EAD-Net: A Novel Lesion Segmentation Method in Diabetic Retinopathy Using Neural Networks

Table 4

Statistical information of lesion areas of e-ophtha, IDRiD, and local intelligent ophthalmology datasets.

Dataset	MAs	HEs	Hard exudates	Soft exudates
E-ophtha	0.01% (148)	—	0.22% (47)	—
IDRiD	0.10% (81)	1.03% (80)	0.90% (81)	0.38% (40)
Local	0.02% (63)	0.91% (84)	0.48% (86)	0.32% (29)

Number1 (Number2) refers to the fact that there are Number2 images of this lesion type in the corresponding dataset, and the average percentage of this lesion area to the total image area is Number1. To maintain data consistency, only the images containing lesions have been used in Number1.