Clinical Study

Glaucoma Diagnostic Accuracy of Machine Learning Classifiers Using Retinal Nerve Fiber Layer and Optic Nerve Data from SD-OCT

Table 4

Areas under the receiver operating characteristic curve (aROCs) of best parameter (BP) and all 23 parameters (AP) obtained with machine learning classifiers and sensitivities (%) with fixed specificities of 80% and 90% for AP.

MLCaROC-BP (CI) [NP]aROC-AP (CI)Specificity 80%-APSpecificity 90%-AP

RAN0.877 (0.810–0.944) [13]0.805 (0.738–0.872)64.949.1
NB0.870 (0.801–0.939) [11]0.818 (0.749–0.939)68.452.6
RBF0.866 (0.796–0.936) [11]0.839 (0.746–0.898)71.963.1
MLP0.843 (0.768–0.918) [11]0.768 (0.693–0.918)49.147.3
ADA0.839 (0.763–0.915) [19]0.839 (0.763–0.915)73.652.6
ENS0.829 (0.751–0.907) [08]0.793 (0.715–0.871)61.456.1
BAG0.828 (0.749–0.907) [12]0.804 (0.725–0.883)57.850.8
SVMG0.825 (0.746–0.904) [10]0.753 (0.674–0.832)56.028.0
SVML0.780 (0.692–0.868) [02]0.690 (0.602–0.778)45.022.5
CTREE0.733 (0.684–0.862) [07]0.687 (0.638–0.736)46.023.0

MLC: machine learning classifier; aROC: area under the ROC curve; BP: best parameter; AP: all parameters; NP: number of parameters; CI: confidence interval of 95%; BAG: bagging; NB: Naive-Bayes; SVML: linear support vector machine; SVMG: Gaussian support vector machine; MLP: multilayer perceptrons; RBF: radial basis function; RAN: random forest; ENS: ensemble selection; CTREE: classification trees; ADA: AdaBoost.