Recently, the 2022 Medical Artificial Intelligence Conference (CMAI 2022) and the 2nd “China Medical Academic Journal Development” high-end forum were held.
Leifeng.com’s “Medical Health AI Nuggets” is the support unit of this conference. Qi Honggang, a professor at the School of Computer Science, University of Chinese Academy of Sciences, and Wang Liqun, a professor at the School of Science, China University of Petroleum (Beijing), served as the hosts of the conference.
This summit forum invited several directors of radiology departments of top hospitals and authoritative experts of artificial intelligence technology to discuss the clinical application and scientific research progress of artificial intelligence technology in medical imaging, and share their research experience.
(We will release the in-depth dialogue and speech content of each speaker in the follow-up, welcome to pay attention)
Huang Wei, Academician of the Chinese Academy of Sciences
Huang Wei, academician of the Chinese Academy of Sciences and editor-in-chief of Research, delivered a speech on behalf of the CMAI conference. He said that the powerful empowering effect of artificial intelligence on various industries has already emerged. Biomedical is a data-intensive, brain-intensive, and knowledge-intensive industry, which needs to rely on powerful analysis and processing capabilities for judgment and diagnosis and treatment. Prospects for artificial intelligence applications.
Over the years, my country has intensively issued a series of policies and regulations on medical artificial intelligence, aiming to establish a fast and accurate intelligent medical system. In the “14th Five-Year Plan”, both artificial intelligence and life and health are listed as priorities in the field of cutting-edge science and technology, which will accelerate a new round of rapid development of my country’s artificial intelligence and life and health sciences.
“Although medical artificial intelligence has entered a period of rapid development, it still faces many challenges. We sincerely hope that this conference will become an opportunity for everyone to collide with ideas, deepen exchanges, and expand future synergy and collaboration in artificial intelligence and biomedical industries. Cooperation.”
Liu Shiyuan, Chairman of the Radiology Branch of the Chinese Medical Association
Liu Shiyuan, Chairman of the Radiology Branch of the Chinese Medical Association and Director of the Department of Imaging Medicine and Nuclear Medicine of Shanghai Changzheng Hospital, as the first guest speaker, shared the latest news based on the upcoming “Report on the Development of Artificial Intelligence in Medical Imaging in China (2021-2022)” The basic situation of industry development.
Liu Shiyuan said that as of May 31 this year, there have been more than 30 three types of medical imaging AI registration certificates approved by the State Food and Drug Administration, covering CT, magnetic resonance, DR and other equipment, including cardiovascular and cerebrovascular, chest diseases, diabetes, bone and bone diseases. Products for joint disease and child development assessment.
“Our AI products, from initial skepticism to overly optimistic vision, have gone through a stage of calm and rationality, and now we have entered a new period of clinical application and commercialization.”
The approved medical imaging AI products can be divided into two major aspects, one is products that optimize the medical imaging workflow, and the other is a disease-centric diagnosis model.
Specifically for the former, it has become the norm for medical imaging equipment to be empowered by deep learning technology. It is expected that by 2023, the penetration rate of AI to CT will increase to about 50%, and the penetration rate of MRI and ultrasound will increase to about 40%.
The latter is the area where the most companies invest a lot of energy in research and development, and the most mature products are pulmonary nodules and coronary CTA.
As of 2022, AI products based on disease models have gradually iterated from lesion detection and segmentation to multi-dimensional, multi-functional, and even multi-task models that combine morphological diagnosis and functional diagnosis, forming disease scenarios as the center. platform application.
Professor Liu Shiyuan introduced that medical imaging AI products have been gradually introduced into hospitals. This year, the AI penetration rate of large hospitals in my country is about 15%, and it is expected to reach more than 30% next year.
It is worth noting that although more than 50% of AI products used in hospitals are obtained through purchase, more than 94% of patients are free trials. Judging from the remaining 5% or so of charging cases, the main methods of charging include diagnosis, consultation, examination and package charging, which have not yet been regarded as separate charging items.
“This shows that AI products are not mature enough to make patients have a strong willingness to buy, and the product form and commercialization model of AI need to be continuously improved.”
