Table 2.

NPC Diagnosis performance (Average, standard deviation and 95% CI) of various models

Model	Precision	Sensitivity	Specificity	F1-score
SwinT	0.9949 (±0.0031) (0.9888–1.0000)	0.9984 (±0.0023) (0.9939–1.0000)*	0.9991 (±0.0006) (0.9979–1.0000)	0.9966 (±0.0015) (0.9937–0.9995)
MaxViT	0.9910 (±0.0145) (0.9626–1.0000)	0.9938 (±0.0072) (0.9797–1.0000)	0.9983 (±0.0027) (0.9930–1.0000)	0.9923 (±0.0061) (0.9803–1.0000)
PoolF	0.9959 (±0.0034) (0.9892–1.0000)*	0.9979 (±0.0022) (0.9936–1.0000)	0.9992 (±0.0006) (0.9980–1.0000)*	0.9969 (±0.0012) (0.9945–0.9993)*
CaiT	0.9949 (±0.0047) (0.9857–1.0000)	0.9954 (±0.0033) (0.9889–1.0000)	0.9991 (±0.0009) (0.9973–1.0000)	0.9951 (±0.0011) (0.9929–0.9973)
ResNet	0.9883 (±0.0066) (0.9754–1.0000)	0.9964 (±0.0029) (0.9907–1.0000)	0.9978 (±0.0012) (0.9954–1.0000)	0.9923 (±0.0024) (0.9876–0.9970)
DenseNet	0.9918 (±0.0033) (0.9853–0.9983)	0.9964 (±0.0043) (0.9880–1.0000)	0.9985 (±0.0006) (0.9973–0.9997)	0.9941 (±0.0027) (0.9888–0.9994)
Xception	0.9949 (±0.0018) (0.9914–0.9984)	0.9969 (±0.0034) (0.9902–1.0000)	0.9991 (±0.0003) (0.9985–0.9997)	0.9958 (±0.0021) (0.9917–0.9999)
ConvNeXt	0.9933 (±0.0061) (0.9813–1.0000)	0.9948 (±0.0075) (0.9801–1.0000)	0.9988 (±0.0011) (0.9966–1.0000)	0.9940 (±0.0037) (0.9867–1.0000)

‘*’ represents the best result of the specific metric.