Table 2 Quantitative evaluation results on internal and external datasets.

From: Preserving privacy in surgical video analysis using a deep learning classifier to identify out-of-body scenes in endoscopic videos

Dataset

Videos (n)

ROC AUC (%)

Average precision (%)

F1-score (%)

Precision (%)

Recall (%)

Test

12

99.97

99.94

99.50

99.69

99.31

Gastric bypassa

10

99.94 ± 0.07

99.42 ± 0.26

96.10 ± 1.53

98.39 ± 0.81

93.96 ± 3.67

Center 1

5

99.99

99.60

97.18

97.82

96.55

Center 2

5

99.89

99.23

95.01

98.96

91.36

Cholecystectomya

20

99.71 ± 0.40

98.66 ± 1.27

94.74 ± 2.17

95.43 ± 5.62

94.37 ± 4.01

Center 3

5

99.83

99.00

92.78

87.20

99.12

Center 4

5

99.92

98.93

96.27

97.79

94.80

Center 5

5

99.12

96.85

92.98

96.95

89.33

Center 6

5

99.97

99.85

96.93

99.79

94.22

  1. aResults are mean ± standard deviation.
  2. ROC AUC, receiver operating characteristic area under the curve.