Ande BioMind: What is a high-IQ medical AI?

How does AI technology empower China's medical digital transformation?

From July 7th to July 10th, the 2021 World Artificial Intelligence Conference (WAIC) with the theme of "Wisdom Connected World, All Wisdom Becomes a City" was held in Shanghai. Many Turing Award winners, more than 30 CEO-level speakers, dozens of top Chinese and foreign academicians, and more than 200 leading scholars in the global AI field-gathered here. Among them, BioMind, a medical artificial intelligence company, has been invited to participate in the World Artificial Intelligence Conference for three consecutive years.

At the "Digital Health, Smart Health" Health Summit Forum, the conference discussed topics such as the development of artificial intelligence medical technology, the empowerment of the biomedical industry, and the new medical infrastructure in the context of the post-epidemic situation. In addition, the conference released the evaluation and acceptance results of the medical track evaluation and acceptance results of the new-generation artificial intelligence industry innovation key tasks of the Ministry of Industry and Information Technology, and BioMind was commended.

How AI technology can empower China’s medical digital transformation is the focus of this forum. Andre BioMind Greater China CEO Li Jingjue said at the forum that as the new medical infrastructure and the country’s major disease control system become more complete, medical labor The smart future will not only empower hospitals' digital transformation, but also contribute to the intelligent construction of the national disease control system and public health capabilities.

During the COVID-19 pandemic in 2020, medical imaging artificial intelligence companies have gradually landed from the "cloud" and played an important role in artificial intelligence screening of lung disease images. In the post-epidemic era, most medical imaging artificial intelligence products that have emerged at this stage are far from meeting clinical needs, and more application scenarios must be explored. Therefore, only AI products that can achieve a precise diagnosis, precise decision-making, and precise treatment of integrated diagnosis and treatment can truly lead to the new era of medical artificial intelligence.

Li Jingjue believes that only high-quality data can produce high-IQ medical AI.

"Medical + AI" is the core of the new era of integrated diagnosis and treatment

A large number of entrepreneurs have emerged in the medical imaging AI industry in China, and many AI imaging products have been developed in the rapid iteration of algorithms and big data technology. This can be seen from the amount of financing: 2015-the first half of 2020, AI imaging The amount of financing is nearly two hundred.

But at the moment when AI image products are blooming everywhere, the crisis will also follow. The advent of the new era of integrated medical artificial intelligence diagnosis and treatment is forcing AI imaging companies to rethink the value of AI imaging products to medical care, and how AI imaging can truly empower medical digital transformation.

Shen Nanpeng, the global managing partner of Sequoia Capital, expressed the same judgment in his keynote speech at the opening ceremony of the conference: If you regard "computing power level" and "application scenario", you can visually regard the two aspects of AI in the field of life. In terms of legs, we can clearly see that the "computing power" leg is very long and strong, and it grows exponentially. The parameter of the largest deep learning model in 2020 is the 100 billion level, and it has reached the trillion level at the beginning of this year. But "under the premise of exponential growth in computing power, there is still much room for improvement in data mining for life segmentation scenarios."

Li Jingjue, CEO of BioMind Greater China, proposed that the core of the new era of integrated medical artificial intelligence diagnosis and treatment is "medical + AI", that is, driven by clinical needs rather than AI technology. "This core contains two major connotations. One is that AI products need to have a deep understanding of clinical needs and clinical pain points, not only to help imaging doctors assist in diagnosis, but also to help clinical department doctors assist decision-making and assist treatment; the other is high-quality data from top medical institutions To become the quality assurance of high-quality AI products, high-quality data can produce AI with high IQ."

Only AI products driven by this kernel can realize an integrated solution of precise diagnosis, precise decision-making and precise treatment.

During the epidemic, BioMind responded quickly and launched the BioMind artificial intelligence imaging-assisted diagnosis system for acute infectious pneumonia, which was sent to designated hospitals in Hubei and various places as soon as possible, and it was able to perform qualitative diagnosis of new coronary pneumonia, and the diagnosis accuracy rate exceeded 94. %, not just pneumonia screening. In addition, BioMind can also realize the simultaneous analysis of multiple diseases based on CT and MR application scenarios. For example, based on chest CT, it can realize the qualitative diagnosis of new coronary pneumonia, SARS, other viral, bacterial, and fungal pneumonia at one time, and Auxiliary diagnosis of tuberculosis, lung cancer, pulmonary nodules, emphysema, etc.

