Giants launch AI_medical image data into "entry"

A doctor who has been in the imaging department of a top three hospital in Guangzhou for 8 years has used his "advanced threshold" to describe his work. "We need to go through historical image comparison and quantitative analysis to make a film for the patient." The basic diagnosis and treatment judgment, and then, determine what kind of treatment plan the patient needs.

In the medical diagnosis, the value of the image cannot be replaced. 90% of medical data are images, from CT, X-ray, magnetic resonance, ultrasound, PET, etc. For example, before a surgery for a cancer patient, a film is taken, based on information such as the condition of the tumor and the degree of stenosis of the blood vessel, to determine the surgical plan, the medication plan, and the follow-up risk.

Advances in artificial intelligence (AI) technology and image recognition technology have made this work a powerful assistant - artificial intelligence medical image analysis system. For the same two-dimensional medical image, the doctor needs to spend ten minutes to observe and reason, and the artificial intelligence can be “read” in tens of seconds after deep learning. With the support of sufficient big data, artificial intelligence is expected to increase the diagnostic speed by 10 times, and thus greatly reduce the cost of diagnosis and treatment.

For doctors, efficient analysis can help them save time on reading, reduce misdiagnosis, and provide richer historical image comparisons. The hospital is also happy to see the digital results of artificial intelligence processing of medical images, and facilitate the construction of medical databases, thereby reducing the cost of medical treatment programs. The collection of capital, technology and medical data is the three shareholder winds of the current artificial intelligence medical image take-off. Can this "reading film" assistant who is still in the internship period be the only one?

Compared with 2016, the artificial intelligence medical industry in 2017 is more enthusiastic and eager to achieve results.

"Everyone is tired of talking. After all, no one can rely on 'chicken blood' to live." Yilan, a medical analyst at Yiou Think Tank, told Caijing reporters, "Is there a kind of application at the application level?" What? I think it is artificial intelligence medical image analysis."

The advantage of artificial intelligence medical imaging is in the data compared to other areas of artificial intelligence medical. The imaging data is not like a medical record. It contains scattered information such as medical history, patient information, symptoms, treatments, recovery, etc. It has a high degree of information integration - a medical pathology film contains a lot of high Value information. Therefore, compared with other medical data, image data processing is less difficult and the processing value is higher.

“The original image of medical images is very dimensional and complex, and artificial intelligence transforms high-dimensional data into a low-dimensional, more problem-solving problem.” Wang Xiaozhe, chief architect of Zero Technology Co., Ltd., on Caijing The reporter said that the medical image data itself fits well with the artificial intelligence representation model algorithm.

The standard of image data is unified, it is easier for the algorithm to "entrance", and the construction of the auxiliary diagnostic model is easier.

The image data of each hospital is not only in one department of the radiology department, but also relates to each clinical department. This also means that the image data does not exist in one information area. The director of the Information Department of Peking University Cancer Hospital told the Caijing reporter that medical image data is the largest in the hospital, and it is standardized and more convenient for machine reading. This is very important.

There are a lot of brains in the medical imaging. As early as 2003, Zhou Zhenyu, senior director of the Department of Clinical Science at Philips Health Technologies, and his mentor had an idea to create an image big data platform, but "in those days, many challenges we could not overcome, such as image quality, data and computers. The mismatch, the logical thinking of diagnosis is not standardized, which led to the failure to achieve true wisdom medical treatment 15 years ago." Zhou Zhenyu told the Caijing reporter.

Just as videologists need to read a large number of clinical medical images, "feeding" pathological image data is the most important way of learning artificial intelligence systems. Compared with Zhou Zhenyu and his mentor, the pathological image data that can be “fed” is more and more sufficient, and the artificial intelligence analysis ability can grow up.

Because the data is relatively abundant, developers are able to gather in vertical areas. In 2016, the Beth Israel Deaconess Medical Center (BIDMC) and Harvard Medical School announced that they have jointly developed an artificial intelligence breast imaging diagnosis and treatment platform. The research and development personnel constantly input a large number of pathological images, let the system complete the cancer cell recognition and health field division of the film, and complete self-improvement in the deep learning technology framework to improve the recognition accuracy and efficiency. The platform's head, Andrew Beck, said that the platform's image analysis accuracy of the patient's breast can reach 92%, combined with the analysis of pathologists, the diagnostic accuracy of 99.5%.

