Table 3.

GON detection accuracy of deep learning models on test dataset.

Model	AUC (95% CI)	P-value
VGG16
Native	0.83 (0.81–0.85)	<0.001*
Transfer Learning	0.89 (0.87–0.91)	0.045*
Inception
Native	0.83 (0.81–0.85)	<0.001*
Transfer Learning	0.90 (0.88–0.92)	0.11*
ResNet50
Native	0.89 (0.87–0.91)	0.01*
Transfer Learning	0.91 (0.90–0.93)	—

*AUC significantly (p < 0.05) lower than the transfer learning ResNet50 model.