Abstract:
The novel addition of artificial intelligence to the current diagnostic methods is the screening and management of glaucoma. The main diagnostic method uses ophthalmoscopy to
assess the optic nerve and tonometry to detect intraocular pressure. Central visual field
tests, on the other hand, provide a clear picture of glaucoma patients’ functional vision
and can provide more detailed information on the disease’s progression. However, the
early identification and prediction of glaucoma is being revolutionized by AI-driven technologies, particularly self-supervised learning (SSL) and Vision Transformer (ViT)-based
models. In order to improve glaucoma detection and classification based on color fundus
photography images, a specially created SSL framework called RetFound uses advanced
self-supervised learning. It learns detailed retinal structures from large-scale unnamed
datasets, which lessens the need for large, fully manually annotated datasets while also
greatly increasing diagnostic accuracy. In order to accurately detect glaucomatous alterations including optic nerve cupping, neuroretinal rim thinning, and retinal nerve fiber
layer abnormalities, the Vision Transformer model further improves the feature extraction
capabilities. In order to distinguish between distinct phases and subtypes of glaucoma
and aid in the diagnosis process, this research investigates binary classification, glaucoma
vs non-glaucoma, and multi-class classification. This study explores the difficulties RetFound faces, AI-powered approaches, and the future of glaucoma treatment; it is setting a
new benchmark for proactive patient care and precise diagnosis.
