Table 4 Method comparison results.

From: A vision transformer architecture for the automated segmentation of retinal lesions in spectral domain optical coherence tomography images

Method

Sensitivity

Specificity

IRF

PED

SRF

MEAN

IRF

PED

SRF

MEAN

U-Net

0.389

(0.078)

0.553

(0.080)

0.534

(0.093)

0.504

(0.082)

0.999

(0.001)

0.994

(0.002)

0.999

(0.001)

0.997

(0.001)

U-Net3+

0.526

(0.047)

0.640

(0.034)

0.600

(0.014)

0.600

(0.033)

0.999

(0.000)

0.995

(0.001)

0.999

(0.000)

0.997

(0.001)

Swin-

UNETR-

24

0.497

(0.033)

0.616

(0.061)

0.589

(0.035)

0.578

(0.048)

0.999

(0.000)

0.993

(0.003)

0.999

(0.001)

0.996

(0.002)

Swin-

UNETR-

48

0.504

(0.039)

0.628

(0.028)

0.587

(0.045)

0.585

(0.035)

0.999

(0.000)

0.993

(0.002)

0.999

(0.001)

0.996

(0.001)

  1. Mean and (std) of the sensitivity and specificity calculated on the test set over a 5-fold CV per class and as a weighted average.
Back to article page