Table 2 Confusion matrices assessing the performance of the AI models, including the realistic performance of different models in the test set.

From: Establishment and validation of an artificial intelligence-based model for real-time detection and classification of colorectal adenoma

 

YOLOv4

Adaptive light-weight YOLOv4

SSD

Predicted class

Predicted class

Predicted class

Adenoma

Others

Adenoma

Others

Adenoma

Others

True class

 Adenoma

145

28

165

39

151

47

 Others

26

133

20

112

32

119

Accuracy (%)

92.95%

95.00%

92.36%

77.68%

88.43%

78.41%

FPR (%)

–

16.35%

–

15.15%

–

21.19%

 

Ensemble model

Proposed YOLOv3

Proposed YOLOv4

Predicted class

Predicted class

Predicted class

Adenoma

Others

Adenoma

Others

Adenoma

Others

True class

 

 Adenoma

178

12

124

32

130

27

 Others

9

101

26

138

20

102

Accuracy (%)

92.70%

81.45%

88.54%

78.23%

65.69%

53.43%

FPR (%)

–

8.18%

–

15.85%

–

16.39%

 

YOLOv7

YOLOv8

Predicted class

Predicted class

Adenoma

Others

Adenoma

Others

True class

 Adenoma

65

55

67

62

 Others

62

43

70

53

Accuracy (%)

76.47%

68.25%

78.82%

84.12%

FPR (%)

–

59.05%

–

56.91%

  1. Accuracy and FPR describe the predictions for adenomas and other types of polyps.
Back to article page