A few days ago, the top academic journal "Cell" published online a new paper on the new crown disease diagnosis. A team of Chinese scientists developed an artificial intelligence tool that can accurately make a diagnosis of neocoronaryngitis based on a CT image of the chest. The paper is a joint collaboration between Tsinghua University, Sun Yat-sen University, and Macau University of Science and Technology, with corresponding authors Zhang Kang, Wang Guangyu, Lin Tianxin, He Jianxing, and Li Weimin, and this tool is now available to medical professionals around the world.
In China, the Xinguang epidemic has been adequately controlled. Globally, however, many countries and regions continue to face the threat of this infectious disease. To prevent the outbreak from expanding and to treat infected people in a timely manner, rapid and accurate diagnostic techniques are key: on the one hand, at a time of high respiratory illness, many people develop inflammation of the lungs. Accurate diagnosis reduces the risk of exposure to neocoronaviruses by eliminating patients infected with other pathogens and allowing them to receive individualized treatment; on the other hand, current data show a high mortality rate among patients with acute respiratory failure, most of whom have had prior lung inflammation. If we can find the first signs of C. neoformans pneumonia, we can treat them before they develop respiratory failure.
In the first line of care, CT imaging is an important diagnostic tool. Compared to molecular tests performed in a standard laboratory, CT scans are faster and provide a more visual view of the pathology of the lungs. As we know, "reading" CT image data is one of the strong points of AI applications in the medical field. As early as 2018, Prof. Zhang Kang's team published an article in the journal Cell, introducing an AI tool capable of distinguishing between bacterial and viral pneumonia in children based on X-ray chest radiographs. The following year his team developed an AI for diagnosing children's medical records.
▲Development process of this AI tool
In this work, scientists developed an AI diagnostic system for neocoronary pneumonia based on a total of more than 530,000 CT images from 3,777 patients. Unlike traditional end-to-end deep learning models, this diagnostic system incorporates two different modeling steps: the first step is a "lung lesion" model based on semantic segmentation, and the second step is to further build an intelligent diagnostic model based on the generated lung-lesion mapping, which takes the patient's entire CT as input. This work can avoid the shortcomings of "black-box" models in practical medical applications, improve the interpretability and generalization performance of AI diagnostic systems, and also improve the accuracy of diagnosis. The AI learned to differentiate between patients with new coronary pneumonia, patients with common pneumonia, and controls by examining CT image data from these patients.
The researchers reported that in self-testing, this AI system diagnosed neocoronal pneumonia with 92.49% accuracy. Using different datasets from different regions, this AI diagnostic system stood up to real-world testing - using both retrospective data and prospective studies, the AI system was able to achieve around 90 percent accuracy, and even using CT data from overseas, the system achieved 84.11 percent accuracy. The results of this series of studies show that this AI diagnostic system developed by the scientists can achieve better performance whether using Chinese data or international data.
▲This system can assist less senior doctors and improve their ability to read films
Compared with human doctors, the researchers found that the AI system outperformed the less senior doctors by a wide margin, while performing about the same as mid-level/senior radiologists. This result suggests that AI systems can assist less experienced physicians in making diagnoses, quickly improving their ability to read films to a level close to that of senior physicians. In countries and regions where medical resources are tight and senior doctors are hard to come by, the importance of this system cannot be overstated.
It is worth mentioning that this AI system also identifies some clinical features related to patient prognosis based on quantitative features of CT and clinical information, and finds that COVID-19 disease does not only affect the function of the respiratory system, but also affects several other organs. In addition, it is expected to provide a more accurate prediction model and survival curve analysis for clinical prognosis, which facilitates timely intervention and treatment by medical personnel.
In summary, with the support of a large number of clinical CT images, the researchers have developed an AI system that can accurately select patients who develop neocoronary pneumonia from those who present with different symptoms. This can be used not only for rapid clinical diagnosis and treatment, but also to assist in the training of less-qualified physicians and help them grow rapidly.
Finally, the scientists announced that in order to better help control the outbreak globally, they will make this AI tool publicly available for free to assist medical professionals from different countries and regions. We salute these scientists and look forward to an early global end to the New Crown epidemic!
To learn more about "500,000 CT images to create this AI can quickly diagnose the new crown pneumonia," more information, please continue to pay attention to the deep space of the science and technology information column, the deep space editorial staff will continue to update you with more science and technology news.
Source: deep space game Editor's responsibility: anonymous Heart of Kings 2 click to try