2018 Medical Artificial Intelligence Report: Investigate the product landing situation of 60 domestic medical AI enterprises, the first generation products has matured.

September 27, 2018

Original author: Luo Shiming

In September 2017, the Artery Network Eggshell Research Institute published the “2017 Medical Big Data and Artificial Intelligence Industry Report”. It reviewed the past and present life of medical artificial intelligence, and made a deep analysis of the ability, application scenario, cost structure and talent situation of medical artificial intelligence, which provided an important role for industrial personages and regulatory agencies. Reference materials have aroused extensive discussion.

After one year, AI enterprises have moved their products to clinic, and the research on the subdivision of diseases/applications has been deepening. AI products have been integrated into all aspects of the medical process.

After years of development, some AI products have basically matured, and the business model is clear and clear. AI technology has begun to expand the application boundaries in the medical field, which has brought many interesting attempts to the medical operation process.

Supervisory departments keep up with the pace of technological development and actively participate in the wave of industrial upgrading. The regulatory thinking has gradually become clear. The Chinese People’s Procuratorate has completed the construction of fundus sugar net photos and lung nodule image databases. There are more than 30 related three categories of products for review. I believe we will soon see three types of AI medical products on the market.

At the same time, we launched a new annual medical artificial intelligence report. This report focuses on the development and application of medical artificial intelligence products and doctors’ use. We surveyed major AI medical companies, interviewed doctors involved in the research and development/use of AI products, and presented the current development status and next research and development directions of medical AI in China from an industry perspective.

Through interviews and investigations, we have reached the following key data and conclusions:

1.The product landing situation of Chinese medical AI enterprises and the pipeline under research.

2.The main medical AI enterprises have matured a generation of AI products, and the business model is becoming clear. The medical AI industry has entered the stage of “leapfrogging and restarting”.

3.The first batch of pulmonary nodule screening and sugar mesh projects have entered the stage of comprehensive positive feedback.

4.Regulatory authorities respond quickly, interact with industry frequently, and review points or recent promulgation.

5.In the next stage, primary health care will be the biggest beneficiary and main battlefield of this wave of artificial intelligence.

6.The main research and development direction of AI enterprises: enhancing the coverage of subdivided diseases and exploring new scenarios according to their own business characteristics.

7.Health medical artificial intelligence technology will become the basic ability of society.

8.Development of Artificial Intelligence Technology and Continuous Expansion of Medical Application Boundary

The rapid progress of modern human society mainly depends on the three industrial revolutions. The first industrial revolution was marked by the improvement of steam engines, the second industrial revolution by the wide application of electricity, and the third industrial revolution by the invention and use of computers. The three revolutions have dramatically changed people’s production and lifestyle, social structure and even the world pattern. Intelligent interconnection technology represented by artificial intelligence is becoming the driving force of the fourth industrial revolution.

Artificial intelligence (AI) is a set of technologies that give computers the ability to perceive, learn, reason and assist in decision-making, thus solving problems in a way similar to that of human beings. In the past, computers could only work according to pre-programmed fixed procedures. With such capabilities, the way computers understand and interact with the world will become more natural and sensitive than before.

Key AI technologies include:

Vision: the ability of a computer to “see” by recognizing the content in a picture or video.

Speech: The ability of a computer to understand what people say and transcribe it into words.

Language: The ability of a computer to grasp many subtle differences and complexities in language (such as slang and idioms), “understand” the meaning of a utterance.

Cognitive ability: The ability of a computer to “reason” by understanding the relationships between people, things, places, events, etc.

These abilities of AI correspond to the medical field. Medical AI system will have various forms of dialogue ability. This will help information flow between individuals, understand patients according to their medical history, and help medical care enterprises provide attractive personalized medical advice to consumers.

Artificial intelligence system has stronger observation and insight than human beings. It can process a lot of medical and patient information quickly, which makes doctors spend more time on patients. Artificial intelligence system can provide assistant suggestions based on comprehensive information, which can help decision-making and reduce human bias. Artificial intelligence system is learning constantly according to the latest information, results and operations, which helps medical professionals to make more wise and timely decisions. Artificial intelligence has few boundaries, which means that the existing medical artificial intelligence capability circle will continue to expand, and this is happening.