The uniqueness of medical care forces artificial intelligence companies to cooperate with hospitals from the beginning. Because in the health system, the hospital is relatively independent and the data is unique, so that all the artificial intelligence enterprises have the opportunity to enter more. In 2017, artificial intelligence companies targeting the medical field frequently announced cooperation projects with hospitals. From the data released from the public, not only the ability of medical image processing and analysis has improved, but also more clinical application cases.

One example is the tracking of lung nodules. Through the artificial intelligence system, not only can the patient be told where there is a pulmonary nodule, but also the probability of malignancy can be predicted through the original data analysis. The patient is screened for the time of review by probability, or whether a needle biopsy or a corresponding genotype is required. Check. These clear messages not only make patients more aware of the condition, but the doctors and patients are easy to communicate, and they also pay for weight loss. This is what the government is willing to see.

According to the data provided by Cai Yifei, the CEO of the independent third-party medical imaging platform, Huiying CEO Hui Xiangfei, the number of readings by Huiyi Huiying has exceeded one million. This is inseparable from the pace of speeding up the actual application scenario. At present, Huiyi Huiying has access to more than 500 grassroots hospitals and more than 200 top three hospitals.

Another artificially intelligent medical company, Airdoc, which focuses on ophthalmology, has collected hundreds of thousands of fundus photographs from domestic and foreign hospitals. In the second half of this year, Airdoc reached a cooperation agreement with Zhejiang Eye Hospital and Shanghai North Hospital, and established a technical base related to artificial intelligence ophthalmology image analysis.

According to incomplete statistics, there are more than 40 startup companies entering the field of artificial intelligence medical imaging. In addition to artificial intelligence companies with high verticality, the actions of Internet giants are becoming more and more obvious.

Given the long-standing data advantages of Internet giants, their involvement may directly affect future changes in the field, and policy blessings have amplified this influence.

On November 15, the Ministry of Science and Technology announced at the kick-off meeting of the new generation of artificial intelligence development planning and major science and technology projects that the first batch of national new generation artificial intelligence open innovation platform list: relying on Baidu to build automatic driving, relying on Alibaba Cloud to build the city brain, Relying on Tencent's construction of medical imaging, relying on the University of Science and Technology to build intelligent voice. This is an artificial intelligence medical image that was first listed as a separate classification in the field of artificial intelligence by the government.

Tencent’s “airborne” artificial intelligence medical imaging market was only three months before the announcement. In August, Tencent launched its first AI medical product, "Shadow", which is mainly used for the screening of early esophageal cancer. At present, Sun Yat-sen University Cancer Hospital, Guangdong Second People's Hospital, Shenzhen Nanshan District People's Hospital have joined this cooperation project.

Whether it is the Internet giant or the more vertical artificial intelligence medical company, the results of sinking to the hospital in 2017 are very significant - ensuring the sufficiency of artificial intelligence data "food" and discovering problems encountered in the application, away from clinical More recently, more commercial prospects, these are all that capital is happy to see.

In addition to the aura of artificial intelligence technology itself, the medical imaging market has reached a new critical point in 2017.

The gap between the growing market demand and medical imaging resources has become significant. The profit of the booming medical imaging market has attracted artificial intelligence companies to flock to it, hoping to become the first "miner" in the crack.

According to Yang Hongfei, CEO of Flint Creation, from the current market size of medical imaging, the patient-side growth rate is high, and the income from image inspection accounts for more than 10% of the total hospital revenue, which is closely related to the proportion of pharmaceutical revenue. Flint created the "Market Map and Industry Development Analysis of Medical Imaging" released in June, pointing out that according to the overall medical expenditure of China in the past five years, the scale of China's medical imaging market will reach about 600 billion to 800 billion yuan in 2020.

This also lifts the "appetite" of capital. According to incomplete statistics, as of press time, there are 3 enterprises in the domestic artificial intelligence medical imaging market that have completed angel round financing, 2 Pre-A round financing companies, 7 A rounds, 3 B rounds, and 1 C+ round.

