Abstract / Summary
Within the selected corpus, there are image-based ophthalmology trials and artificial intelligence validations for bounded tasks. Fundus photography and optical coherence tomography allow detection, classification, and referral tasks to be delimited; however, algorithmic performance alone does not equate to clinical benefit, gap closure, or safe implementation. The most mature evidence concentrates on diabetic-retinopathy screening. A specific system received FDA De Novo authorization and, in its pivotal trial, reported 87.2% sensitivity and 90.7% specificity for detecting more-than-mild diabetic retinopathy in a defined population and indication. Those figures do not transfer to other tools or to Latin America. In retina and OCT, studies and reviews support classification and referral tasks, but validation depends on equipment, capture protocols, prevalence, and reference standard. In glaucoma, an SD-OCT meta-analysis reported a pooled sensitivity of 0.91 and specificity of 0.90, along with heterogeneity that limits any reading of universal readiness. The direct regional evidence identified includes two Mexican evaluations of AI-human support or comparison: a randomized reader experiment for diabetic-retinopathy images and an evaluation of 435 patients in tertiary care for glaucoma and retinal disease. Both works report performance in Mexican settings, not clinical outcomes, routine implementation, or validity for all of Latin America. A Brazilian economic model evaluated teleophthalmology without AI and contributes system context, not algorithmic effectiveness. For Latin American clinical teams, the current responsible use is support for screening, reading, prioritization, and referral within a governed care pathway. It is not yet possible to promise autonomous comprehensive ocular diagnosis, outcome benefit, reduction of inequities, or generalization across cameras, populations, and systems. The responsible agenda begins with retrospective feasibility, image governance, definition of the human response, and subgroup evaluation. Only after clinical and institutional approval would it be appropriate to design a silent prospective evaluation.