Table 4.

AROC values obtained with other models used to diagnose glaucoma.

	all eyes (N = 110)	N and G groups	mN and mG groups
a CNN with 16 layers, similar to VGG16	86.3 [79.9–93.0]	81.8 [71.2–91.4]	91.2 [83.5–99.0]
Random Forests	77.5 [69.6–85.4]	76.8 [65.9–87.7]	78.3 [66.9–89.6]
Support Vector Machine	71.1 [62.7–79.5]	75.1 [64.1–86.1]	66.2 [53.0–79.5]

AROC [95% confidence interval] values were calculated by training using (i) CNN with 16 layers, similarly to VGG16, (ii) support vector machine, and (iii) Random Forest, using all of the training dataset, and validating using the testing dataset.

AROC: area under the receiver operating characteristic curve, CNN: convolutional neural network, G: non-highly myopic glaucoma patients, N: non-highly myopic normative subjects, mG: highly myopic glaucoma patients and N: highly myopic normative subjects.