On February 12, 2019, a scientific research achievement of diagnosing pediatric diseases by using artificial intelligence technology was announced: the system has a diagnostic accuracy rate for 55 common pediatric diseases and some critical and severe diseases, which has exceeded that of ordinary young doctors.
This is also a major breakthrough for Chinese institutions in AI artificial diagnosis and treatment up to now.
The initiator of this research is Yitu Medical Ni Hao’s team ( the company’s algorithm team ). As the only medical artificial intelligence company covering full-link medical intelligence in China at present, Yitu Medical is also the first company to carry out medical artificial intelligence landing practice in China. Its products cover many fields such as intelligent medical imaging, intelligent clinical big data, intelligent outpatient optimization, intelligent quality control, etc.
According to the response of medical professionals in Yitu, the system began to be put into operation in Guangzhou Women’s and Children’s Medical Center in December 2018.
Will this achievement of Yitu Medical Care drive China’s AI medical investment to the next climax? Looking at the whole AI medical field, the giants are still entering and distributing, and capital is still entering in high profile. AI research statistics show that China’s artificial intelligence medical market will exceed 38 billion yuan in 2018 and is expected to reach 10 billion US dollars by 2020.
As a matter of fact, the problems in this field are also more prominent. For example, the product functions of most AI medical imaging companies are limited to a single disease subdivision, but this does not seem to hinder the hot trend in this field.
In January 2019, AI medical imaging became the most mature field of artificial intelligence medicine in China, according to the White Paper on Artificial Intelligence Medicine released by Shanghai Jiaotong University’s Institute of Artificial Intelligence and Shanghai Health and Health Development Research Center.
Ni Hao, president of Yitu Medical, believes that in 2019 the AI medical industry, the bubble will gradually be swept away and the players who have worked hard will still have high tides.
According to Sun Xin, director of the medical department of Guangzhou Women’s and Children’s Center, in just 20 days, doctors have actually used artificial intelligence technology to diagnose pediatric disease intelligence system to carry out 30,276 auxiliary diagnoses, with a coincidence rate of 87.4%.
Sun Xin said that it treated patients like human doctors. Doctors input the patient’s chief complaint, symptoms, personal disease history, physical examination, laboratory test results, imaging test results, medication and other information into the medical record text, and the system automatically converts the free medical record text into standardized, standardized and structured data. After ” reading” the medical record, the system gives the diagnosis result.
Taking respiratory diseases as an example, the diagnostic accuracy rates for upper respiratory diseases and lower respiratory diseases are 89% and 87% respectively, while in the diagnosis of upper respiratory diseases, the accuracy rates for acute laryngitis and sinusitis are 86% and 96% respectively, and the diagnostic accuracy rates for different types of asthma are 83% to 97%. At the same time, it has high diagnostic accuracy for common systemic diseases and diseases with higher risk, such as infectious mononucleosis ( 90% ), chicken pox ( 93% ).
Different from the previous artificial intelligence system, the system can not only look at pictures statically, but also ” read” and learn more data than ever before. On February 12, this feature, as the first research achievement in the world, was published in a top medical journal, namely, the research achievement on the clinical intelligent diagnosis of natural language processing ( NLP ) technology based on Chinese text electronic medical record ( EMR ).
Regarding this achievement, other AI medical professionals in China said that the technical index is not the decisive factor, but the clinical application is the key point. In the future, enterprises will mainly compete for practicality.
It is understood that in cooperation with Guangzhou Women and Children’s Center, Yitu collected 1.362 million electronic medical records of 567,498 outpatients from January 2016 to July 2017, and extracted 101.6 million data points covering 55 common diseases in pediatrics for initial diagnosis.
According to Ni Hao, president of Yitu Medical, the core technology of this achievement is to deconstruct the electronic medical record data through in-depth learning technology and medical knowledge map, thus constructing a high-quality intelligent disease database and establishing various diagnostic models on this basis.
It is understood that the data in the medical scene in this intelligent system include image data, electronic medical record data, laboratory test data, etc. Data in different formats correspond to different branches of artificial intelligence technology – image recognition, voice recognition and NLP technology, among which NLP technology is especially complex. Microsoft founder Bill Gates, for example, called it ” the pearl in the crown of intelligent field”. ” When doctors make a diagnosis, they must not only look at the image report, but also the medical records and test data. This is a comprehensive data analysis process.” Fang Cong, vice president of Yitu Medical, stressed many times in an interview with this reporter that whether a product can really meet clinical needs is the key. Different scenarios have different requirements for algorithms, and a single algorithm is difficult to meet all clinical scenarios. AI Company needs to break through the difficult problems of text data conversion, complex image extraction, data standardization and so on in order to become a first-class medical artificial intelligence company.
