Skip to main content

Table 2 Main properties and results of each artificial intelligence model

From: Artificial intelligence for differentiating COVID-19 from other viral pneumonias on CT: comparative analysis of different models based on quantitative and radiomic approaches

Model

Type of features

Number of radiological featuresa

Number of relevant radiological featuresa

Test set

Accuracy (mean ± standard deviation)

Test set

AUC (mean ± standard deviation)

Model1

RF1 (2L)

21

10

0.713 ± 0.004

0.768 ± 0.032

Model2

QM (2L and 4 GS)

102

26

0.724 ± 0.006

0.800 ± 0.026

Model3

RF1 + RF2 (2L)

141

24

0.776 ± 0.003

0.867 ± 0.008

Model4

RF1 + QM (2L and 4 GS) + RF2 (2L)

241

32

0.796 ± 0.005

0.870 ± 0.011

  1. Principal characteristics of each model developed. The type of features, the number of initial radiological features, and the final relevant radiological features after LASSO regression used for building each classifier are reported together with results of accuracy and AUC obtained in the test set. Mean and standard deviation values of results were calculated after a 4-fold cross-validation iterated ten times. aPatient age and sex were added as clinical metrics in all models. AUC Area under the receiving operating characteristic curve, 2L Two lungs, GS Geometrical subdivisions, QM Quantitative metrics, RF1 First-order radiomic features, RF2 Second-order radiomic features