Pixel-Boundary-Dependent Segmentation Method for Early Detection of Diabetic Retinopathy

Table 3

Comparative performance of the published technique and the proposed method with IDRiD dataset.

Metrics	Lesion localization [10]	Attentive RefineNet [14]	Multifractal geometry [19]	Ridgelet features [20]	Proposed DS-KL
Accuracy	0.912	0.9217	0.985	0.910	0.9767
Dice value	0.821	0.813	0.868	0.890	0.9245
Sensitivity	0.940	0.8936	0.960	0.909	0.9865
Specificity	0.959	0.9328	0.973	0.910	0.9976