Li Hongjun, Director of the Department of Radiology, Beijing You’an Hospital
Li Hongjun, director of the Radiology Department of Beijing You’an Hospital, shared the theme of “The Role and Value of Medical Imaging in Medical Big Data”.
Li Hongjun introduced the development of medical imaging, the connotation of medical imaging data, the value of medical imaging data, and the mining and utilization of medical imaging big data.
Li Hongjun said that in recent years, the application of AI and the upgrade of AI algorithms have driven the entire medical imaging to enter a new stage, including the selection of regions of interest, segmentation and image processing to reduce image interference factors, improve algorithm efficiency, and improve results. more precise.
Taking the extraction of coronary tree as an example, AI’s understanding of the global structure, compensation of effective information, and repair of weak signal breaks can achieve the most effective generation, actively remove and repair artifacts, and display in all-round 3D Morphology of the entire coronary image.
Li Hongjun believes that the occurrence and development of each disease is not a single data change, but a multi-omics change. “Our radiomics must be integrated with multiple data models such as clinical data characteristics, proteomics, genomics, metabolomics, and socioomics, so that we can comprehensively and objectively reflect the occurrence, development and prognosis of individual diseases.”
This also means that the traditional morphological imaging diagnosis mode can no longer meet the requirements of precision medicine.
Li Hongjun said that the early AI was only based on image and data characteristics for early warning and prediction of diseases, which deviates from the biological meaning. The combination of radiogenomics and AI will be the development and extension of morphological imaging, which can solve diseases that are invisible to the naked eye. , to achieve the diagnosis of diseases without symptoms and signs.
Wu Jian, Professor of Zhejiang University and Changjiang Scholar
Wu Jian, a professor of Zhejiang University and a Changjiang scholar, shared the market background, industry status, bottlenecks and difficulties, solutions and stage results of AI-assisted ECG diagnosis with the theme of “Artificial Intelligence ECG-assisted Diagnosis”.
Wu Jian introduced that the application of ECG examinations in my country is very demanding. At least 250 million people undergo ECG examinations every year, but they are faced with the problems of insufficient ECG and cardiovascular doctors and low accuracy of equipment detection.
The project research done by Wu Jian mainly focuses on four major goals, namely, the use of AI algorithms combined with signal processing methods to perform automatic ECG analysis, establish a monitoring model for abnormal ECG events, establish a cardiovascular disease discrimination model, and provide doctors with ECG marking tools. And establish an auxiliary diagnosis platform.
During the exploration of AI-assisted ECG diagnosis, Wu Jian’s team found that they encountered bottlenecks in five aspects: data collection, data cleaning, data labeling, heartbeat recognition and model building.
In response to these bottlenecks, the algorithm framework developed by Wu Jian’s team has innovative points such as convolution feature description global information, frequency domain analysis feature supplementary detail information, fast and accurate, batch operation and so on.
At present, Wu Jian’s team has obtained more than 2 million pieces of ECG data, and organized more than 100 types of labels, covering 99% of ECG diagnosis categories. , the overall accuracy rate reaches 95%, and the F1 reaches 91%.
In addition, the team also successfully developed an ECG band labeling tool and an intelligent ECG-assisted diagnosis system.
Cui Guangbin, Director of the Department of Radiology, Tangdu Hospital, Air Force Military Medical University
Cui Guangbin, director of the Department of Radiology, Tangdu Hospital, Air Force Military Medical University, shared the theme of “The Application Status, Challenges and Prospects of AI in Pulmonary Nodules”.
Cui Guangbin said that lung cancer is the type of cancer with the largest number of new cases and deaths in my country. In order to reduce the incidence and mortality of cancer, the “Healthy China Action” requires the promotion of early cancer screening, early diagnosis and early treatment.
The configuration and popularization of CT have made the basic hardware conditions for CT lung cancer screening in my country. However, in public hospitals of different levels, there are pain points such as shortage of radiologists and lack of experience in reading images and diagnosis. Medical imaging AI is to solve these pain points. an important method.
With the implementation of medical imaging AI in hospitals, Cui Guangbin found that there is a certain disconnect between AI products and actual clinical application needs. AI companies can’t be fully satisfied either.”
