Medical, Medical quasi-Intelligence, Jukun Science and Technology, Ande Medical Intelligence, and Qiang Lian Zhichuang. At the commercial level, medical AI is actually equivalent to the supplier of device manufacturers. Dai Ying, vice president of GE Medical China, told 36Kr Holdings that at present, the form of cooperation with medical AI enterprises is mainly an one-off buyout, but they want to go to the form of pay-per-view in the future. Generally speaking, medical AI has been commercialized in an all-round way, and players not only continue to compete for in-hospital scenes, but also try to find more commercialized out-of-hospital scenes based on the uniqueness of their own products.

With the rise of giants, the road lies ahead.

In five or six years, the medical AI industry has experienced competitive competition and then to the big waves to clean up the sand, and now the industry is moving towards a new stage of the rise of giants. The road taken by the head enterprises quietly constitutes the current barriers and competitive advantages. If it goes well, the four companies that have submitted their forms this year may all log on to the Hong Kong Stock Exchange, and with the realization of the listing tide, the entire medical AI imaging track will also move towards a broader market and get more attention. Jukun Technology, which started by husband and wife, has reached a high valuation of US $1.5 billion in just four years when the company raised 700 million yuan in the last round of financing before handing over the prospectus.

It is assumed that the valuations of Healthcare, Airdoc and Keya Healthcare are not far behind, of which the valuation of Healthcare is estimated to reach 1 billion US dollars. Although the upstream algorithms are open source, for new entrants, they will face two barriers if they want to catch up with the real leaders in industries such as imaginary healthcare and Keya healthcare. First of all, in terms of data, high-quality data is essential to ensure the safety and effectiveness of AI medical products.

In general, the more data, the better the performance of the model. In addition, the clinical data of images need to be classified and labeled by medical experts to train deep learning algorithms. At present, head enterprises have accumulated a large number of high-quality clinical data, and can also verify and improve the performance of their AI medical products through efficient algorithm training. For example, Airdoc disclosed that it already has 3.7 million retina image data of real-world users.This clinical verification through high-quality data is the core entry threshold for new market participants.

The second is in the landing and commercialization of products. Once the hospital recognizes the clinical advantages of a company's product, it is likely to insist on using the product, and for doctors, it is easy to produce stickiness in the use of a certain product in a large amount of work every day, and then generate awareness of the medical AI corporate brand. As a result, manufacturers that have established hospital channels, especially top hospitals, will enjoy a strong first-mover advantage and may further consolidate their dominant position by expanding their AI healthcare portfolio.

For new entrants, they will have to pay a great deal of business, marketing and time costs before they are expected to establish a broad hospital network. In the face of these two major barriers, for new entrants, a direct realistic question is, how much money should be spent in the market in order to recreate an assumption and a Koya? It may become the norm for startups in this field to cut from the subdivision scenarios that AI has not yet covered. But at the same time, this does not mean that the head company can rest easy. A more fierce business war may still be after the listing.

A chairman of AI, who spoke on condition of anonymity, said, "if everyone stays at selling software modules, it means that all companies are robbing these 1000 public hospitals. After these 1000 hospitals are full, the market will soon become a red sea." This means that if we want to, we will rely on our core products to make money in the future, and then other products will be used to hit competitors and drive down market prices. Therefore, it is really beneficial for the industry to stock up the future product line now. "

In Zhao Yang's view, the mainstream commercial direction of selling to hospitals is not sustainable. "at a glance, AI medical products will only be a charging item for hospitals in the future, and at that time, with the expansion of the scale of use, there is a high probability that they will be purchased with volume. "the wider application of AI medical products is in out-of-hospital scenarios, such as helping B-end enterprises to promote sales, establishing professional service capabilities for users, doing out-of-hospital follow-up, health management of sub-healthy people, and so on.

With the coming wave of listings, the focus outside the industry is on the wealth story of this track-what kind of price can the secondary market offer? This may be directly related to the value of the entrepreneurs and investors of the entire track.But people who are in the middle of the track are more rational, and a new industry naturally has its difficulties.

Presumably Medical founder Chen Kuan talked about his entrepreneurial feelings in this industry in an interview with 36Kr Holdings: "along the road of AI Healthcare in the past few years, we have experienced many difficult times, such as product review, certification, commercialization, and so on, and it is not easy to keep going, but once all this has been realized, it seems to have become a natural experience. "Looking back at the trend of listing, many players have not been able to hold on to it so far.

In this year's financing boom, such as 900 million yuan in health care financing, 700 million yuan in Jukun technology financing, and hundreds of millions of yuan in Shenrui Medical and quasi-intelligent financing, companies such as Yitu Medical, Tuma Shenwei, voxel Technology and other enterprises that were once famous were no longer heard. Zhang Dalei, chairman of Airdoc, revealed to 36Kr Holdings that during the last round of financing, some financial investors had estimated the investment intention of 30 billion yuan to Airdoc, but Zhang Dalei chose to refuse."in my understanding, the valuation should match the team, the development of the company, the industry, and so on," he said. "

At present, what is more critical than valuation is how much value artificial intelligence has created for the medical industry, how many positive patients have been screened, how many doctors' needs have been met, and what kind of health monitoring network has been established for users. From a longer-term time perspective, AI has only solved one problem in the specific aspects of medical care, from early screening, to clinical diagnosis, diagnosis and treatment, and then to chronic disease management and health management in the later stage, AI medical treatment has a long way to go.