This is the main goal of Ande Medical Intelligence BioMind in the new era of "medical + AI" diagnosis and treatment integration: not only to help clinicians shorten diagnosis time, improve clinical diagnosis efficiency, but more importantly, based on the qualitative analysis capabilities of multiple diseases and The extremely high accuracy rate can provide basic-level clinicians with auxiliary diagnosis and decision-making to the greatest extent, truly help graded diagnosis and treatment, realize the sinking of high-quality medical resources, and improve the diagnosis and treatment capabilities of basic-level hospitals.

Clinical needs determine the direction of medical AI

Aiming at the first connotation of the new era of integrated medical artificial intelligence diagnosis and treatment, "in-depth understanding of clinical needs and pain points", the BioMind R&D team of Ande BioMind has been rooted in the hospital for a long time since its inception, in-depth communication and close cooperation with clinical experts, truly Achieve from the clinic to the clinic.

Through in-depth communication, the BioMind team found that a major clinical pain point is that the more complex and difficult to diagnose diseases, the more they need the help of AI image-assisted diagnosis, especially at the grassroots level.

Take intracranial tumors as an example. Intracranial tumors are a general term that includes dozens of brain tumors. The complexity of the disease is high, and the mortality rate in the world is gradually increasing. There is an urgent need for AI image-assisted diagnosis of intracranial tumors in primary clinics, but there are only a handful of domestic companies that study MR intracranial tumors.

This is because the imaging of intracranial tumors has the phenomenon of "the same disease with different shadows" and "same shadow with different diseases". The diagnosis cannot be made by imaging alone. The patient's medical history, clinical symptoms, signs and other related auxiliary examinations must be used as Judgments based. This has brought great difficulties to the diagnosis of primary clinicians, and the misdiagnosis rate is high. Some clinicians often use pathological examination or needle biopsy after a craniotomy to determine what tumor is, but in fact, this requires extremely high technical requirements and is even more difficult for primary clinicians.

Due to the dual considerations of the pain points of clinical needs and the difficulty of research and development, the BioMind team of Ande Medical Intelligence chose to conduct research from intracranial tumors first. The research goal is that in addition to screening for disease, AI imaging can also perform specific qualitative determination, that is, "what kind of intracranial tumor is it?", to provide doctors with very accurate auxiliary diagnosis, based on different diseases, doctors will choose in time Correct treatment methods to improve patient prognosis.

At the same time, there is still a big demand for clinics: a complete form of clinical multitasking products that meet the clinical scenario is more needed. Many grassroots clinicians look forward to scanning one part at a time, allowing AI imaging to assist in the diagnosis of all diseases that may occur in that part.

But in fact, most of the current AI imaging products are cut from a single disease in a single location, and are a simple AI model for a single disease. The "single-soldier combat" model is very limited inefficiency in the diagnosis field for grassroots doctors.

Based on the above clinical pain points and needs, in June 2018, Ande Medical Intelligence BioMind, in conjunction with Tiantan Hospital, National Neurological Disease Clinical Medicine Research Center, National Neurological Disease Quality Control Center, Chinese Stroke Society, and Neurological Disease Artificial Intelligence Center, jointly released The world's first neuroimaging artificial intelligence-assisted diagnosis product-BioMind.

BioMind can identify 27 kinds of intracranial tumors, cerebrovascular malformations, aneurysms, cerebral small vessel disease, ischemic stroke, cerebral hemorrhage, brain trauma, ischemic penumbra, and atherosclerosis through CT and MR images. There are more than 60 kinds of brain diseases, such as arterial stenosis, and the diagnosis accuracy rate exceeds 90%, and the diagnosis accuracy rate of some diseases exceeds 96%. It helps doctors make a rapid diagnosis and improves doctors' diagnosis and treatment capabilities.

High-quality data determines the height of medical AI

Medical clinical diagnosis is by no means only relying on single-dimensional data. AI imaging products on the market that are only grown by learning image data are still far from being “accurate”.

Aiming at the second connotation of the new era of integrated medical artificial intelligence diagnosis and treatment, "High-quality data from top medical institutions become the quality assurance of high-quality AI products. Only high-quality data can give birth to high-IQ AI." As early as December 2017, Ande BioMind and Beijing Tiantan Hospital jointly established the "Neurological Disease Artificial Intelligence Research Center" to carry out comprehensive and in-depth cooperation.

