Artificial intelligence "trains" out of the imaging department "Da Na"
Just after the Spring Festival holiday, a message made the artificial intelligence (AI) circle exciting. In the latest issue of "Cell" magazine on February 22nd, the research of Zhang Kang team of Guangzhou Women and Children Medical Center was published on the cover of the magazine. The result is an accurate diagnosis of eye diseases and pneumonia. AI system. Let's take a look at the related content with the network communication Xiaobian. For AI based on data-based ingredients, there is nothing better to "chew" than medical imaging data. At the press conference held on February 23, the R&D team introduced that the new generation AI platform is “powerfulâ€, capable of reading both X-ray and ultrasound data, as well as CT (X-ray tomography) and MR (Magnetic resonance). Image; can diagnose macular degeneration and diabetic retinal macular edema in 30 seconds, and can also analyze and judge the pathogen type of children's pneumonia within a few seconds, the accuracy and sensitivity of diagnosis are achieved More than 90. It is understood that this is not only the first time that the Chinese research team has published research results on medical artificial intelligence in the top biomedical journals, but also the world's first use of such large and well-marked high-quality data for migration learning and highly accurate diagnosis. To achieve a breakthrough in the use of AI to accurately recommend treatment. "In the future, we will continue to increase the retinal diseases that this system can diagnose, and we will also add diseases including other systems such as tumors." Director of the Human Genome Medicine Institute of the University of California, San Diego, and Director of the Genetic Testing Center of the Guangzhou Women and Children Medical Center. Zhang Kang said. In ophthalmic treatment, retinal OCT (optical coherence tomography) imaging technology is one of the most commonly used diagnostic techniques. By obtaining high-resolution images of retinal tissue, doctors can accurately target age-related macular degeneration and diabetic macular edema. Diagnosis of blind eye diseases and treatment options. Based on the universality of OCT technology, if AI technology can be used to process these images, it will undoubtedly greatly enhance the efficiency and accuracy of the diagnosis. To this end, the Zhang Kang team acquired more than 200,000 images of OCT and trained a deep learning algorithm using 100,000 images from nearly 5,000 patients. After a lot of iterative training, the accuracy of this algorithm has reached the current optimal value. “After learning the OCT image data of more than 200,000 cases, the accuracy of the AI ​​platform for the diagnosis of macular degeneration and macular edema was 96.6%, the sensitivity was 97.8%, and the specificity was 97.4%.†According to Zhang Kang, the new generation AI The platform can not only realize the identification and severity assessment of common retinal diseases based on OCT data, but also realize the differential analysis and rapid and accurate determination of the pathogen type of children's pneumonia based on chest X-ray image data. Then, why does the diagnostic level of the AI ​​platform after “learning training†increase rapidly? This is the innovation of the algorithm applied in this research—migration learning. The so-called "migration learning" is to transfer the trained model parameters to the new model to help the new model training, that is, use the existing knowledge to learn new knowledge and find the similarity between existing knowledge and new knowledge. . This is actually equivalent to giving the opposite. "For example, you have never seen a tiger in the past, but when you meet three tigers, you will know the fourth one." Medical imaging artificial intelligence expert, Huiying Huiying CEO Chai Xiangfei explained to the reporter of Journal of Chinese Academy of Sciences "When we build a basic understanding of a thing, it is relatively easy to learn new things, and with a small sample we can have a knowledge transfer. This is migration learning." Compared with most other learning models, "from scratch", migration learning uses convolutional neural network (CNN) to learn the characteristics of learning target task input data based on existing trained source task parameters, and obtain new ones. Network model and its parameters. Taking medical image learning as an example, the system will recognize the characteristics of the image in the target system, construct the new system model and parameters from the similarity of the input image data from the structure and parameters of the source system imported by the researchers. Xue Yu, a professor at the School of Life Science and Technology at Huazhong University of Science and Technology, said that the traditional machine learning algorithm training data set is large and feature extraction is difficult. The result is that the data set is not accurate, and the prediction accuracy is improved after the data is larger. If the set is big, it will not be allowed. The advantage of deep learning is that the larger the data set, the higher the accuracy, and the feature extraction ability is much stronger than the traditional machine learning algorithm. "CNN is a kind of method in deep learning algorithm. It has advantages in processing image data. This research strategy is to let the machine learn the characteristics of 1000 kinds of pictures and then build models, and then carry out migration learning for the problems that need to be studied, in this case, The training set is large enough to be accurate and high.