Room: Track 2
Purpose: Chest CT imaging has been added as an auxiliary method to detect the novel coronavirus disease 2019 (COVID-19) with high sensitivity to identify suspected patients prior to the occurrence of positive nuclear acid detection results. However, because of the significant overlap in CT manifestations between COVID-19 and other coronavirus and influenza viral pneumonias, the specificity of using CT to detect the COVID-19 is extremely low. In this study, we developed a machine learning-enabled algorithm combining lung CT radiomics and clinical factors to significantly improve the early detection of patients with COVID-19.
Methods: A total of 109 patients with COVID-19 were included in this study. The patients with influenza viral pneumonia were used as control. The 1766 radiomics features were extracted from disease lesion regions on CT images. The Mann-Whitney U test was applied to assess the differences of CT radiomics features between COVID-19 and non-COVID-19 groups. The LASSO regression were employed to select optimal features. A predictive model was constructed to combine clinical factors and optimal radiomics features using a 5-fold cross validation method and the random forest machine. The utility of the model constructed was evaluated by the ROC analysis in the validation cohort.
Results: A total of 17 radiomics features show significant differences between COVID-19 and non-COVID-19 groups. The predictive model using the optimal radiomics features alone shows an excellent performances in early identification of COVID-19 with an AUC value of 0.867. Furthermore, the model combining radiomics and clinical factors showed an improved detection rate with an AUC value of 0.937 in the validation cohort.
Conclusion: The random forest machine-enabled diagnostic algorithm depicting the correlation of CT radiomics and clinical factors with COVID-19 could be used to accurately identify patients with COVID-19, which would have important diagnostic and therapeutic implications from a precision medicine perspective.