Table 3.

Evaluation of different algorithms on retinal OCT classification.

Algorithm	Acc	Recall	Specificity	F1	AUC	Parameters (M)	Test time
ResNet50a(He et al., 2016)	0.941	0.941	0.9803	0.9411	0.9923	23.595	4.81 s±315 ms
InceptionV3a(Szegedy et al., 2016)	0.96	0.96	0.9867	0.9599	0.9947	21.811	4.78 s±277 ms
Xceptiona(Chollet, 2017)	0.95	0.95	0.9833	0.9501	0.9961	20.87	4.81 s±285 ms
EfficientNetV2B3a(Tan and Le, 2021)	0.928	0.928	0.976	0.9273	0.9929	12.937	5.39 s±204 ms
Huanga(Huang et al., 2019)	0.884	0.846	N/A	N/A	N/A	N/A	N/A
GABNeta	0.965	0.965	0.9883	0.965	0.9969	9.361	7.26 s±353 ms
FN-F1-OCTb(Ai et al., 2022)	0.985	0.985	0.995	0.985	0.99	99.717	18.1 s±831 ms
FN-Weight-OCTb(Ai et al., 2022)	0.984	0.984	0.995	0.984	0.99	99.717	15.6 s±419 ms
FN-Auto-OCTb(Ai et al., 2022)	0.987	0.987	0.996	0.987	0.991	99.774	15.8 s±451 ms
Kermanyb(Kermany et al., 2018)	0.961	0.961	0.987	0.961	0.99	N/A	N/A
Hwangb(Hwang et al., 2019)	0.9693	N/A	N/A	N/A	N/A	N/A	N/A
Sinhab(Sinha et al., 2023)	0.944	0.944	0.9815	0.9448	N/A	N/A	N/A
EfficientNetV2B3+GABb	0.978	0.978	0.9927	0.9781	0.9983	18.281	5.53 s± 94.5 ms
Xception+GABb	0.99	0.99	0.9967	0.99	0.9994	30.354	4.88 s±240 ms

^aNon-transfer learning methods.

^bTransfer learning methods.

Value in bold means the best of the same class.

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