Table 3 Detailed results of applying the 1STL model with color normalization and partially balanced dataset.

From: Deep learning-based histotype diagnosis of ovarian carcinoma whole-slide pathology images

  

HGSC Concordance

CCOC Concordance

ENOC Concordance

LGSC Concordance

MUC Concordance

Cohen’s Kappa

F1 Score

AUC

Balanced Concordance

Internal Training Dataset

Model 1

82.02%

88.89%

81.63%

87.50%

54.55%

0.7393

0.7838

0.9404

78.92%

Model 2

84.72%

81.25%

72.73%

100.00%

78.57%

0.7448

0.8188

0.9547

83.45%

Model 3

82.95%

94.12%

67.31%

86.67%

77.78%

0.7294

0.7775

0.9473

81.76%

Mean

83.23%

88.09%

73.89%

91.39%

70.30%

0.7378

0.7934

0.9475

81.38%

External Test Dataset

Model 1

93.55%

100.00%

50.00%

100.00%

80.00%

0.7985

0.8377

0.9599

84.71%

Model 2

77.42%

100.00%

40.00%

100.00%

80.00%

0.6669

0.7124

0.9223

79.48%

Model 3

93.55%

100.00%

70.00%

50.00%

80.00%

0.7988

0.8022

0.9586

78.71%

Mean

88.17%

100.00%

53.33%

83.33%

80.00%

0.7547

0.7841

0.9469

80.97%

Ensemble Model

93.55%

100.00%

50.00%

75.00%

80.00%

0.7722

0.8085

0.9592

79.71%

  1. For the Internal Training Dataset, Models 1–3 refer to the models trained and tested on the three cross-validation splits of the Internal Training Dataset. For the External Test Dataset, Models 1–3 refer to the three models trained based on the cross-validation splits on the Internal Dataset and tested with the External Test Dataset.
  2. Bolded values refer to either the mean or ensemble results across Models 1–3.
Back to article page