Table 4 Class-wise performance metrics for RP, STGD, and healthy eyes using different inputs.

From: A deep learning model for diagnosis of inherited retinal diseases

  

RP

STGD

Healthy eyes

Training mode

Support

66

38

58

Train images

250

86

284

Test images

66

38

58

Total images

316

124

342

Total patient

158

62

171

Prevalence

0.4074

0.2345

0.3580

Single-input

(CFP)

Precision

0.9412

0.9143

0.9661

Sensitivity

0.9697

0.8421

0.9828

Specificity

0.9583

0.9758

0.9808

NPV

0.9787

0.9528

0.9903

F1

0.9552

0.8767

0.9744

AUC

0.9813

0.9644

0.9973

Single-input

(IR)

Precision

0.8919

0.9688

0.9508

Sensitivity

0.8684

0.9394

1

Specificity

0.9677

0.9792

0.9712

NPV

0.96

0.9592

1

F1

0.88

0.9538

0.9748

AUC

0.9824

0.9902

0.9988

Multi-input

(CFP + IR)

Precision

0.9412

0.9444

1

Sensitivity

0.9697

0.8947

1

Specificity

0.9583

0.9839

1

NPV

0.9787

0.9683

1

F1

0.9552

0.9189

1

AUC

0.9905

0.9868

1

  1. Signifiacnce value bold.
Back to article page