ARTIFICIAL INTELLIGENCE IN OPHTHALMOLOGY: CURRENT APPLICATIONS IN SCREENING, DIAGNOSIS, AND DISEASE MONITORING
TL;DR
Artificial intelligence should currently be considered primarily as a decision -support technology that complements rather than replaces ophthalmologists for its safe and effective implementation.
Abstract
Artificial intelligence has rapidly become an important technological component of modern ophthalmology, particularly in the analysis of ophthalmic images and the automated detection of ocular pathology. The increasing availability of digital fundus photography, optical coherence tomography, optical coherence tomography angiography, and other imaging modalities has created large datasets suitable for machine learning and deep learning applications. AI-based systems are currently being investigated and imple mented for screening, diagnosis, classification, prognosis, and longitudinal monitoring of multiple ophthalmic diseases. The most established applications involve diabetic retinopathy, glaucoma, age-related macular degeneration, diabetic macular edema, and retinopathy of prematurity. Deep learning algorithms can identify characteristic retinal lesions, assess optic nerve head changes, quantify retinal fluid, and detect structural abnormalities with performance approaching that of experienced ophthalmic graders in selected clinical tasks. Despite these advances, several limitations remain, including dataset bias, variability between imaging devices, insufficient external validation, limited interpretability, regulatory challenges, and uncertainty regarding clinical responsibility for AI- generated decisions. [10,11] Therefore, AI should currently be considered primarily as a decision -support technology that complements rather than replaces ophthalmologists. Proper integration of AI into clinical workflows, continuous validation, transparency, and appropriate human oversight will be essential for its safe and effective implementation.