In 2018, the approval signal released by the Food and Drug Administration became a "year-end gift" for medical AI players, and with the acceleration of industry reshuffle and the opening of the market environment, the circle of friends of "medical + AI" will become more and more lively.
Early in the morning of December 26, the circle of friends was swept away by a piece of news: at the just-past "public welfare training on the registration and declaration of artificial intelligence medical devices", the Food and Drug Administration gave a detailed explanation of the process, indicators and key points of the approval of medical artificial intelligence devices, and it was said that the approval channel was released in mid-December.
Medical AI has been budding and hot since around 2014, and players have been suffering from the cautious and cautious approval and landing of new technologies at the policy level. This training means that enterprises in the field of medical AI are no longer "headless flies" in applying for approval of three types of devices, and have a clear direction. Ji Xin, founder of Shanggong Medical Trust, sighed: "The approval department began to formulate a process and promote the development of medical AI, which is a major progress for the country and the industry. ”
In Ji Xin's view, the misdiagnosis of medical AI, just like the doctor to learn to accumulate experience is the only way, the startup company does not have to worry too much about the product can not get the approval for a while, but can not be positioned wrong; the worst business model is not "can not make money", but "rush to make money"; the biggest opponent in the future of the startup is not "BAT", but the enterprise itself.
Medical AI companies, do they do technology or platforms?
Should healthcare AI companies focus on technology or positioning platforms? Until then, it needs to be clear whether medical AI-assisted screening technology has enough maturity to support the cost investment of startups.
A few months ago, the internationally renowned "cancer expert" IBM Watson was exposed to the news of "misdiagnosis" and abandoned by partners, causing the market to even begin to doubt the feasibility of medical AI. Some industry insiders have analyzed that on the one hand, this is due to the lack of "Chinese characteristics" in its products, on the other hand, medical AI products are still not fully mature in China.
When the early artificial intelligence technology first began to spread in the medical field, in addition to the doctor's "good assistant" surgical robot, the related products on diabetic retinal screening and lung nodule screening were the hot "stars" in the field of medical AI. "Especially in 2016 and 2017, when due to the public dataset of sugar network screening in the world, the product was well trained, coupled with the open ai algorithm, medical AI players only need to build a model, practice with a standard database, and do a little laboratory stuff, and the product can come out." Ji Xin said frankly.
Ji Xin calls it the first stage of AI-assisted diagnosis, but in the field of sugar network screening, it is far more simple for a product to be called a "clinical product".
A major indicator of the maturity of AI-assisted diagnostic products in the market judgment is "robustness", which is an important step from laboratory products to complex application scenarios. Taking sugar network screening as an example, in China, the camera brands of top three hospitals and grass-roots hospitals are very different, and the internal and external doctors of the same top three hospitals and even different doctors in the same department also take different fundus photos, which puts forward higher requirements for AI diagnostic products for domestic sugar network screening.
In April 2018, the US FDA approved the IDx-DR, the first AI diagnostic device developed by IDx for sugar network screening, which inspired many medical AI troops at the time. But Ji Xin said that IDx-DR only supports the sugar net auxiliary screening of the Tuopukang NW-400 camera. This means that IDx-DR is not "robust" if it is to be used domestically.
The starting point of returning to AI-assisted diagnosis is still as a "compass for doctors to diagnose and treat", "GPS in the medical link", which cannot become a substitute role, and the quality of domestic medical AI products is uneven, and the current intelligence of medical AI is obvious.
In Ji Xin's view, in the future, medical AI companies will be divided into two categories. One is a pure technology enterprise, customers can call or directly purchase their technology on the platform, grafted into the hospital; the other is the technology plus services of technical medical companies including Shanggong Medical Information, for which technology/system is only a tool and an element of the service. In addition, for medical AI products, the industry may also incubate "agent" pure service companies.
"Therefore, Shanggong Medical Information is bound to develop towards the platform, but the core technology and products are the threshold of medical AI startups and cannot be lost." Ji Xin told Yiou Health, "Including the construction platform of the medical association, the ophthalmology management platform, the referral platform, etc., and finally rely on platform services to connect hospitals, governments, enterprises and patients horizontally and vertically." ”
The "pit" of subsidies still has to stand firm
The exploration of "business models" has always been a hot topic of discussion in the medical AI plate. Although some of the faster medical AI companies in China have obtained approval for Class II devices, they are still far from "large-scale realization" and "sustained profitability".
Ji Xin said frankly that subject to the maturity of the market and policies, medical AI companies still have to "burn money" for a long time. At present, it is more worth considering for enterprises: how to "paste the subsidy value"?
