Table 5.

Evaluation of DR classification algorithms on fundus photograph dataset.

Classification algorithm	Transfer learning	Attention mechanism	Acc	Recall	Specificity	F1	AUC
VGG19 (Wu and Hu, 2019)	-	-	0.51	N/A	N/A	N/A	N/A
ResNet50 (Wu and Hu, 2019)	-	-	0.49	N/A	N/A	N/A	N/A
Inception V3 (Wu and Hu, 2019)	-	-	0.61	N/A	N/A	N/A	N/A
DR-IIXRN (Ai et al., 2021)	-	-	0.793	0.7933	0.8778	0.7602	0.7602
Xception	NO	NO	0.7479	0.7479	0.8307	0.6808	0.7307
	YES	NO	0.7901	0.7901	0.8767	0.7539	0.8093
	YES	YES	0.7939	0.7939	0.8711	0.7487	0.7942
EfficientNetV2B3	NO	NO	0.7358	0.7358	0.8271	0.6628	0.7127
	YES	NO	0.797	0.797	0.8776	0.7555	0.8025
	YES	YES	0.804	0.804	0.8831	0.7653	0.8109
GABNet	-	NO	0.6877	0.6876	0.807	0.6242	0.5987
	-	YES	0.7607	0.7607	0.8398	0.6954	0.743

Value in bold means the best of the same class.

“N/A” Means that the metric was not displayed in the comparison article.

“-” Means that the algorithm does not have this feature or property.