Out of the search for auxiliary diagnosis and new business path, how are the head enterprises of medical image AI building a new ecosystem?

In 1987, time delay neural network (TDNN) proposed by Alexander Waibel can be traced back to the first CNN algorithm in history But limited by limited computing power and data, CNN has not become the focus of scholars Until 19 years later, Geoffrey Hinton at the University of Toronto redefined deep learning, CNN's representational learning ability became a hot topic again Then, CNN's data mining, machine learning and other capabilities are gradually mature in image processing applications People who are interested in it gradually begin to realize that such a new technology may have sufficient development space in medical treatment Therefore, the first people to enter the medical imaging AI are a group of Internet people Since 2015, a large number of medical imaging AI companies have been set up and are flocking to the blue ocean market of imaging AI But after entering, people wake up: how to overcome the deep barriers without sufficient medical knowledge? Recalling the scene at that time, Chen Kuan thought of technology as a wry smile: "at that time, I didn't want to see my own products." The transformation began in 2017 When a group of Internet users continue to work together with doctors, deeply understand the work process of doctors, and the country begins to realize the importance of AI gradually, AI products begin to enter clinical trials The mutual assistance bridge between doctors and entrepreneurs has been established, and AI enterprises have experienced rapid development in the next two years 2019 is a year for medical AI to become universal We can see that more intelligent medical oriented enterprises are emerging Meanwhile, health management, community management and even medical beauty are all trying to introduce AI carefully In this regard, the arterial network has conducted research on artificial intelligence, and a set of data may help us better understand the development of the whole medical AI in 2019 According to the data of the knowledge base of the arterial network, from December 24, 2018 to August 20, 2019, the number of global medical AI financing events totaled 107, with a total fundraising amount of 13.958 billion yuan (excluding the financing events with an investment amount of "undisclosed"), and the specific round distribution and financing amount segmentation are shown in the figure below From this chart, we can see clearly that the financing events of medical AI in 2019 gather in round B and round C By contrast, there are no successful IPOs and only four angel rounds The distribution of the whole event is "large in the middle, small at both ends" This may mean that the head AI enterprises have established sufficient barriers, and it is difficult for new enterprises to enter The scale of medical AI enterprises is also expanded As can be seen from the above figure, AI enterprises with more than 50 employees occupy only 1 / 4 of the total, and high-quality resources are obviously gathered Among them, image AI enterprises are the first ones to participate in artificial intelligence From the dynamic of these enterprises, we may see the clue of the whole industry How about the development of these enterprises? As shown in the figure above, with the completion of b-round voxel technology and c-round extrapolation technology by the end of 2018, the head image enterprises have all received a new round of funds Now that 2019 has passed half a year, the products of pulmonary nodules, breast screening and other products of head imaging enterprises have been mature enough With data and hospital layout gradually establishing their own barriers, corresponding AI products enter the clinical stage of approval In the process of waiting for approval, the top enterprises are constantly exploring the potential value of existing resources and the potential commercialization possibility of AI Now the key question is, who will be the payer of image AI? Who may be the payer of image AI? Liu Shiyuan, director of imaging medicine and nuclear medicine of Changzheng Hospital, once said: "in the future, the work of imaging department must be intelligent and the report is structured." Potentially, AI must be part of imaging, but not today Looking away from the imaging department, is there still room for AI development? The answer is yes, pharmaceutical companies, patients and insurance companies may be potential payers of AI image, and the hospital is not only the radiology department willing to pay for AI image products Let's talk about the hospital first Different hospitals have different needs for products If AI products want to enter the top three hospitals, they must grasp the two key needs of doctors in the top three hospitals - efficiency needs and scientific research needs Now the mature CT lung, CT liver and brain MRI products are to meet the doctor's pursuit of reading film efficiency, but this only accounts for a part of the working time of imaging doctors More close to the needs of doctors is the needs of scientific research - helping doctors to control data in the era of big data For township hospitals with less medical ability, AI enterprises can build private cloud and cloud PACS connecting medical consortia, and also cultivate doctors' reading ability and reporting ability in the hospital by teaching Although this part of the market is huge, due to the limitations of the hospital's own scale, there will be corresponding restrictions on the construction expenditure of smart hospitals, which will affect the willingness to pay for AI enterprises As for the willingness to pay for AI auxiliary tools, township hospitals are still stronger than the top three hospitals The most likely paid medical institutions include private specialized hospitals, hybrid hospitals and third-party image centers These three types of hospitals lack the medical resources to compete with the top three hospitals, need AI to supplement, and lack the trust of the common people in the system, and need AI to add color to the technical strength of the hospital In general, selling AI equipped equipment to hospitals, building a cloud PACS platform, or selling non AI image related services (such as digital film) will take on the main revenue source of AI enterprises for a long time, while the commercialization of AI auxiliary diagnosis business still has a way to go Talk about pharmaceutical companies AI image enterprises can meet the needs of three types of pharmaceutical enterprises One is to act as