According to Fang Qian, as early as November 2018, the top-level conference in the field of natural language processing EMNLP2018, the picture-based paper preco: a large-scale dataset in prechoolvo – cable for coherence resolution was employed as an Oral article. In June of the same year, Yitu Medical and West China Hospital jointly released the country’s first lung cancer clinical research intelligent disease database, which is also the largest lung cancer intelligent disease database in the field of lung cancer today. It is reported that the results of the first phase have been put into clinical trials in dozens of top hospitals in China, and a multi-center clinical trial led by West China Hospital is about to start.
In this study, the amount of data included is the largest amount of real clinical data processed by artificial intelligence enterprises based on NLP technology so far. A total of 1.36 million high-quality real electronic medical records from 567,000 children patients were studied, and more than 101 million data points were collected. In this scientific research achievement, a system framework for data mining of electronic medical cases is proposed and tested according to chart medicine. EMR data is deconstructed through NLP technology and medical knowledge mapping technology, thus a high-quality intelligent disease database is constructed. Based on this, a disease diagnosis system covering more than 80% of common pediatric diseases is established by using logisticregressionclassifiers.
Liang Huiying, director of the data center of Guangzhou Women’s and Children’s Medical Center, revealed that after three months of improvement and iteration, the intelligent system will have more than 30,000 calls in the first quarter of 2019. In the future, the imagination of the system will be even greater. It can be used as triage program. For example, when a patient comes to the emergency department, the nurse can obtain his vital signs, basic medical history and physical examination data, and then input them into the model, allowing the algorithm to generate predictive diagnosis and helping doctors to screen which patients to give priority to diagnosis and treatment. It can also help doctors diagnose complex or rare diseases. In this way, doctors can use AI – generated diagnosis to help broaden differential diagnosis and consider diagnostic possibilities that may not appear immediately.
According to Ni Hao, a number of products based on NLP technology such as ” AI guided diagnosis”, ” AI pre – interrogation”, ” AI pre – examination”, ” AI auxiliary diagnosis” have been put on the ground in hospitals, including Guangzhou Women’s and Children’s Center, Shanghai Children’s Medical Center, Zhongshan Hospital affiliated to Xiamen University, Second Hospital affiliated to Wenzhou Medical University and other medical institutions.
So far, the layout of the three major technical fields of artificial intelligence, namely image recognition, speech recognition and natural language understanding, has been preliminarily completed by the three-year-old Yitu medical service.
Perhaps, in the world of Yitu Medical, the layout of AI medical will be wider and deeper in the future.
Breaking the Single Malpractice
In AI medical circles, people are used to calling 2016 ” the first year”.
This year, AlphaGo defeated Li Shishi and almost all the people knew about artificial intelligence.
By 2017, IBM’s Watson tumor assistant diagnosis system will enter China, and AI medical treatment will set off a boom in the entire medical field.
Medical imaging has become an indispensable part of China’s artificial intelligence industry due to its data advantages and industry characteristics. In the past three years, most enterprises have expanded from single image recognition to multiple branches of artificial intelligence technology. From the detection of a single lesion to the subdivision of diseases, to all-position diagnosis; From several hospitals to the ” arms race” of thousands of tertiary hospitals in China.
If Ni Hao is shown, he feels that the consciousness of the founder of Yitu Medical Treatment woke up earlier. In 2012, two rooms and one hall in a residential building near Shanghai Jiaotong University rang with intense keyboard tapping all day long and remained brightly lit until late at night.
” Ideaischeap. Thedevilisinstead – tail”, written by Zhu Long, co-founder of Etu Science and Technology and UCLA Doctor of Statistics, on a blackboard with colorful graffiti, ” Superman”, ” Robot” and ” AI” are scattered in simple strokes. That year, he established Etu Science and Technology jointly with Lin Chenxi.
In 2013, the first set of face recognition algorithms of Yitu Technology landed and were adopted by Suzhou Public Security Bureau. At the same time, it obtained the angel wheel financing from real fund. For the next four years, Eto Technology has been pioneering in a particularly vertical area like security.
Until 2016, Yitu set up a medical subsidiary.
It is understood that the first product of Yitu Medical has entered Zhejiang Provincial People’s Hospital, which is also the first artificial intelligence medical solution connected with clinical workflow in the history of AI medical development in China. At present, the artificial intelligence solution of Yitu Medical has entered nearly 200 3A hospitals in the country.
Along the way, the artificial intelligence solution for medical treatment based on maps has also increased from a single lung product to three major product lines now, including imaging system, intelligent data system and internet paediatric system, with accompanying diseases including lung, breast, stroke, paediatric, growth and development, pathology and ultrasound.
In fact, the development process of Yitu medical care is behind the development of AI medical care in China.
On July 20, 2017, the State Council issued the ” Development Plan for New Generation of Artificial Intelligence”, formally proposing the top-level strategic plan for the development of artificial intelligence in China. Artificial intelligence rose to the national strategy. On April 28, 2018, the general office of the state Council officially issued the ” opinions on promoting the development of” internet + medical health ” ( hereinafter referred to as” opinions ” ), which clearly stated that artificial intelligence assisted diagnosis system can improve the efficiency of medical services; The new edition of the Classification Catalogue of Medical Devices, which was implemented on August 1, 2018, sets a path for AI medical products to go on the market.
