Table 2 Performance of the three different regression models for VF prediction from OCT data (including augmented data using MICE and SMOTE) with fivefold cross-validation (report on test results with fivefold cross-validation, mean ± standard deviation).

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

ML Model

Folds

MD

VFI

PSD

R

RMSE (dB)

MAE (dB)

R

RMSE

MAE

R

RMSE (dB)

MAE (dB)

XGBoost

Fold-1

0.80

3.14

2.24

0.89

7.27

4.29

0.85

1.95

1.32

Fold-2

0.74

3.35

2.30

0.72

9.09

4.67

0.69

2.51

1.70

Fold-3

0.75

3.50

2.47

0.85

7.72

4.75

0.76

2.48

1.66

Fold-4

0.73

3.80

2.80

0.75

10.05

4.90

0.77

2.43

1.61

Fold-5

0.66

4.43

2.91

0.81

7.44

4.79

0.75

2.34

1.47

Mean ± STD

0.74 ± 0.05

3.64 ± 0.5

2.54 ± 0.3

0.80 ± 0.07

8.31 ± 1.21

4.68 ± 0.23

0.76 ± 0.06

2.34 ± 0.23

1.55 ± 0.16

95% CI

0.69–0.78

3.21–4.08

2.28–2.8

0.74–0.86

7.26–9.37

4.47–4.88

0.71–0.81

2.14–2.54

1.42–1.69

SVM

Fold-1

0.79

3.23

1.95

0.75

12.29

5.89

0.76

2.51

1.57

Fold-2

0.71

3.46

2.38

0.73

9.80

4.88

0.71

2.56

1.63

Fold-3

0.80

3.26

2.24

0.78

10.66

5.86

0.83

2.51

1.69

Fold-4

0.77

3.24

2.36

0.68

12.45

6.03

0.75

2.33

1.66

Fold-5

0.65

4.62

2.92

0.66

10.03

5.47

0.68

2.56

1.62

Mean ± STD

0.74 ± 0.06

3.56 ± 0.6

2.37 ± 0.35

0.72 ± 0.05

11.05 ± 1.25

5.63 ± 0.47

0.75 ± 0.06

2.49 ± 0.10

1.63 ± 0.05

95% CI

0.69–0.80

3.03–4.09

2.06–2.68

0.68–0.76

9.95–12.14

5.22–6.03

0.7–0.79

2.41–2.58

1.59–1.67

RF

Fold-1

0.78

2.80

2.12

0.80

6.73

4.41

0.74

2.76

1.90

Fold-2

0.75

4.11

2.66

0.80

10.51

5.97

0.75

2.20

1.64

Fold-3

0.76

3.34

2.31

0.75

8.66

5.23

0.65

2.57

1.81

Fold-4

0.85

2.67

1.83

0.84

6.86

4.16

0.86

1.92

1.34

Fold-5

0.67

3.61

2.49

0.66

10.04

6.18

0.70

2.63

1.76

Mean ± STD

0.76 ± 0.07

3.30 ± 0.59

2.28 ± 0.32

0.77 ± 0.07

8.56 ± 1.75

5.19 ± 0.90

0.74 ± 0.08

2.42 ± 0.35

1.69 ± 0.22

95% CI

0.70–0.82

2.78–3.82

2.00–2.56

0.71–0.83

7.03–10.09

4.40–5.98

0.67–0.80

2.11–2.72

1.50–1.88

  1. Boldface numbers indicate the best performance for that metric. Input features included RNFL, GC-IPL and MC thickness data in the spatial and frequency domain. 95% CI = Mean ± 1.96 × Standard Error (SE), SE = STD/\(\sqrt{\text{n}}\), n = 5, CI:Confidence Interval, MD mean deviation, VFI visual field index, PSD pattern standard deviation, RMSE root mean squared error, MAE mean absolute error.
Back to article page