Abstract

Background: Childhood visual impairment can adversely affect visual development, educational attainment, psychosocial well-being, and quality of life. Early identification of refractive errors, amblyopia risk factors, strabismus, and other ocular abnormalities is therefore an important component of paediatric eye care. However, conventional screening programmes may be constrained by shortages of trained personnel, limited infrastructure, difficulties in examining young children, and inadequate referral follow-up. Artificial Intelligence (AI), particularly machine learning, deep learning, and computer vision, offers potential approaches to improving the accessibility and consistency of paediatric vision screening.
Objective: This review examines current evidence on AI-assisted paediatric vision screening, with particular emphasis on refractive error detection, amblyopia risk assessment, strabismus identification, smartphone-based screening, and potential public health applications.
Methods: A narrative review approach was used to synthesise relevant published research, clinical validation studies, systematic reviews, and World Health Organization guidance on paediatric vision screening and AI-enabled ophthalmic assessment. Evidence was examined according to screening modality, target condition, diagnostic performance, clinical applicability, and implementation challenges.
Results: Published studies demonstrate promising applications of AI-assisted screening. A smartphonebased deep-learning system for detecting visual impairment in children younger than four years reported an area under the receiver operating characteristic curve of 0.940 in internal validation and 0.843 in external validation. A separate deep-learning photoscreening study reported 90.8% accuracy for identifying amblyopia risk factors. Smartphone-based systems have also demonstrated potential for refractive error classification and strabismus detection. Nevertheless, performance varies according to the algorithm, target condition, patient population, reference standard, and screening environment. Limited external validation, population bias, inconsistent reporting, data privacy concerns, and uncertainty regarding cost-effectiveness remain important barriers to implementation.
Conclusion: AI-assisted vision screening may complement conventional paediatric optometric services by supporting early risk identification, task-sharing, and referral prioritisation. Its implementation should be guided by prospective clinical validation, transparent performance reporting, human oversight, data protection, and reliable referral pathways. AI should be regarded as an adjunct to, rather than a replacement for, comprehensive paediatric eye examination.

Keywords:Artificial intelligence; Paediatric optometry; Vision screening; Machine learning; Deep learning; Refractive error; Amblyopia; Strabismus; School eye health and Public health