Table 3.

Performance for four-class classification.

Model	VGG	ResNet-macro F1	ResNet-ACC	DenseNet
Macro-average precision	0.360 ± 0.143	0.444 ± 0.071	0.340 ± 0.033	0.393 ± 0.110
Macro-average recall	0.342 ± 0.068	0.445 ± 0.054	0.328 ± 0.041	0.379 ± 0.067
Macro-average F1 score	0.304 ± 0.080	0.429 ± 0.062	0.364 ± 0.029	0.350 ± 0.095
Accuracy	0.691 ± 0.070	0.667 ± 0.078	0.712 ± 0.056	0.699 ± 0.064

Bold fonts represent the best performance among the methods.

ResNet-macro F1 ResNet early stopped with macro-average F1-score, ResNet-ACC ResNet early stopped with accuracy.