The increasing use of Optical Coherence Tomography (OCT) and OCT angiography (OCTA) in ophthalmology has generated large and complex imaging datasets, making manual analysis challenging, particularly for early or preclinical disease detection. Artificial intelligence (AI) has enabled automated OCT analysis for segmentation, classification, and biomarker extraction; however, conventional convolutional neural networks (CNNs) are limited by their dependence on large annotated datasets and restricted global context modeling. Unlike previous reviews of OCT-AI that mainly focus on CNN-based methods or general disease classification, this review specifically examines how transformer-based, self-supervised, and foundation models support reliable early disease detection, where structural and microvascular changes are subtle and labeled data are scarce. We analyze their applications in diabetic retinopathy, age-related macular degeneration, and glaucoma, and critically compare them with CNN-based approaches. Key challenges related to data scarcity, inter-device variability, generalization, and clinical translation are discussed, along with future directions toward robust clinical deployment.
Sajid et al. (Mon,) studied this question.