Figure 1

Box plots of evaluation metrics AUC, accuracy, sensitivity, and specificity obtained through tenfold cross-validation on the test data. This figure presents the box plot of the evaluation metrics AUC, accuracy, sensitivity, and specificity obtained through tenfold cross-validation on the test data. Subfigure (a) illustrates the results for the Macular-OCT model, while subfigure (b) displays the results for the ONH-OCT model. The AUC values box plot reveals that, for both macular-OCT and ONH-OCT datasets, the 2D CNN model delivers superior overall results. Additionally, the 2D model yields robust results for other essential metrics such as accuracy and sensitivity. Among the 3D CNN models, the pre-trained 3D-ResNet18 outperformed the 3D-ResNet18 model trained from scratch. Notably, the 3D-CNN-Encoder exhibits superior performance compared to the 3D-ResNet18 models when tested on the ONH-OCT dataset, while delivering subpar performance on the macular-OCT dataset. AUC area under the receiver operating characteristic curve, OCT optical coherence tomography, macular-OCT macural-OCT, ONH-OCT optic nerve head OCT.