Table 6.

The comparison results of our proposed approach and existing methods on the Covid-CT dataset. We report Accuracy, AUC, and F1 score for performance comparison.

Methods	Acc	AUC	F1
DensNet169 [22]	0.83	0.87	0.81
DenseNet169 (self-trans) [22]${}^{†}$	0.83	0.90	0.84
Redesigned Covid-Net [23]	0.79	0.85	0.79
M-Inception [9]	0.81	0.88	0.82
Evidential Covid-Net [14]	0.81	0.87	0.81
Ours (Res50+ImageNet+CRE (${\lambda }_1^{*}$=0.38))	0.84	0.90	0.86

$†$ For a fair comparison, we retrained the DenseNet169 (self-trans) model with same settings, except for the pre-training on the additional LUNA dataset.