cro to reprocess pathological data; the other is to screen patients in clinical trials with the help of big data tools; and the third is to help pharmaceutical enterprises carry out accurate digital marketing through community cooperation The most easy channel to get cash comes from digital marketing AI image enterprises can cooperate with the medical department to conduct real world data research and assist in the validation of drug treatment effect and standardized diagnosis and treatment mode; they can also cooperate with the marketing department and sales department to build academic platform and assist in drug sales Since these three types of businesses do not need the approval of the drug administration from the image AI enterprises, but rely on the resources they integrate in the long-term layout, it is possible that these cooperation will be one of the important revenue sources of the AI image enterprises Compared with the above two models, it is much more difficult for patients to become stable paying objects Although in many top three hospitals, MDT consultation and auxiliary diagnosis based on AI have entered the charging list of hospitals However, due to the uncertainty of AI technology core, it is difficult for enterprises to require doctors to recommend AI products to patients At the same time, due to the lack of market education, when choosing AI services and non AI services, most patients prefer to choose traditional services without AI core But with the advancement of technology, policy and patients' cognition, patients are likely to become stable payers in the future So, how do image head enterprises seek development in the dilemma? Arterial network interviewed nearly ten image AI head enterprises, trying to tease out the logic It should be noted that some of the following enterprises involve a number of breakthroughs, while this paper only selects a part of each enterprise's innovative business as a case study Medical institutions as a breakthrough point are the starting point for AI enterprises Almost all the initial AI image products are designed to serve doctors So even though these products are still unprofitable, they set the stage for the promotion of new products Under the premise of ensuring the continuous iteration of existing products, all enterprises are extending from the image >>>>Conjecture Technology: clinical practice is not the only way for doctors to exert their value As an experience discipline, doctors hope to spend more time on experience sorting and exploration of unknown possibilities, and comb them into papers for more like-minded people to exchange and learn " Xi Weiling, President of marketing for speculative technology, once said With the continuous promotion of cooperation with doctors, the researchers have a deeper understanding of doctor's treatment process and demand, and then know how to help doctors standardize data structure and use data, and customize algorithm model for scientific research Such a set of thinking concretization has created the research platform of today's infer scholar center AI scholars Only the contribution status of RSNA was counted In 2019, the research platform assisted several clinical cooperation hospitals to complete more than 300 contributions of RSNA, involving the comparison of artificial and AI, pathology, efficacy evaluation, scanning parameters, high and low dose comparison of 12% In this way, it can effectively transform the experience of researchers in communication with doctors, greatly expand the value and application scenarios of deep learning in the medical field, and realize the artificial intelligence from "auxiliary clinical diagnosis" to "auxiliary clinical research" Through such an AI based research platform, it is assumed that it can provide original AI research services for all departments and diseases in the hospital The platform has dozens of AI algorithms and single disease data management capabilities, which can serve the scientific research, teaching and industrial transformation of the whole hospital For the hypothesis, this model will help them to obtain more medical resources and give AI more creativity >>>>Shenrui medical: the value of intangible assets is difficult to measure and speculate The Shenrui Research Institute, led by Professor ieeefellow Yu Yizhou, the chief scientist of Shenrui medical, more relies on its own scientific research strength to establish all-round scientific research cooperation with medical institutions and doctors In the balance between reality and foresight, Yu Yizhou should not only lead the team to do technical research on specific projects, but also cooperate with external universities and scientific research institutes to carry out forward-looking technology research and development These scientific research cooperation covers a wide range of fields, including common breast cancer, pulmonary nodules, stroke scenes, and pancreatic cancer, which is relatively small in the AI community, laying a solid foundation for the whole disease strategy of Shenzhen Rui Medical Co., Ltd In addition to the medical scene, Shenzhen Rui medical also has many scientific research achievements in the field of basic computer vision Only in 2019, eight papers of Shenzhen Rui Research Institute were selected into the top conference of artificial intelligence cvpr2019, which achieved innovative breakthroughs in image recognition, medical image analysis and other technologies, and ranked among the top scientific and technological companies in the number of papers published in China At the top international conference on medical image analysis (MICCAI) held in October and the top international conference on computer vision (iccv) held in November this year, another 10 scientific research papers on the field of medical artificial intelligence of Shenzhen Rui Research Institute were included In addition, Shen Rui has two papers included by ER and Mia Up to now, Shenzhen Rui Research Institute has published more than 50 top academic papers, with a cumulative impact factor of more than 80 and a paper acceptance rate of more than 50% Sunray Research Institute has published nearly 30 papers on top journals and conferences of artificial intelligence and machine learning (such as science robotics, tpami, TCYB, tip, ICML, CVPR, iccv, ECCV, AAAI, etc.), including three top international conferences in the field of computer vision and pattern recognition, especially in the top conference CVPR (top 10 of Google 2019 academic list) which has attracted great attention for two consecutive years Academic achievements published