Table 6.

Performance comparison of eight models on 5-fold-validation set

Models	ACC	AUC	$F_1$	Rec	Prec
CovidVGG${}^{\mathrm{a}}$	0.771 ± 0.127	0.693 ± 0.237	0.524 ± 0.315	0.562 ± 0.281	0.500 ± 0.340
VGG19	0.784 ± 0.133	0.688 ± 0.237	0.534 ± 0.324	0.569 ± 0.288	0.512 ± 0.351
ResNet50	0.956 ± 0.007	0.992 ± 0.002	0.943 ± 0.008	0.935 ± 0.010	0.952 ± 0.007
AlexNet	0.937 ± 0.004	0.984 ± 0.003	0.917 ± 0.007	0.915 ± 0.008	0.921 ± 0.016
Single-CNN${}^{\mathrm{a}}$	0.937 ± 0.026	0.987 ± 0.003	0.927 ± 0.024	0.928 ± 0.019	0.931 ± 0.023
Double-CNN${}^{\mathrm{a}}$	0.938 ± 0.036	0.985 ± 0.008	0.921 ± 0.047	0.931 ± 0.020	0.916 ± 0.066
TIN	0.972 ± 0.008	0.997 ± 0.001	0.959 ± 0.010	0.966 ± 0.004	0.954 ± 0.020
DNN	0.963 ± 0.007	0.991 ± 0.002	0.952 ± 0.007	0.951 ± 0.003	0.954 ± 0.012
OriginalViT	0.964 ± 0.003	0.995 ± 0.001	0.953 ± 0.003	0.957 ± 0.009	0.950 ± 0.009
CovidViT${}^{\mathrm{a}}$	0.982 ± 0.001	0.996 ± 0.002	0.976 ± 0.001	0.973 ± 0.004	0.978 ± 0.005

The best results are in bold

${}^{\mathrm{a}}$Represents our proposed models