An Effective Method for Detecting and Classifying Diabetic Retinopathy Lesions Based on Deep Learning

Table 4

Results obtained by using MESSIDOR dataset and different pretrained models on the proposed method.

Model	TP	FN	TN	FP	ACC	AUC	SEN
VGG16	220	0	660	0	100	100	100
VGG19	220	0	660	0	100	100	100
DenseNet201	217	3	650	10	98.5	100	98.5
DenseNet121	195	25	658	2	96.9	97.6	88.6
DenseNet169	192	28	647	13	95.3	91.3	87.2
MOBILENET	192	28	574	86	87.0	94.5	87.2
NASNet	190	30	570	90	86.3	96.5	86.4
InceptionV3	200	20	593	67	90.1	94.2	90.1
InceptionResNetV2	192	28	574	86	87.0	87.0	87.2
Resnet50	186	34	560	100	84.7	89.8	84.5