The novel coronavirus pneumonia auxiliary diagnosis artificial intelligence model was successfully established, with a total accuracy of 83%.

Authors: in novel coronavirus pneumonia, the key to screening pneumonia cases quickly and accurately is to take appropriate isolation and treatment measures Laboratory nucleic acid detection is the gold standard of diagnosis, but because it is time-consuming and prone to false-positive, sometimes it needs to be combined with computed tomography (CT) as a diagnostic tool In February 17th, a novel coronavirus pneumonia artificial intelligence model was developed by Xu Bo and professor of Tumour Hospital Affiliated to Tianjin Medical University, and the national supercomputer Tianjin center in the medRxiv preprint platform, which was co developed by CT The total accuracy of novel coronavirus pneumonia and other viral pneumonia is 83%, and it is fast and efficient Https://doi.org/10.1101/2020.02.14.20023028 according to the imaging mode, there are many features that can identify virus pathogens, which are related to their specific pathogenesis The sign of covid-19 is the bilateral distribution of plaque shadow and opaque ground glass Based on this, the researchers collected 453 cases of covid-19 confirmed by pathogens and CT images previously diagnosed as typical viral pneumonia, modified the concept transfer learning model to establish the algorithm, and then carried out internal and external verification The total accuracy of internal verification was 82.9%, the specificity was 80.5%, and the sensitivity was 84%; the total accuracy of external test data set was 73.1%, the specificity was 67%, and the sensitivity was 74% These results demonstrate the principle of using artificial intelligence to extract radiology features for timely and accurate diagnosis of covid-19 Deep learning algorithm framework this is the first study to apply artificial intelligence technology to CT images to effectively screen covid-19 The time of each case is about 2 seconds, and it can be remotely operated through a shared public platform Although there are still some limitations, the researchers believe that further optimization and testing can improve the accuracy, specificity and sensitivity by linking the hierarchical characteristics of CT images with the characteristics of other factors (such as genetic, epidemiological and clinical information) The platform can be used to assist clinical diagnosis and contribute to covid-19 disease control Novel coronavirus pneumonia: [1] A deep learning algorithm using CT images to screen for for Xu Bo deep team, the National Supercomputing Center successfully established a new crown pneumonia auxiliary diagnostic artificial intelligence model, Jeni turtle.