Taking CT screening for the new crown as an example, where AI focuses on improvement is also what doctors can do with the naked eye, such as the scope of lesions, which is of little practical significance. The bedside films taken by X-ray machines for critical cases are overlapping images, and manual review will have many uncertain factors hindering diagnosis. This is an area where AI is useful, but it has not yet been resolved.
Peng Shaoliang, Professor of Hunan University and Changjiang Scholar
Peng Shaoliang, professor of Hunan University and Changjiang scholar, shared the theme of “Supercomputing-based Metaverse Digital Therapy and Electronic Medicine”.
Shaoliang Peng introduced in detail the role of digital therapy and a number of digital therapy cases around the world.
Peng Shaoliang believes that digital therapy has many advantages, which can accelerate the effect of treatment, shorten the cycle of treatment, and reduce the cost of treatment. Compared with traditional chemical drugs, the development speed of digital therapy is faster.
“The research and development of a new drug takes more than 5 to 10 years, and the minimum cost is 1 billion US dollars. On the contrary, digital therapy is a software, which does not require such a long time and such a large cost. In the future, we only need to verify the validity of data and algorithms. .”
Peng Shaoliang said that my country’s exploration in the field of Metaverse medical treatment and digital therapy is still blank. He hoped that hospital societies, medical companies, IT game companies, etc. will work together to establish the first domestic Metaverse Medical and Digital Therapy Alliance, focusing on adolescent depression and Alzheimer’s disease. and a series of diseases with less international distribution, and launched the first digital prescription standard and electronic medicine in China.
Lu Jie, Vice President of Beijing Xuanwu Hospital
Professor Lu Jie is the vice president of Xuanwu Hospital of Capital Medical University and the director of the Department of Radiology and Nuclear Medicine. At the meeting, he shared the topic of “Application Research of Artificial Intelligence in MRI Imaging of Brain Demyelinating Diseases”.
Lu Jie introduced that multiple sclerosis (MS) and neuromyelitis optica (NMOSD) are common brain demyelinating diseases, and they are also common diseases of the nervous system that cause disability among young and middle-aged people. There are about 2.8 million multiple sclerosis patients in the world. About 42,000 people.
Differential diagnosis of MS and NMOSD is challenging due to similar clinical symptoms, overlapping laboratory results, and long diagnosis cycle, especially for primary hospitals and junior physicians.
In clinical practice, magnetic resonance imaging (MRI) evaluation is an important part of the diagnosis of MS and NMOSD. With the development of artificial intelligence technology in recent years, the application of artificial intelligence in MRI imaging of brain demyelinating disease has also made great progress.
Lu Jie pointed out in the report that artificial intelligence technology can mine high-level quantitative features that cannot be recognized by the naked eye in image images, and the artificial intelligence model based on topology will have important value in predicting the prognosis of brain demyelinating disease.
Liu Yong, a professor at Beijing University of Posts and Telecommunications
Professor Liu Yong is a professor at the School of Artificial Intelligence of Beijing University of Posts and Telecommunications. He shared the title of “Research on Radiomic Characterization of Alzheimer’s Disease Based on Magnetic Resonance and PET Imaging”.
He said that the study of machine learning methods on medical images has been widely carried out, and researchers have been working hard to explore objective quantitative, reproducible and biologically meaningful imaging markers of neuropsychiatric diseases.
Liu Yong’s research team has been conducting research on how to characterize abnormal brain imaging in Alzheimer’s disease (AD) for more than ten years, exploring the feasibility of using magnetic resonance imaging to study early imaging markers of AD.
“We can’t change age, family history and genetics. One of the things researchers can do is to find clues as soon as possible and provide a little help for the early identification of AD.” Liu Yong pointed out at the end of the report, “If we do this A little bit, we may be able to bring a little bit of benefit to more patients and families.”
Zhang Daoqiang, professor of Nanjing University of Aeronautics and Astronautics
Professor Zhang Daoqiang is a professor at Nanjing University of Aeronautics and Astronautics. He shared the title of “Research Progress in Brain Imaging Intelligent Computing and Its Several Applications”.
Zhang Daoqiang introduced that the best intervention stage for Alzheimer’s disease is the incubation period before symptoms appear and the stage of mild cognitive impairment. Once the patient enters the stage of dementia, effective treatment will no longer be possible. Therefore, early detection and early intervention becomes particularly important.
