A few days ago, the top academic journal Cell published a latest paper on the diagnosis of diseases in COVID-19. A research team led by scientists in China has developed an artificial intelligence tool, which can accurately diagnose COVID-19 based on CT images of the chest. This paper was jointly completed by Tsinghua University, Sun Yat-sen University and Macau University of Science and Technology. Correspondents are Zhang Kang, Wang Guangyu, Lin Tianxin, He Jianxing and Li Weimin. This tool has been made public to medical staff all over the world.
In China, the COVID-19 epidemic has been completely controlled. On a global scale, many countries and regions still face the threat of this infectious disease. In order to prevent the spread of the epidemic and treat the infected people in time, rapid and accurate diagnosis technology is the key: on the one hand, in the high incidence season of respiratory diseases, many people will have lung inflammation. Accurate diagnosis can exclude patients infected with other pathogens, which is convenient for them to receive separate treatment, thus reducing the risk of exposure to COVID-19; On the other hand, from the current data, the mortality rate of patients with acute respiratory failure is very high, and most of them had lung inflammation before. If we can find signs of COVID-19 at the first time, we can treat respiratory failure before it appears.
In the first-line treatment, CT image is an important diagnostic tool. Compared with molecular detection in standard laboratory, CT scanning is faster and can observe the pathological features of lung more intuitively. We know that "reading" CT image data is one of AI's strengths in the medical field. As early as 20 18, Professor Zhang Kang's team published an article in Cell magazine, introducing an AI tool that can distinguish bacterial pneumonia from viral pneumonia in children according to X-ray chest films. The next year, his team developed an artificial intelligence to diagnose children's medical records.
▲ The development process of this AI tool
In this work, scientists have developed an artificial intelligence diagnosis system in COVID-19, which is based on more than 530,000 CT images of 3,777 patients. Different from the traditional end-to-end deep learning model, this diagnosis system combines two different models. The first step is the "lung lesion" model based on semantic segmentation, and the second step is to establish an intelligent diagnosis model based on the generated lung lesion map and the whole CT of the patient as input. This work can avoid the defects of "black box" model in practical medical application, improve the interpretability and generalization performance of artificial intelligence diagnosis system, and also improve the accuracy of diagnosis. By studying CT image data from COVID-19 patients, common pneumonia patients and control group, AI learned to distinguish them.
The researchers reported that the accuracy of this artificial intelligence system in diagnosing COVID-19 reached 92.49% in the self-inspection. Using different data sets in different regions, this AI diagnostic system has withstood the test of the real world-using retrospective data and prospective research, the accuracy of this AI system can reach about 90%, and even using overseas CT data, this system has reached 84. 1 1%. This series of research results show that the AI diagnostic system developed by scientists can achieve good performance whether using China data or international data.
▲ This system can assist junior doctors and improve their film reading ability.
Compared with human doctors, researchers found that the performance of AI system far exceeds that of junior doctors and is similar to that of intermediate/senior radiologists. This result shows that AI system can help doctors with less experience to make diagnosis and quickly improve their reading ability to the level of senior doctors. In countries and regions where medical resources are tight and senior doctors are hard to find, the significance of this system is self-evident.
It is worth mentioning that this AI system also identifies some clinical features related to the prognosis of patients based on the quantitative characteristics and clinical data of CT, and finds that coronavirus pneumonia-19 disease will not only affect the function of respiratory system, but also affect many other organs. In addition, it is also expected to provide more accurate prediction model and survival curve analysis for clinical prognosis, which is convenient for medical staff to intervene and treat in time.
To sum up, with the support of a large number of clinical CT images, researchers have developed an AI system, which can accurately select COVID-19 patients from patients with different symptoms. This can not only be used for rapid clinical diagnosis and treatment, but also help to train doctors with less experience and help them grow up quickly.
Finally, scientists announced that in order to better help the world control the epidemic, they will provide this AI tool free of charge to assist medical personnel in different countries and regions. We salute these scientists and look forward to an early end to the COVID-19 epidemic around the world!
For more information about "building this AI with 500,000 CT images can quickly diagnose COVID-19", please continue to pay attention to the column of deep space science and technology information, and deep space Bian Xiao will continue to update more science and technology news for you.
This article Source: Deep Space Game Editor: Heart of the Anonymous King 2 Click on the demo.