. 2021 May 6;118:108005. doi: 10.1016/j.patcog.2021.108005

Table 3.

The quantitative results on our 3D dataset, which are the mean values of 3-fold cross validation. We report Sensitivity (Sen.), Specificity (Spec.) and AUC for each class, as well as Accuracy, AUC, and F1 score for overall performance (unit:%). The values in brackets are 95% confidence intervals.

Methods	COVID-19			H1N1			CAP			Healthy			Acc	AUC	F1
	Sen.	Spec.	AUC	Sen.	Spec.	AUC	Sen.	Spec.	AUC	Sen.	Spec.	AUC
COVNet [4]	87.00	79.89	92.18	91.08	97.88	98.63	65.57	88.51	86.59	33.94	90.76	79.25	68.16 [65.42, 70.79]	89.39	66.66
DeCovNet [3]	85.73	92.03	95.14	98.94	99.67	99.89	67.22	96.32	91.43	82.04	91.11	94.53	84.89 [82.77, 86.89]	95.44	83.94
Ours (Res18+PSP)	87.37	96.10	97.87	98.95	99.18	99.47	65.63	97.78	93.84	91.61	90.38	96.88	88.14 [86.14, 89.89]	97.20	86.71
Ours (Res18+CRE)	90.73	96.45	98.20	98.95	99.34	99.84	66.40	98.82	94.16	95.60	92.02	96.75	90.51 [88.76, 92.13]	97.42	88.95
Ours (Res18+PSP+CRE) $†$	90.75	97.51	98.63	98.75	99.34	99.94	72.15	98.08	94.60	92.79	91.84	97.18	90.51 [88.76, 92.26]	97.78	89.42
Res18 ( $m$ =2, $d$ =64)	90.32	91.82	96.73	98.43	98.69	99.66	59.00	99.11	95.67	88.79	92.01	95.91	87.01 [85.14, 88.89]	97.14	85.32
Res18 ( $m$ =1, $d$ =64)	87.38	96.27	97.65	97.89	98.20	98.95	52.50	98.67	91.87	98.00	89.84	96.71	87.89 [86.02, 89.76]	96.54	84.90
Res18 ( $m$ =1, $d$ =128)	87.37	93.95	96.92	100.00	99.67	100.00	64.70	97.05	92.54	88.00	91.46	96.15	87.13 [85.14, 89.14]	96.58	85.67
Res18+CRE ( $λ_{1} = 0.05$ )	90.32	93.96	98.00	98.95	98.36	99.79	62.38	98.08	95.16	89.98	92.92	97.12	88.01 [86.02, 89.89]	97.70	86.12
Res18+CRE ( $λ_{1} = 0.50$ )	89.47	95.38	98.20	99.48	98.85	99.97	63.09	98.38	94.82	90.79	90.93	96.72	88.26 [86.39, 90.14]	97.59	86.64
Res18+CRE ( $λ_{1} = 0.75$ )	90.32	94.49	97.90	99.48	99.01	99.86	63.05	98.53	94.75	90.41	91.65	96.67	88.39 [86.52, 90.14]	97.47	86.76
Res10+CRE	91.58	93.96	98.28	99.47	98.85	99.94	65.61	98.53	95.77	89.21	92.92	96.98	88.76 [86.89, 90.51]	97.65	87.36
Res34+CRE	91.16	90.41	96.36	99.47	98.69	99.80	58.23	97.79	92.41	82.41	92.73	96.13	85.39 [83.15, 87.39]	96.40	83.62

$†$ Ours (Res18+PSP+CRE) is with $m$ =2, $d$ =64, the learned adaptive weight $λ_{1}^{*} = 0.28$ , and ResNet18 backbone.