Among them, the three largest financings were concentrated in May and the second half. In May, Etu Technology announced the completion of the 380 million C-round financing, led by Gaochun Capital Group, Yunfeng Fund, Sequoia Capital, Gaochun Capital, and Zhenge Fund. In September, the company announced that it has completed 1.2. Billion B round of financing, led by Qiming Venture Capital, Yuansheng Capital, Sequoia China joint investment; in October, Huiyi Huiying announced the completion of "hundreds of billions" of Series B financing, the investors are Datai Capital and others 2 investment institutions.

In the view of Wang Haoquan, president of consulting firm Frost & Sullivan China, domestic capital is closely entwined with artificial intelligence medical images. "Artificial intelligence imaging is inevitable for medical advancement. This improvement is visible and the most popular at the moment. It is expected that capital enthusiasm will not be cold by 2018." Wang Yiquan told the Caijing reporter.

However, it seems that it is not easy to cut a piece of cake that looks like a huge cake. Even the leading IBM Watson, its Watson for Oncology has not reported profitability.

The work of the tumor-assisted diagnosis and treatment solution Watson for Oncology is to provide doctors with medical advice and medical treatment results, and to generate treatment plans to improve the accuracy of doctors' diagnosis and treatment. Backed by IBM's own information analysis expertise and market recognition. degree. Watson for Oncology has entered China, the United States, the Netherlands, South Korea, Thailand and India. However, the current IBM Watson earnings situation has not been publicly reported for the time being.

Compared with Watson, a number of domestic artificial intelligence medical imaging companies are still in the application stage of disease screening, that is, to determine whether there is a certain type of disease in the image, "know it", "know why" and "know how to make it In the three steps, most artificial intelligence medical imaging companies are still in the first step of exploration.

"As far as the current situation is concerned, the results achieved by AI are far from expected." Wei Ruili, director of the ophthalmology department of Shanghai Changzheng Hospital, told the Caijing reporter: "AI is mainly used for screening. In actual use, the doctor will review it again. Once again, just as the patient took the diagnosis report of the local hospital, we saw that we still have to reconsider."

In addition, domestic companies are still concentrated in the field of diseases where medical image analysis is relatively simple, and the value is relatively low. Taking the lungs as an example, lung cancer recognition is a popular field of artificial intelligence medical images. This is because the lung image recognition has a natural contrast, which is a relatively easy to overcome direction, but does not have deep analysis ability for specific symptoms of lung cancer.

Liu Zaiyi, a professor of radiology at the Guangdong Provincial People's Hospital, felt deeply about this. "Most of the lung cancer cases in our hospital are reviewed. The lungs in the third and fourth phases have many metastases, combined exudation, and atelectasis. Computers It is difficult to achieve automatic comparison of these features. These assisted doctor products may indeed reduce some workload in the clinic, but the help and application scenarios for doctors are better than the school."

The areas that are more likely to break through, means that the competition is more intense, and the risk of being crowded out by the giants is higher. In the interviews of Caijing reporters in the industry, the profit model and profitability issues are still the problems in their minds.

In November, Everbright Securities analyzed that the service-oriented medical imaging downstream industry needs to be innovative in its service model. Enterprises that quickly obtain sufficient resources in telemedicine imaging diagnostics and independent imaging centers will have greater advantages in image intelligence diagnosis in the future.

Judging from the market reaction of the giants, IBM has tried many times to open up the Chinese market, and Ali and Tencent, which have already been active, have a strong impact on the market. Shang Yi, a medical analyst at the billion-dollar think tank, told the Caijing reporter. "According to my contact with the capital, I can expect the giants to step up their actions in 2018. A round of 'big fish to eat small fish' will begin soon. ."

Wang Haoquan, president of consulting firm Frost & Sullivan China, believes that even if the giants do not integrate the artificial intelligence medical imaging market so quickly, the killing between small companies will be very fierce. Who can win the unpredictable, and even possibly share the bicycle industry? Similarly, it is the representative of the capital game.

In 2017, the combination of artificial intelligence and medical treatment began to deepen and refine. As the earliest and most competitive “battlefield”, many problems encountered in the artificial intelligence medical imaging industry are difficult to find empirical reference, and this also reflects its distance to taste. This delicious "first soup" of artificial intelligence medical treatment is the closest.