†Xue Yu commented, “The theoretical training set is constantly increasing, and the accuracy can completely exceed the diagnosis of any top expert.†For AI based on data-based ingredients, nothing needs to be "chewed" more than medical imaging data. In medical treatment, more than 80% of the data comes from CT, X-ray, MR, ultrasound and other medical images. AI can use this massive data to generate algorithm models to ensure the maximum inclusiveness of the model. However, in Chai Xiangfei's view, there is a notable feature in the medical field that medical data has no way to have a rich source like image data such as faces, fingerprints, and license plates. "In fact, the data of medical imaging is very limited, especially for single diseases. Each of us can't take a film every year, such as interstitial pneumonia or a fracture in a certain part. There may be only tens of thousands in the country every year. Patients, and also scattered in various regions and hospitals, data access is very difficult." Chai Xiangfei said. Just as radiologists need to read a large number of clinical medical images, "feeding" pathological image data is also the most important way to learn AI systems. The more adequate the pathological image data of “feedingâ€, the stronger the analytical ability of AI. "You can get very good data to know what problems exist in the algorithm, and achieve the best results through repeated calculations by AI." Zhang Kang also pointed out that AI applications in the medical field, data acquisition is a big challenge. . “Chinese hospitals have a large number of patients, but if they are not purified and are not labeled with high quality, such data will not be expected to be directly input into the computer.†In addition, although most radiology departments have completed millions of imaging examinations, and the degree of structuring is high, most of them do not have doctor's annotation information. The professionalism of medical imaging determines its particularity. Most of the image data can only be relied on professional and experienced practitioners in the medical field. It is difficult to outsource the labeling tasks like voice data, text data or natural images. Go out. Not only that, Zhang Kang also pointed out that the AI ​​medical field has always been monopolized by several large IT companies. If the blockade of data and technology is formed, it will also limit the development and application of AI in the medical industry. At present, video has become the main breakthrough for AI in the medical field. However, Chai Xiangfei believes that this is not easy to break through. The combination of AI and medical scenes still has a long way to go. AI developers and engineers are on the medical industry. The strangeness is the biggest challenge. AI medical imaging is different from other fields that only need theoretical talents or applied talents. It requires a large number of combined talents of medical workers. Chai Xiangfei, who has had many years of research experience in the United States, deeply feels that there are still major differences in the cultivation of talents in this cross-disciplinary field at home and abroad. "In the United States, engineering students have seven or eight years of hospital work experience, engaged in joint development, and then handed the results to the device manufacturers for commercialization. However, there are very few people with relevant experience in the country, and a large number of doctors are interested and willing to Often the engineering background is insufficient, and some doctors are eager to do industrialization, but the business experience and ability are relatively insufficient." Chai Xiangfei said. In order to cultivate more compound talents, Huiyi Huiying launched the “Excellent Talents Program†to deliver excellent medical and computer talents from China to Stanford University and other top universities in the world for further study and improve the comprehensive talents in the medical field in China. The competitiveness of the global market. Xia Huimin, director of the Guangzhou Women and Children Medical Center, said that the contradiction between the growing quality of medical resources and the lack of training of professional medical personnel is one of the pain points facing the hospital. Studying better technical means and platforms can not only solve the problem of insufficient medical service capacity to a certain extent, but also improve the fairness and accessibility of health services. For this AI system developed by the research team, Zhang Kang hopes that the future can be applied to areas including primary care, community health care, family doctors, emergency rooms, etc., to form a large-scale automated triage system.
BIOTEPT motor Cost effective and food safe. Stainless steel IEC motors from BIOTEPT excel in power, efficiency and food safety. With the IEC stainless steel motors you benefit from a cost-effective and food-safe drive that is suitable for applications where hygiene plays an important role.
DAYTON, LEESON, MARATHON MOTORS and U.S. MOTORS
IEC Serie – IEC Stainless Steel Motors: main features
Iec Stainless Steel Motor,Waterproof Electric Motor,Waterproof Motor,Water Resistant Motor Ningbo Biote Mechanical Electrical Co.,Ltd , https://www.biotept.com
