"47 years old, female, pulmonary nodule examination. Through the artificial intelligence assisted diagnosis system, the doctor can indeed detect some smaller lung nodules, effectively reducing the possibility of missed diagnosis. In addition, to some extent, the system provides the knot Information such as position, size, and density can help report writing to diagnosis.†This artificial intelligence scene does not have the difference between Ke Jie and AlphaGo, but people can see the artificial intelligence to medical empowerment.
Since the artificial intelligence in the domestic boom, the medical imaging AI segment has been joined by many players, including Shen Rui Medical, Yitu Technology, Pushing Technology, Huiyi Huiying, Tuoma Shenwei and so on. The startup company was once considered to be one of the most accessible areas for medical AI, and capital has an eager anticipation. According to data from the venture capital data platform, Whale's data shows that in the past five years, AI+ medical applications have completed 86 financing projects, of which 31% are images, accounting for the first. But are these medical imaging AI companies actually starting to make money? Is the landing scene clear? Why do you mostly get together with "lung nodules"? What are the resistances of today's development?
Landing scene: Should medical imaging AI be oriented to primary hospitals? Or is it a top three hospital?
Today's medical imaging AI companies mostly like to declare their own cooperation with the top three hospitals and top hospitals. In terms of Shenrui Medical, its products have entered more than 200 hospitals, most of which are top three hospitals; from the Huihui Huiying official website, they have already reached cooperation with more than 700 top hospitals in China; according to media reports, Tuoma Shenwei also Has cooperated with 50 top three hospitals.
What is the significance of these numbers? Should medical AI really play its value in primary care? In China, the distribution of medical resources is uneven, the excellent medical personnel of primary medical care are seriously lacking, lack of high-tech medical equipment, and the clinical diagnosis and treatment ability is relatively weak. In the course of daily medical treatment, misdiagnosis of primary hospitals has also occurred from time to time. In contrast, the top three hospitals are in a saturated state regardless of medical equipment or doctor resources. Why do these AI products choose the top three hospitals? Where is the living space? Will the willingness to pay not be low?
In this regard, Shen Rui Medical CEO Qiao Wei said that primary care does need more such products. However, let the doctors of the top three hospitals accept it first, and then let them affect the grassroots. This is a process in itself.
Industry investors also pointed out that the cooperation between the company and the top three hospitals is based on two considerations: first, it is conducive to the formation of high-quality labeled data training models; second, to play the role of the top hospitals, The brand endorsement of the product. Nowadays, these enterprises and hospitals are mainly based on scientific research cooperation and help the department to build a data platform. The paid payment scene is not clear. This is a problem that everyone faces together.
In fact, one of the major reasons for not paying for the scene today is that the current algorithm is still immature, the accuracy is not up to the expectation of this tool, and a lot of data is needed to optimize and iterate. Ma Jun, dean of Tongren Hospital affiliated to Shanghai Jiaotong University School of Medicine, said in his conversation with Yiou Health that for medical AI, clinicians now have both open-ended and troublesome attitudes because the ideal state of the tool is Improve the productivity of doctors, but there are not so many software that actually improve performance. In the process of working with these teams, experts and clinicians not only need to contribute their wisdom, but also affect the efficiency of their work.
Product: Get together with "pulmonary nodules", medical imaging AI equals lung nodule screening?
In addition to clear payment scenarios, it is one of the important factors to realize the commercialization of products in the same category. However, in the field of medical imaging AI, due to the characteristics of open data sets and lung nodules for observation, almost all enterprises are engaged in lung nodule projects, and even the activities of human-computer collaboration "reading films" are mostly For example, lung nodules, is the product too homogenous?
Relevant industry investors believe that the diagnosis in the medical field is the result of comprehensive inquiry. The diagnosis of any disease is a multi-dimensional and multi-indicator process, which is still not achieved for medical imaging AI. Therefore, enterprises should not be screened first, and the indicators of their operating projects are as single as possible.
It is understood that for the lung nodule project, the indicators that need to be diagnosed are relatively simple, and have a public data set, then, the enterprise is also understandable from the lung nodules. The lung nodule project cannot represent medical images or the entire artificial intelligence medical treatment.
From the perspective of products, Shenrui Medical takes “two legs†to walk, and does all the product lines, and the technology is deep. In terms of technology, in addition to the open lung nodule project, stroke, heart, fracture, abdomen, breast, etc. also have a layout. It is reported that the diseases developed by Huiyi Huiying also include leak diagnosis of chest CT, mammography, and nuclear magnetic analysis of cerebral hemorrhage.
In addition, AI Medical Corporation does not have to limit its vision to medical imaging. Some people believe that AI is also promising in disease treatment, drug development, and health management.
Dilemma: What are the bottlenecks that hinder the development of medical imaging AI companies?
For medical imaging AI products, the application is based on the deep learning of massive data by the machine. Without data, even if it is a “smart woman, it is difficult to get rid of riceâ€, which is one of the difficulties that medical imaging AI companies have been facing.
In fact, the dilemma faced by enterprises is not only the problem of the amount of data, but also the quality of the data. For doctors' diagnosis, medical data is related and needs to refer to various data, but obtaining a complete data structure itself is very difficult. On the one hand, due to historical reasons, the retention of hospital data itself is incomplete. On the other hand, a complete data structure includes medical history, family history, and information on the treatment link, but now the company is doing different links, focusing on a certain database, the data link cannot be opened, and the complete data link cannot be obtained.
Qiao Wei believes that for medical imaging AI enterprises, the opening of the data link is a direction and a goal for future development, and now this is the biggest bottleneck for enterprise development.
In addition, national policies are also a major impediment to the development of medical imaging AI companies. According to the new version of the "Medical Device Classification Catalogue" issued by the State Food and Drug Administration in September 2017, it will be implemented on August 1, 2018. There is a definition of such artificial intelligence software, which means that medical imaging AI companies have The requirement of "holding the certificate".
So far, the State Food and Drug Administration has issued Class II certificates to companies such as Tuma Shenwei, Shenrui, Imagination, and Intranet. There are no enterprises that have obtained Class III certificates. That is to say, the products of such enterprises only have auxiliary diagnostic functions, and the products that provide clear diagnostic prompts need to obtain three types of certificates. The demand for "small assistant" products is naturally far less than the products that can be actually diagnosed, the most valuable. The product is definitely still in the diagnostic segment. It is reported that 11 companies have jointly established the certification standards for three types of medical devices in the relevant departments. It is expected that the first batch of companies that have obtained three types of certificates will be born in 2019.
Some industry insiders pointed out that the approval of such products in the national level has been paid attention to, but the charging mode of the traditional medical scene for pure software diagnostic services has not been fully educated and accepted. What form is charged after the certificate is obtained? Do you join the hospital fee list? These issues are still unclear.
It can be said that the domestic medical imaging AI enterprises have started shortly and are still in the stage of slowly accepting the market. However, as technology matures and companies accelerate, the industry's "gold standard" will eventually emerge, and the end of the free trial era will be in the near future.
Xi'an Healthway Biotech Co.,Ltd , https://www.xianhealthway.com