Table 3.

Quantitative evaluation metrics for four cross-validation cases on the DXR4 dataset (Kermany et al., 2018; Cohen et al., 2020; Jaeger et al., 2014), and the resulting average and standard deviation. Superscript max/min indicates the highest/lowest score among the cross validation sets.

Classification performance of ResNet-50 in DXR4 dataset: COVID-19, pneumonia, tuberculosis and healthy
Metric (%)	Cross validation #1	Cross validation #2	Cross validation #3	Cross validation #4	Average
Recall	62.7${}^{\max }$	61.9	62.4	61.5${}^{\min }$	62.0 ± 0.5
Precision	81.0${}^{\max }$	78.9	77.1${}^{\min }$	77.6	78.6 ± 1.7
AUC-ROC	94.1${}^{\max }$	93.2	92.7	92.6${}^{\min }$	93.2 ± 0.7
AUC-ROC macro	89.9	89.0${}^{\min }$	91.0${}^{\max }$	89.5	89.9 ± 0.9
AUC-ROC micro	88.1${}^{\min }$	88.8	90.4${}^{\max }$	88.8	89.0 ± 1.0
AUC-ROC weighted	87.3${}^{\min }$	88.4	89.6${}^{\max }$	88.2	88.4 ± 0.9
F1	70.7${}^{\max }$	69.9	69.0	68.6${}^{\min }$	69.5 ± 0.9
F1 macro	70.7${}^{\max }$	70.0	69.1	68.6${}^{\min }$	69.6 ± 0.9
F1 micro	70.5${}^{\max }$	69.8	68.8	68.4${}^{\min }$	69.4 ± 0.9
F1 weighted	70.7${}^{\max }$	70.1	69.0	68.7${}^{\min }$	69.6 ± 0.9

Classification performance of DenseNet-121 in DXR4 dataset: COVID-19, pneumonia, tuberculosis and healthy
Metric (%)	Cross validation #1	Cross validation #2	Cross validation #3	Cross validation #4	Average
Recall	63.3${}^{\max }$	62.4	62.3${}^{\min }$	62.9	62.7 ± 0.5
Precision	80.1	80.8${}^{\max }$	79.3	76.8${}^{\min }$	79.3 ± 1.7
AUC-ROC	93.8${}^{\max }$	91.2	89.0${}^{\min }$	89.8	91.0 ± 2.1
AUC-ROC macro	90.1${}^{\min }$	92.5${}^{\max }$	90.2	91.5	91.1 ± 1.1
AUC-ROC micro	89.2	91.2${}^{\max }$	88.8${}^{\min }$	89.2	89.6 ± 1.1
AUC-ROC weighted	88.7	90.6${}^{\max }$	87.6	87.2${}^{\min }$	88.5 ± 1.5
F1	70.7${}^{\max }$	70.4	69.8	69.2${}^{\min }$	70.0 ± 0.6
F1 macro	70.0${}^{\max }$	69.6	69.0	68.3${}^{\min }$	69.3 ± 0.7
F1 micro	70.7${}^{\max }$	70.4	69.8	69.2${}^{\min }$	70.0 ± 0.6
F1 weighted	70.4${}^{\max }$	69.9	69.6	69.0${}^{\min }$	69.8 ± 0.6

Classification performance of DenResCov-19 in DXR4 dataset: COVID-19, pneumonia, tuberculosis and healthy
Metric (%)	Cross validation #1	Cross validation #2	Cross validation #3	Cross validation #4	Average
Recall	70.0	71.0${}^{\max }$	67.0${}^{\min }$	70.7	69.7 ± 1.8
Precision	80.0${}^{\min }$	83.0	86.0${}^{\max }$	82.6	82.9 ± 2.4
AUC-ROC	93.9${}^{\min }$	95.0	96.0${}^{\max }$	95.2	95.0 ± 0.8
AUC-ROC macro	94.7${}^{\min }$	94.7	94.7${}^{\max }$	94.8	95.6 ± 0.1
AUC-ROC micro	93.9${}^{\min }$	94.4	98.2${}^{\max }$	93.9	95.1 ± 2.1
AUC-ROC weighted	93.3${}^{\min }$	94.1	98.0${}^{\max }$	93.6	94.7 ± 1.8
F1	75.0${}^{\min }$	76.5${}^{\max }$	75.3	76.2	75.8 ± 0.7
F1 macro	76.2	77.6${}^{\max }$	76.1${}^{\min }$	77.1	76.7 ± 0.7
F1 micro	74.9${}^{\min }$	76.3${}^{\max }$	75.3	76.2	75.6 ± 0.7
F1 weighted	75.0${}^{\min }$	76.5${}^{\max }$	75.3	76.2	75.7 ± 0.7