Research in context
Evidence before this study: Fast and accurate screening large numbers of cases with viral pneumonia symptoms for appropriate quarantine and treatment measures is a priority to control the spread of Corona Virus Disease (COVID-19). Pathogenic laboratory testing is the diagnostic gold standard but it is time-consuming with significant false positive results. Computed tomography （CT） is also a major diagnostic tool for COVID-19. Although typical CT images may help early screening of suspected cases, the images of various viral pneumonias are similar and they overlap with other infectious and inflammatory lung diseases. Artificial intelligent technologies such as deep learning might be able to extract COVID-19’s graphical features and provide a clinical diagnosis. However, there is no published work exploring this possibility.
Added value of this study: Our study represents the first study to apply artificial intelligence technologies to CT images for effectively screening for COVID-19. We employed a modified Inception migration-learning model to establish the algorithm. The internal validation achieved a total accuracy of 82·9% with specificity of 80·5% and sensitivity of 84%. The external testing dataset showed a total accuracy of 73·1% with specificity of 67% and sensitivity of 74%. The time for each case is about 2 seconds and can be done remotely via a shared public platform.
Implications of all the available evidence: To control the outbreak of COVID-19,it is crucial to develop fast, accurate, safe and non-invasive methods for early diagnosis. We present a proof-of-principle of such as method. We believe that with more data included for further optimization and testing, the accuracy, specificity and sensitivity can be improved. Such a platform can be used to assist clinical diagnosis.
Abstract
Background: To control the spread of Corona Virus Disease (COVID-19), screening large numbers of suspected cases for appropriate quarantine and treatment measures is a priority. Pathogenic laboratory testing is the diagnostic gold standard but it is time consuming with significant false positive results. Fast and accurate diagnostic methods are urgently needed to combat the disease. Based on COVID-19 radiographical changes in CT images, we hypothesized that deep learning methods might be able to extract COVID-19’s graphical features and provide a clinical diagnosis ahead of the pathogenic test, thus saving critical time for disease control.
Methods：We collected 453 CT images of pathogen-confirmed COVID-19 cases along with previously diagnosed with typical viral pneumonia. We modified the Inception migration-learning model to establish the algorithm, followed by internal and external validation.
Findings: The internal validation achieved a total accuracy of 82·9% with specificity of 80·5% and sensitivity of 84%. The external testing dataset showed a total accuracy of 73·1% with specificity of 67% and sensitivity of 74%.
Interpretation: These results demonstrate the proof-of-principle for using artificial intelligence to extract radiological features for timely and accurate COVID-19 diagnosis.
Funding: No funding is involved in the execution of the project.