1. Trends in Medical Artificial Intelligence Policy: Regulators and Enterprises Join Hands to Establish Standard Databases

In early April 2018. FDA (Food and Drug Administration of the United States) approved the software program of IDx-DR, the first self-contained AI diagnostic equipment developed by IDx company for first-line medical treatment, which can be used without the participation of professional doctors. In this case, diabetic retinopathy is diagnosed by viewing the retinal photographs. The product has been approved for 21 years. It took seven years for IDX and FDA to communicate on how to evaluate the system and ensure its accuracy and security. Several AI products recently approved by the United States are all based on the certification process of CLASS II, which is proved to be safe and effective by comparing with the traditional CDSS (Clinical Decision Support System). China’s laws and regulations are relatively stricter, and the path control of clinical evaluation is very stringent. On August 1, 2018, China’s new edition of Catalogue of Medical Devices came into force. Medical software was approved according to two or three categories of medical devices. The Catalogue points out that if the diagnostic software provides diagnostic suggestions through its algorithm, it only has the function of assistant diagnosis and does not directly give diagnostic conclusions. The related products in this sub-catalogue are managed according to the second category of medical devices. If the diagnostic software automatically identifies the lesions and provides clear diagnostic hints through its algorithm, the risk level is relatively high. The related products in this subdirectory are managed according to the third category of medical devices. Therefore, most of the AI products we see now belong to the third category of medical devices. To cope with this policy, most enterprises in our country adopt the method of adding or deleting diagnostic functions and declare the products of two or three kinds of instruments at the same time. At present, many enterprises have taken the lead in obtaining two kinds of certificates. At present, Hirschner Heterogeneity, Yasen Science and Technology, Huiyi Ying, Tuma Shenwei, speculation, Shenrui, Airdoc, Yitu Medical and other well-known artisans. Intelligent enterprises are actively carrying out the declaration of three types of medical devices. According to Itu Medical, all of their product matrices are certified in three categories. Airdoc sent the first server in China that loaded AI software for inspection. At present, none of the products has obtained three types of certificates. According to the registration process of medical devices, the product from declaration to final approval should go through six steps: product finalization, testing, clinical trials, registration declaration, technical review, administrative approval and so on. At present, most of the medical artificial intelligence products declaring the three types of devices remain in the initial stage of registration and declaration. As a technical support institution of national supervision, the Chinese Procuratorate undertakes the quality evaluation and research of medical artificial intelligence products. With the advantages of rich experience in software testing of medical devices, the optoelectronic laboratory has set up a special AI team to undertake this work.

The examination and approval of AI medical products has always been a matter of great concern to the industry. Ren Haiping, Director of the Optical, Mechanical and Electrical Medical Device Inspection Department of China Institute of Food and Drug Identification, said in a public speech: “What we have done is to compare the characteristics of some new products, especially the new product platforms and inspection methods without national and trade standards.” Research work on Evaluation criteria, etc. The Chinese Procuratorate has accepted 30 to 40 applications for AI products from all over the country, and we have done our preliminary work. Director Ren Haiping’s point of view can be summed up in three aspects. First, not all products on the market need to carry out clinical experiments. The data set used from the real world can be used for pre-clinical evaluation and clinical evaluation, and the clinical evaluation can be used for prospective and retrospective clinical evaluation. The state has certain requirements on the quality evaluation of AI medical devices, including AI, medical devices and software. At present, the quality evaluation of AI by the Chinese Procuratorate is mainly based on the following three guiding principles:

• Guidelines for Technical Review of Medical Device Software Registration
• Guiding Principles for Registration Technology of Mobile Medical Devices
• Guiding Principles for Technical Review of Medical Device Network Security Registration

The inspection system of AI medical devices planned by the Chinese People’s Procuratorate has the following four steps: standard data, phantom testing, software performance and simulated confrontation. Two databases, color fundus image and lung CT image, have been established.

The process of database construction includes three steps: data collection, image annotation and data management. Data governance in software design and development requires the same.

Fundus Image Standard Database

The establishment of the standard database of fundus imaging was relatively early. At present, a database containing 6327 cases has been formed.

Pulmonary Imaging Standard Database

The construction of the standard database of lung imaging started in February, 2018, and began to recruit pulmonary nodule image calibration experts nationwide in April. The online examination and training of these experts were completed in early May. The off-line closed calibration was completed on June 10. Twenty-four calibration experts and 15 arbitration experts jointly completed the case calibration.

