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

Table 3

Comparison with U-net on local intelligent ophthalmology dataset.

Lesion type	Model	Lesion based results
		SE	SP	PR	ACC	F1
MAs	U-net	13.17	99.97	54.07	99.90	21.19
	EAD-Net	17.32	99.98	59.26	99.91	26.82
HEs	U-net	73.43	99.93	80.21	99.83	76.67
	EAD-Net	83.59	99.95	87.75	99.89	85.62
Hard exudates	U-net	68.38	99.99	98.42	99.96	80.70
	EAD-Net	84.60	99.99	93.51	99.98	88.83
Soft exudates	U-net	76.89	99.99	98.86	99.98	86.50
	EAD-Net	84.92	99.99	92.78	99.98	88.68

SE: sensitivity; SP: specificity; PR: precision; ACC: accuracy; F1: F1 score.