Fig 1.
Flowchart of the study.
Table 1.
Distribution of lesions according to patient groups, lesion types and lobes.
Table 2.
ROC analysis results of size, shape and first order texture parameters to discriminate COVID-19 related lesions from atypical pneumonia.
Table 3.
Level of statistical significance (p values) of the first order texture parameters between the pneumonic lesion types of whole study population and between the same type lesions of COVID-19 and atypical pneumonia groups.
Table 4.
ROC analysis results of second order texture parameters in discrimination of COVID-19 from atypical pneumonia.
Table 5.
Level of statistical significance (p values) of the second order texture parameters between the same type lesions of COVID-19 and atypical pneumonia groups.
Fig 2.
Frequency distribution plot (histogram) of COVID-19-related lesions.
(a) ground glass opacity, (b) crazy paving, and (c) consolidation. Note that the histogram of the consolidation is right-skewed and (d) mean skewness value is negative.
Fig 3.
(a) Gray scale CT image and (b) Skewness map of a 2.6 mL GGO lesion showed that high density septal components lead a right-shift.
Table 6.
Model comparison in COVID-19 and atypical pneumonia prediction.
Fig 4.
Decision curve analysis of the models with highest accuracy.
(a) PSG Model-3, and (b) NSG Model-7 None (thick line): Net benefit if all patients were accepted as atypical pneumonia, All (thin line): Net benefit if all patients were accepted as COVID-19, Model (dashed line): Net benefit if patients were managed according to model. Note that PSG model was not superior to manage all patients as COVID-19 in lower threshold probabilities.
Fig 5.
An algorithm to predict pneumonic lesions with dedicated logistic regression models.
AP Atypical pneumonia, PSG Positive skewness group, NSG Negative skewness group, Number of, ⊖ Falsely predicted patients, and ⊕ Correctly predicted patients. Note the distribution of 20 patients with both types of lesions that black circled in PSG and white circled in NSG model predictions.