At the same time, BioMind has also established close strategic partnerships with top scientific research and medical institutions such as the Massachusetts Institute of Technology (MIT) Artificial Intelligence Laboratory and the National University of Singapore.

Li Jingjue said, “The multi-dimensional data that combines medical record data + imaging data + pathological data from top hospitals is high-quality data.”

With AI technology alone without high-quality data, there have been cases of failure in foreign countries. IBM Watson Health and Google Health, which were once considered to be the pioneers of medical AI, both failed due to data acquisition problems. Watson Health was unable to obtain high-quality data in the later period because of the high error rate. In 2016, Google's research institution DeepMind cooperated with three British hospitals to obtain medical data of 1.6 million patients, and then obtained and used data due to violations of regulations. Be investigated.

China is a populous country and generates a large amount of medical data every year. This is precisely the inherent advantage that foreign AI imaging companies do not have.

Through scientific research cooperation with Tiantan Hospital, BioMind uses deep learning algorithm models to systematically train millions of imaging cases in the past ten years, incorporating the clinical experience of top hospital experts. While using high-quality data from top hospitals, in order to ensure data security, Ande BioMind dispatched a research and development team to work in the hospital, so that all training data is not discharged from the hospital, and the data is cleaned and desensitized training is all in the hospital. The intranet is completed.

In addition, the core technical team of Ande Medical Intelligence BioMind comes from the National University of Singapore, Harvard University, Massachusetts Institute of Technology, Tsinghua University, Chinese Academy of Sciences and other top universities in the world. The blessing of a strong team is the technical level of the products of Ande Medical Intelligence. Provides a strong guarantee.

High-quality data and a strong technical team have become the source of continuous innovation of Ande Medical Intelligence. The solid product strength has helped Ande Yizhi successfully open the international market. In 2018, Ande Yizhi obtained product certifications from more than a dozen countries including the European Union CE and Singapore. The products are sold to Germany, Poland, Luxembourg, Singapore, etc. Countries and regions.

Application scenarios determine the future of medical AI

Li Jingjue said at the forum that the value of medical AI lies in solving clinical needs. In recent years, many companies have failed on this track precisely because they did not really understand the market demand and only did what they wanted to do. "If AI products cannot be fully embedded in medical scenarios, products that only improve efficiency are software tools, not real artificial intelligence."

Most AI image products on the market exist in the form of immature software, and the algorithm model is still in the training and optimization stage, and large-scale applications have not been completed.

In July 2020, BioMind, as one of the joint member units of the international project, became the only target of the national-level medical artificial intelligence public platform project of the Ministry of Industry and Information Technology of the People's Republic of China-"A public service platform for artificial intelligence assisted diagnosis in the medical and health industry" , The winning bid amount reached 168 million yuan.

At present, BioMind has gradually realized the application of accurate auxiliary diagnosis and decision-making assistance covering multiple parts of the body, such as the head, neck, heart, blood vessels, and breast. Before 2023, Ande Medical Intelligence will complete the whole-body system imaging AI-assisted diagnosis And a comprehensive layout of decision-making assistance products for multiple diseases.

But Ande Medical Intelligence is not satisfied with this. Li Jingjue said that Ande BioMind is gradually moving from auxiliary diagnosis to precise diagnosis, precise decision-making and precise treatment, trying to create an AI precision diagnosis and treatment integrated system that is more needed by grassroots clinics to help the development of grassroots hospitals.

For example, the BioMind cerebrovascular disease clinical diagnosis and treatment assistant decision-making system ("Cerebrovascular Disease CDSS") developed by Ande Medical Intelligence BioMind, assists doctors in the whole process of screening-diagnosis-treatment-prognosis, and helps doctors achieve Scientific and reasonable clinical decision-making for discharge.

Li Jingjue believes that artificial intelligence is most likely to take the technology of top hospitals to the grassroots level and improve the level of diagnosis and treatment of grassroots doctors. Ande BioMind has been providing global solutions that support the comprehensive improvement of imaging and clinical capabilities. In the future, it will continue to base itself on medical treatment and deepen algorithms to provide imaging doctors and clinicians with integrated diagnosis and treatment intelligent solutions that are more suitable for clinical application scenarios.