Skip to main content

Table 5 Diagnostic performance of the commonly used algorithms in the literatures

From: Diabetic retinopathy risk prediction for fundus examination using sparse learning: a cross-sectional study

Methods* AUC (95% CI) Accuracy (%) (95% CI) Sensitivity (%) (95% CI) Specificity (%) (95% CI) PPV (%) NPV (%)
(A) Internal validation group (N = 163)
  SVM (RBF kernel) 0.83 (0.76-0.88) 74.8 (67.3-81.2) 71.0 (63.3-77.7) 75.8 (68.3-82.0) 40.7 91.7
  ANN 0.79 (0.72-0.85) 71.2 (63.5-77.9) 80.6 (73.6-86.3) 68.9 (61.2-75.9) 37.9 93.8
  Random Forest 0.80 (0.73-0.85) 72.4 (64.8-79.0) 87.1 (80.7-91.7) 68.9 (61.2-75.9) 39.7 95.8
  Naïve Bayes 0.76 (0.69-0.82) 74.2 (66.7-80.7) 74.2 (66.6-80.6) 74.2 (66.7-80.7) 40.4 92.5
  k-Nearest Neighbors 0.52 (0.45-0.59) 71.2 (63.5-77.9) 16.1 (11.0-22.8) 84.1 (77.4-89.3) 19.2 81.0
(B) External validation group (N = 562)
  SVM 0.81 (0.78-0.84) 74.1 (70.1-77.7) 75.7 (71.8-79.2) 73.7 (69.7-77.3) 42.6 92.1
  ANN 0.79 (0.76-0.83) 71.9 (67.8-75.6) 81.1 (77.5-84.3) 69.5 (65.4-73.3) 40.7 93.4
  Random Forest 0.76 (0.72-0.79) 71.1 (67.1-74.9) 69.4 (65.3-73.2) 71.6 (67.5-75.3) 38.7 90.0
  Naïve Bayes 0.73 (0.69-0.77) 70.6 (66.5-74.3) 69.4 (65.3-73.2) 70.9 (66.8-74.6) 38.1 89.9
  k-Nearest Neighbors 0.52 (0.48-0.57) 73.7 (69.7-77.3) 16.2 (13.3-19.6) 88.6 (85.5-91.1) 26.9 80.3
  1. *The models were trained and validated in scenario 3 without feature selection. The optimal conditions of each method were obtained in the 5-fold cross validation.
  2. ANN Artificial neural network, AUC Area under the receiver operating characteristic curve, CI Confidence interval, NPV Negative predictive value, PPV Positive predictive value, RBF Radial basis function, SVM Support vector machine.