Speaking of this topic, it is necessary to return to the application for enterprise product qualifications. Usually, medical AI companies must apply for approval and inspection after polishing out their products, and this link is responsible for by the Central Inspection Institute. However, before that, the Chinese People's Procuratorate did not formulate a gold standard for the "examination" of medical AI auxiliary diagnostic devices, nor did it have a database on single diseases. Since 2017, Shanggong Medical Information and the China Inspection Institute have taken the lead in polishing the "gold standard" and database for screening AI products on sugar networks, which were finalized in May 2018. After passing the approval of the Central Inspection Institute, the enterprise will send the medical AI product to the clinical inspection and approval, and finally package the results and send it to the Food and Drug Administration for approval again, and only after passing can it obtain the Second or Third Class device certificate.
While this process of assessing accuracy and safety, IDx-DR went for seven years. In this process, the data, talent, computing power, and operating costs of medical AI companies are expensive, which extremely tests the patience of startups and investors. "After getting the device certificate, it does not mean that the company's profit window has been opened." Ji Xin said, "In the end, it is necessary to charge from the hospital end, and it is also necessary to pass the medical insurance bureau, through the medical insurance bureau pricing, and then enter the hospital." ”
At present, it is revealed that the test standards and databases established by the Central Inspection Institute on the two major diseases of sugar network screening and lung nodule screening have only been able to take the first few steps by medical AI companies, including Shanggong Medical Information, and are currently preparing relevant standards for glaucoma. "The gold standard and database for other diseases have not been introduced, and startups have not been able to go behind the process." Ji Xin said.
Secondly, startups can not avoid "burning money" in the process of landing. At present, the landing direction of medical AI products is generally to G, to H, to B three. At the government level, including assisting the intelligent review of medical insurance, creating a big data supervision platform, assisting the CDC to integrate information reports, etc.; medical institutions are also the directions that have landed the most at present, playing a role in helping doctors improve diagnostic efficiency and assisting grass-roots doctors; enterprises, companies are also trying to broaden service channels, such as Speculative Technology, Huiyi Huiying and other companies have applied AI-assisted screening products to physical examination institutions and even non-public medical institutions.
Because of the contact with public hospitals, many startups play by reaching cooperation with the local government first, which is more conducive to their landing at the grassroots level. At present, Shanggong Medical Information mainly through the medical consortium from the public tertiary hospital to the primary medical institutions, through the top-down three-level system, in addition to providing AI sugar network screening technology for medical institutions, but also for the medical consortium to provide information storage cloud platform and referral solutions.
Services like this for public hospitals and governments are unlikely to generate high and stable revenues for startups in the short term. "But what healthcare AI companies are most afraid of is 'rushing to make money.'" Ji Xin said. Therefore, he is not optimistic about the practice of some startups that cannot withstand the huge and fierce "subsidy war" and seek profitable ways for non-main business.
The "circle of friends" has just become lively
In 2017, there were more than 100 medical AI startups at one time, according to the statistics of Yiou Think Tank, as of July 31, 2017, there were 131 medical artificial intelligence companies in China, and more than 100 similar financings. Although in the first half of 2018, 18 companies in the medical AI field were invested, with a total amount of more than 3.1 billion yuan, Ji Xin also clearly felt that the heat was much lower, and many startups have become anonymous since then.
He blamed it on the psychology of entrepreneurs "rubbing hot spots and lack of evaluation", entrepreneurs eat the "meal" of the capital market, in the cold environment of the capital market, but it is the winter of speculators, "capital is the smartest." He said.
"If you want to really cultivate deeply in medical treatment, there are too many market cultivation in addition to technology." Ji Xin said. Admittedly, the approval of the products of medical AI startups is only the first step, and the more favorable thinking for their commercialization is how to innovate ways to seize the scene. In the track of combining eye and diabetes, the application space of medical AI can be extended to the field of pan-health, involving sugar control drugs upwards, and extending the health management platform scenario downwards.
The medical AI track in 2018 has gradually run out of the echelon, but startups still can't take it lightly. Sun Chao, founding partner of Chongshan Capital, once told Yiou Health that in addition to BAT, domestic and foreign equipment giants with CT and MRI production and research and development as the main industry are also doing some "industrial upgrading things" in recent years. For example, in 2018, GE launched its artificial intelligence cloud imaging solution, and as the depth of technology continues to develop, it may be one of the next directions for this wave of players to enter the hospital service port with "software + hardware".
![To see how strong the AI is, someone took it to play a "script kill"[fig]](data:image/gif;base64,R0lGODlhAQABAIAAAP///wAAACwAAAAAAQABAAACAkQBADs=)