Abstract
Objectives: To develop and validate a radiological nomogram combining radiological and clinical characteristics for differentiating mycoplasma pneumonia and bacterial pneumonia with similar CT findings. Methods: A total of 100 cases of pneumonia patients receiving chest CT scan were retrospectively analyzed, including 60 patients with mycoplasma pneumonia and 40 patients with bacterial pneumonia. The patients were divided into the train set (n = 70) and the test set (n = 30). The features were extracted from chest CT images of each patient by AK analysis software, then univarite analysis, spearman correlation analysis, and least absolute shrinkage and selection operator (LASSO) were utilized for dimension reduction in training set. A radiomics model was built by multivariable logistic regression based on the selected features, and a radiomics-clinical multivariable logistic regression model was built by combining imaging radiomics and clinical risk factors (age and temperature). ROC, AUC, sensitivity, specificity, and accuracy were calculated to validate the two models. The nomogram of the radiomics-clinical was built and evaluated by calibration curve. The clinical benefit of the two models was measured by using decision curve. Results: A total of 396 texture features were extracted from each chest CT image, and 10 valuable features were screened out. In the radiomics model, the AUC, sensitivity, specificity, and accuracy for the train set is 0.877, 0.762, 0.821, 78.6%, and for the test set it is 0.810, 0.667, 0.750 and 70.0%, respectively. In the radiomics-clinical model, the AUC, sensitivity, specificity, and accuracy for the train set is 0.905, 0.976, 0.714, 87.1%, and for the test set is is 0.847, 0.889, 0.667 and 80.0%, respectively. Decision curve analysis shows that both the two models increase the clinical benefits of the patients, and the radiomics-clinical model gains higher clinical benefits, compared to the radiomics model. Conclusion: The radiomics-clinical nomogram had good performance in identifying mycoplasma pneumonia and bacterial pneumonias, which would be helpful in clinical decision-making.