Artificial Intelligence and Deep Learning in Ophthalmology

Abstract

Artificial intelligence (AI), in particular deep learning (DL), has gained significant interest recently from healthcare systems. DL has been widely applied to detect and classify major diseases in ophthalmology, including diabetic retinopathy (DR), age-related macular degeneration (AMD), glaucoma, and retinopathy of prematurity based on fundus photographs; cataract and anterior segment diseases, glaucoma, and retinal diseases based on optical coherence tomography (OCT) scans; and glaucoma progression based on visual fields. The substantial progress of AI in ophthalmology has involved the identification of clear public health (e.g., DR screening) and clinical (e.g., prediction of the need to treat AMD) unmet needs, the targeted development of the AI algorithms using both retrospective and prospective clinical and imaging data, and designing the application interface for clinical deployment. In ophthalmology, there has also been significant experience of applying AI algorithms in “real-world” clinical situations, as well as the submission and approval by governmental regulatory agencies (e.g., Food and Drug Administration). Future research is warranted to address not only technical issues (e.g., explainability of the “black box”) but also a range of nontechnical challenges, such as increasing the awareness and acceptance of physician and patient, issues relating to global collaboration and data sharing, medical ethics, financial and reimbursement systems, and integration of AI algorithms in clinical settings with diverse electronic health records.