RÉSUMÉ
ObjectivesBased on the changes of lung lesions in patients with COVID-19 at different stages, a nomogram model describing CT image features was established by radiomics method to explore its efficacy in predicting the progression of the disease. MethodsThis retrospective study enrolled 136 patients with COVID-19 pneumonia who received at least two CTs including three cohorts (training cohort and validation cohort 1 and 2). Patients in the training cohort were divided into three groups according to time between onset of fever symptoms and the first CT. The clinical manifestations and CT features of each group were analyzed and compared. A nomogram to predict disease progression was constructed according to the CT features of the patients, and its performance was evaluated. ResultsThe training cohort consisted of 41 patients.A nomogram was generated to predict disease progression based on three CT features: irregular strip shadow, air bronchial sign, and the proportion of lesions with irregular shape ≥50%. AUC(95%CI)=0.906(0.817,0.995).The C index of the training cohort was 0.906, and the C index of the internal verification was 0.892. AUC(95%CI)of the validation cohort 1 (34 cases) =0.889(0.793,0.984);AUC(95%CI)of the validation cohort 2 (61 cases)=0.876(0.706,1.000).The calibration curves show that the predicted values of the nomogram are in good agreement with the observed values. ConclusionThe nomogram model based on CT radiomics can predict the outcome of lung lesions in patients with high sensitivity and specificity.According to the changes of CT image characteristics of patients with COVID-19, lung lesions will be improved when the proportion of irregular cable shadow, air bronchogram and irregular lesions is greater than 50%.