Zhang Daoqiang’s research is based on brain imaging to construct a brain network, and to mine, analyze and classify the brain network, so as to realize the technology of Alzheimer’s disease diagnosis.
Among them, the representative “brain connectomics” refers to the discipline that uses multimodal neuroimaging technology and network analysis methods to describe the structure and functional connection patterns of the living human brain. Three with valid connection. In the work process, the brain network is first constructed using brain images, and then features are extracted from the brain network, and finally the extracted features are classified.
In the report, Zhang Daoqiang also shared his team’s research progress and achievements in applications such as brain network classification, imaging genetics, brain cognition and brain decoding.
Lei Baiying, professor at the School of Biomedical Engineering, Shenzhen University
Professor Lei Baiying is a professor at the School of Biomedical Engineering, Shenzhen University, and shared the topic “Intelligent Diagnosis for Clinical Application”.
Lei Baiying’s research on intelligent diagnosis mainly focuses on two common brain diseases, Alzheimer’s disease and Parkinson’s disease. Lei Baiying said that in order to improve the diagnosis accuracy of brain diseases, his team proposed to build a longitudinal analysis model with regularization of multiple relationships to improve the accuracy of intelligent diagnosis and assist doctors in clinical diagnosis.
In the research, according to the different characteristics of single-time-point data, multi-time-point data, and multi-template data, different core methods are used for research, and they are applied to Alzheimer’s disease, mild cognitive impairment and autism respectively. in clinical diagnosis.
In addition, Lei Baiying’s team also explored deep learning in the early diagnosis of Alzheimer’s disease. Using the second-order statistics of MRI, the high-order pooling scheme was incorporated into the classifier, combined with tensor training, high-order pooling and semi-supervision. The learned GAN network is used for diagnosis.
Li Xiaomeng, Assistant Professor, Hong Kong University of Science and Technology
Prof. Xiaomeng Li, Assistant Professor of the Department of Electronic and Computer Engineering, Hong Kong University of Science and Technology, shared the title of “Empowering Clinical Decision-making by AI-based Medical Image Analysis”.
Li Xiaomeng introduced the team’s research contents such as medical image classification, segmentation and detection, medical image reconstruction, and image prediction through efficient annotation, as well as generalization research using models, such as the protection of hospital data privacy through federated learning research.
In view of the situation that there are a large number of unlabeled data and labeled data in the training data set, Li Xiaomeng and her team proposed a semi-supervised learning method to segment medical images based on the self-integration model of rotation consistency.
In addition, since weakly supervised learning can obtain pixel-level segmentation results in image segmentation, it also has very important application scenarios in medical imaging, such as gland segmentation in case images. In the sharing, Li Xiaomeng introduced the team’s method of weakly supervised learning on natural images.
“We discovered how to use existing deep learning models on natural images to make them more useful in medical imaging,” said Li Xiaomeng.
Zhao Di, associate researcher at the Institute of Computing Technology, Chinese Academy of Sciences
Professor Zhao Di is an associate researcher of the Institute of Computing Technology, Chinese Academy of Sciences, and shared the topic “Neuromorphic Computing for Medical Imaging Analysis”.
Zhao Di introduced that artificial intelligence has gone through two ups and downs since the emergence of artificial intelligence for 60 years, and now it has entered the eve of the real outbreak.
As an important part of the development stage of deep learning, convolutional neural network (CNN) image recognition is currently one of the main methods of medical image analysis. into people’s sight.
The third-generation neural network spiking neural network (SNN) is also called in-brain computing or neuromorphic computing. Compared with CNN, the power consumption of SNN has been reduced by orders of magnitude. Zhao Di believes that SNN is a possible direction for the future development of artificial intelligence.
Therefore, the spiking convolutional neural network (SCN) that combines SNN and CNN has great potential for development. In the case that the classification and target detection and segmentation accuracy are relatively close, the energy consumption of SCN is much lower than that of CNN.
Zhao Di said that the development of in-brain computing will greatly promote research in the field of medical and health care.
Leifeng Network Leifeng Network
This article is reprinted from: https://www.leiphone.com/category/healthai/I8yMlUw4Ls2TvOBq.html
This site is for inclusion only, and the copyright belongs to the original author.