Table 3 Diagnostic performance of random forest with LASSO for discriminating cytomegalovirus esophagitis from herpes simplex virus esophagitis.

From: Machine learning approach for differentiating cytomegalovirus esophagitis from herpes simplex virus esophagitis

Random forest with LASSO

Sen. (%)

Spec. (%)

PPV (%)

NPV (%)

Acc (%)

AUC

HSB color model

Without clinical factor* (per-ROI/per-image/per-patient)

99.3 (0.8)

98.8 (0.9)

96.5 (2.7)

99.8 (0.3)

99.0 (0.7)

0.999 (0.001)

99.7 (0.9)

99.7 (0.5)

99.5 (0.7)

99.8 (0.5)

99.7 (0.5)

1.0 (0.001)

100 (0.0)

100 (0.0)

100 (0.0)

100 (0.0)

100 (0.0)

1.0 (0.0)

With clinical factor* (per-ROI/per-image/per-patient)

99.2 (0.8)

98.8 (0.7)

96.2 (2.8)

99.7 (0.3)

98.9 (0.7)

0.999 (0.001)

99.8 (0.6)

99.4 (0.8)

99.1 (1.2)

99.8 (0.4)

99.6 (0.6)

1.0 (0.001)

100 (0.0)

100 (0.0)

100 (0.0)

100 (0.0)

100 (0.0)

1.0 (0.0)

RGB color model

Without clinical factor* (per-ROI/per-image/per-patient)

72.0 (13.0)

67.1 (12.9)

42.6 (9.8)

88.4 (5.1)

68.5 (7.8)

0.745 (0.066)

71.8 (14.1)

68.5 (13.6)

59.9 (8.7)

81.0 (7.0)

70.0 (6.0)

0.737 (0.073)

78.0 (16.6)

74.6 (18.0)

72.6 (13.3)

84.6 (9.8)

76.1 (8.0)

0.752 (0.093)

With clinical factor* (per-ROI/per-image/per-patient)

65.2 (16.8)

73.4 (12.4)

45.3 (11.8)

87.1 (6.1)

71.2 (7.0)

0.741 (0.082)

68.9 (13.8)

72.7 (16.3)

63.9 (12.5)

79.9 (6.4)

71.2 (6.9)

0.749 (0.076)

77.1 (16.1)

75.0 (16.3)

72.6 (13.7)

83.9 (8.7)

76.1 (7.1)

0.769 (0.075)

  1. Results were obtained per-ROI (top), per-image (center), and per-patient (bottom), and presented as average (standard deviation) of five repeated five-fold cross-validation.
  2. ROI region of interest, HSB hue–saturation–brightness, RGB red–green–blue, Sen. sensitivity, Spec. specificity, PPV positive predictive value, NPV negative predictive value, Acc. accuracy, AUC area under the ROC curve, ROC receiver operating characteristic.
  3. *Clinical factor: previous history of transplantation.