After the on-site calibration of lung nodule images on 10 June, in order to enter the testing stage as soon as possible, the AI team of optoelectronic Laboratory of the Chinese People’s Procuratorate issued a notice on 15 June to solicit opinions on the testing scheme of lung nodule AI products, covering 11 enterprises sent to the Chinese People’s Procuratorate for inspection.

The 11 enterprises are: Health Education Technology, Tulu Shenwei, Zero Krypton Technology, Ito Technology, Yunji Technology, Shenrui Medicine, Huiyi Medicine, Presumption Technology, Yasen Technology, Point Biology c, Wing Spread Picture.

It is understood that “in-depth learning assistant decision-making medical device software review points” will soon enter the stage of soliciting opinions.

III. Analysis of Artificial Intelligence Landing: Leap Forward, First Generation Product Maturity

It has been 7 years since Watson laid down the business development direction of its medical treatment in 2011. In these seven years, the development of AI has been booming. Numerous enterprises have followed the trend and deep learning algorithms have undergone several generations, but after the tide, the remains of numerous pioneers have been left behind. Now, survivors and later generations have gradually formed a leading camp in the field of medical artificial intelligence. In the process of in-depth learning, enterprises can report a gratifying accuracy figure for their AI products. However, the new era is no longer an algorithmic era, and the measurement of AI is no longer a number or a man-machine scale. The results of the war can be judged. If you want to survive in this circle, you have to go to the hospital. Nowadays, the competition of AI medical industry focuses on landing link. Continue our classification of medical artificial intelligence application fields in “2017 Medical Big Data and Artificial Intelligence Industry Report”: medical imaging, case literature analysis, virtual assistant, new drug research and development, hospital management, health management, gene, disease prediction and diagnosis, intelligent devices, for medical artificial intelligence enterprises. This paper classifies and counts the current product directions of 108 domestic medical AI enterprises which are active in the market, and interviews and investigates the main medical AI enterprises to understand their current product application situation and the next generation product development status, and obtains the following contents. We have counted the main product lines of 108 medical AI. We have obtained the following AI product directions and disease maps.

We can see that most AI enterprises choose to launch products in the four directions of medical imaging, case literature analysis, health management and virtual assistant. The two hot directions of lung nodule screening and sugar screen are far ahead, but at the same time, quite a number of enterprises have focused on cardiovascular diseases. AI Enterprise products show a decentralized trend. According to the Global Market Insight data report, according to the application division, drug R&D has the largest share in the global medical AI market, accounting for 35%. The smart medical imaging market is the second largest market segment, and will grow at a rate of more than 40%, reaching $2.5 billion in 2024, accounting for 25%. These two areas are also the most widely used artificial intelligence in various medical scenarios. Next, we will take these two scenarios as the main line to introduce the landing of artificial intelligence in China.

1. Step by Step, Medical Imaging Comes to the Forefront

The application of artificial intelligence in medical imaging mainly includes classification of image or examination, location of organs, regions or markers, detection of target and pathology, segmentation of organizational structure, segmentation of focus area, and image registration. The main diseases are pulmonary nodules, sugar web, stroke, etc. There are three main directions: disease screening, focus delineation and three-dimensional imaging of organs. We will discuss the application of AI in lung screening, sugar screen, lesion delineation and three-dimensional imaging of organs. We made a simple statistics on the enterprises involved in diseases, and interviewed and investigated the landing situation of the products on behalf of the company. Pulmonary nodule screening is undoubtedly the hottest direction of medical artificial intelligence products. As of July 2018, incomplete statistics showed that more than 20 AI companies have produced specific products in the field of pulmonary nodule screening, and most of them have received investment. The number of new lung cancer patients in China is the first in the world every year, the number of lung cancer-related deaths is the first in the world every year, the demand for early screening is strong, and low-dose spiral CT is being widely promoted. In terms of image quality, chest CT images have thin layers, clear vision, few interference factors, and the characteristics of lesions can be followed, which are ideal for intelligent image interpretation. In addition to the scarcity of Chinese imaging doctors and the strong promotion of national policies, the application foundation in this field is perfect. In 2017, the major AI enterprises that focus on Pulmonary Nodules Detection all submitted brilliant answers, and the sensitivity soared, 95%, 96.5%, 98.8%… Pixel differences, which are hard to detect with the naked eye, are invisible to AI’s powerful arithmetic.

(The above content is an excerpt from the “2018 Medical Artificial Intelligence Report: Leap Forward and Beyond”, reproduced the automatic pulse network, deleted if invaded.)