Skip to main content

Table 5 Performance analysis of all algorithms on HRF databases with respect to the measuring metrics, as well as training time per epoch and prediction time on one image (in second)

From: Retinal vessel segmentation with constrained-based nonnegative matrix factorization and 3D modified attention U-Net

Methods

Accuracy

Specificity

Sensitivity

Precision

Training

Prediction

Soares et al. [13]

0.9373

0.9671

0.6253

0.6444

–

6

U-Net [1]

0.9577

0.9701

0.6577

0.7308

12

10

AG-UNet [35]

0.9600

0.9875

0.8297

0.8268

29

13

IterNet [26]

0.9623

0.9743

0.7524

0.7284

428

53

DenseNet [36]

0.9639

0.9812

0.7610

0.7747

67

22

V-GAN [37]

0.9617

0.9858

0.8196

0.8213

107

32

M-GAN [38]

0.9700

0.9931

0.6948

0.8800

112

36

Ours

0.9688

0.9903

0.7451

0.8947

89

17