With the encouragement of policies and the stimulation of the overall size of the medical service market, some BAT giants and data companies have laid out the AI medical field one after another. For example, it is assumed that science and technology, advanced medical science and technology, and medical science and technology will be combined together.
Medical imaging can become a hot topic in AI medical treatment, which has its own characteristics. Imaging is the most basic link in modern diagnosis and treatment. With the rapid development of imaging technology, imaging technologies such as X – ray, CT, MRI and ultrasound have gradually become important means of tumor detection, staging and follow – up, 80% of medical data come from medical images. At the same time, artificial intelligence technology also has a broader landing scene in the field of medical imaging. Analysis of a large number of imaging data and functional imaging data not only improves the accuracy of disease diagnosis, but also increases the complexity of disease diagnosis and dependence on physician experience.
In the past three years, with the continuous entry of AI medical enterprises, AI medical application scenarios have expanded from the previous single lung nodule detection to breast cancer, orthopaedics, pediatrics and other fields. However, at this level, there is still a problem that most AI medical imaging companies have made remarkable achievements in many subdivided areas, such as lung nodule screening, diabetic eye disease, hemorrhagic stroke, bone age detection of children, but their application scope is relatively narrow. For example, in the field of lung nodule detection, most AI applications can only look at lung nodules, but lung images also include pneumonia, pulmonary tuberculosis, chronic obstructive pulmonary disease, bronchiectasis, pleural effusion and other diseases.
In Fang Qian’s view, a mature product should be based on a large number of real medical data to develop products, deeply understand clinical pain points, be able to integrate into clinical workflow, and help doctors solve practical problems. Take the intelligent detection in the field of pulmonary nodules as an example. In the doctor’s daily work flow, pulmonary nodules are only a focus examined by the imaging doctor when he looks at chest radiographs. Chest CT is not only used to look at pulmonary nodules. Therefore, AI system that can only detect pulmonary nodules cannot become the doctor’s real artificial intelligence assistant.
Fang Cong said that for a long time, most of the pulmonary imaging diagnostic products of medical AI enterprises have been limited to a single task of nodule detection. Nodules only account for 60% of all lesions seen on pulmonary CT images, and the rest of the lesions accounting for about 40% of the total number of lesions can not be detected, such as plaque, streak, cystic shadow, etc., and mediastinal and pleural lesions closely related to the lung can not be detected.
” Pain point, itch point, Sao point”
Relevant statistics show that [c3] based on the market scale of the medical industry, the market prospect of medical artificial intelligence is also very broad. In 2017, the scale of China’s medical artificial intelligence market will exceed 13.6 billion yuan. With the deepening of application, the market scale is expected to exceed 50 billion yuan in 2019. In 2018, the scale of China’s artificial intelligence market exceeded 38 billion yuan.
Public data show that [c4] As the population aging trend highlights and the prevalence rate rises, China’s medical expenditure has increased rapidly in recent years. From 2012 to 2016, the total medical expenditure increased from 281.2 billion yuan to 460 billion yuan.
With the help of artificial intelligence technology, it is expected to greatly improve the efficiency and level of medical diagnosis and treatment, reduce the related cost investment of medical institutions, liberate more medical resources, and bring considerable benefits to patients. Therefore, based on the market scale of the medical industry, the market prospect of medical artificial intelligence is also very broad.
Zhang Kan, director of science and education department of Shanghai municipal health Committee, said at the ” sino – german cooperation AI medical forum” that artificial intelligence is a ” disruptive innovation” for the medical industry. Combining artificial intelligence to realize intelligent medical treatment is not only an innovative exploration, but also a general trend under the background of the times.
According to the global trend, AI medical treatment is developing rapidly in many countries. As of the first half of 2018, the U.S. Food and Drug Administration ( FDA ) has approved nine products related to artificial intelligence, including automatic monitoring and early warning products and auxiliary diagnosis products, which have been applied in many hospitals. Japanese hospitals began to experiment with and try out artificial intelligence systems, especially in the field of image-assisted diagnosis, so as to improve the supply capacity of medical services in Japan.
Relevant data show that by 2017, AI medical imaging will already be the undisputed star in the capital market. According to rough statistics, the number of financing enterprises of AI medical imaging enterprises in 2017 will reach 19, with a total scale of more than 1 billion yuan.
In January 2019, the ” White Paper on Artificial Intelligence Medicine” released in Shanghai on the 9th by Shanghai Jiaotong University’s Artificial Intelligence Research Institute, Shanghai Health and Health Development Research Center and Shanghai Jiaotong University’s Medical College showed that 19 provinces and cities across the country have issued artificial intelligence plans, and AI medical imaging has become the most mature field of artificial intelligence medicine in China.
But bubbles always coexist with tuyeres. ” When an industry becomes very hot, it must be