Table 5 Local prediction performance of VF threshold sensitivity values from OCT data (excluding any augmentation using MICE and SMOTE).

From: Predicting visual field global and local parameters from OCT measurements using explainable machine learning

ML model

Folds

Threshold sensitivity (dB)

Grayscale pixel values

MAE (original TS)

MAE (normalised TS)

MAE

MSSI

XGBoost

Fold-1

3.13

2.60

9.50

0.74

Fold-2

2.98

2.51

10.77

0.71

Fold-3

3.39

2.85

11.2

0.71

Fold-4

3.06

2.65

10.83

0.71

Fold-5

3.20

2.66

10.84

0.71

Mean ± STD

3.15 ± 0.16

2.66 ± 0.13

10.63 ± 0.65

0.72 ± 0.01

95% CI

3.01–3.29

2.55–2.77

10.06–11.2

0.71–0.72

SVM

Fold-1

3.24

2.82

9.32

0.77

Fold-2

3.00

2.59

10.31

0.75

Fold-3

3.34

2.93

11.34

0.75

Fold-4

3.13

2.76

9.63

0.76

Fold-5

3.35

2.86

9.87

0.76

Mean ± STD

3.21 ± 0.15

2.79 ± 0.13

10.09 ± 0.79

0.76 ± 0.01

95% CI

3.08–3.34

2.68–2.9

9.4–10.78

0.75–0.76

RF

Fold-1

2.84

2.40

9.48

0.80

Fold-2

2.87

2.41

10.54

0.78

Fold-3

3.18

2.70

10.68

0.78

Fold-4

2.77

2.41

10.77

0.77

Fold-5

2.98

2.54

10.19

0.78

Mean ± STD

2.93 ± 0.16

2.49 ± 0.13

10.33 ± 0.53

0.78 ± 0.01

95% CI

2.79–3.07

2.38–2.61

9.87–10.79

0.77–0.79

  1. Boldface numbers indicate the best performance per metric. The results are obtained from the clean data without MICE and SMOTE-based augmentation.
Back to article page