Figure 1
From: Deep learning-based identification of eyes at risk for glaucoma surgery
Schematic of our deep learning model. Data augmentation techniques—random horizontal flip, zoom, rotation, and skew augmentation—were first applied to the VF-OCT stack. Then, spatially aligned VF and OCT images were input into the Vision Transformer (ViT). ViT-extracted features were then concatenated with VF, OCT, clinical and demographic data, and fed into a fully connected classifier to predict the occurrence of glaucoma surgery within the specified time horizon. This ViT architecture was described by Dosovitskiy et al.
