Artificial intelligence innovations in low vision rehabilitation: a systematic review of emerging technologies
TL;DR
AI assistive technologies for low vision demonstrate technical feasibility with promising performance, however, important evidence gaps limit clinical application: the absence of prospective validation, limited comparative effectiveness data, and minimal patient-centered outcomes.
Abstract
Low vision is one of the major types of visual impairment worldwide and when diagnosed early can help with its management. Artificial intelligence (AI) has recently shown promising outcomes in all fields including health, with low vision rehabilitation being one of its significant applications. This systematic review aims to provide an up-to-date and comprehensive evaluation of the application of AI in the management of low vision. The systematic review followed the Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) guidelines. All relevant published papers were retrieved through searches from PubMed, Scopus, Web of Science, and IEEE Xplore, without date restrictions. All retrieved articles from electronic databases were collated and managed in Covidence software. Nineteen studies were included, consisting of six application domains, which include assistive navigation and wayfinding ( n = 8), environmental interaction and object recognition ( n = 5), reading and text access ( n = 2), educational accessibility ( n = 1), medical applications ( n = 3), and digital accessibility optimization ( n = 1). Navigation systems demonstrated the strongest performance, with leading systems achieving sub-meter location error and 100% task success rates in real-world testing. Object recognition accuracies reached 97%, hazard detection achieved 92.0% dynamic accuracy with temporal filtering, and multimodal environmental understanding received risk assessment ratings of 9.40/10. Recurring technical limitations included sensitivity to lighting conditions, processing latency, and connectivity dependency. AI assistive technologies for low vision demonstrate technical feasibility with promising performance. However, important evidence gaps limit clinical application: the absence of prospective validation, limited comparative effectiveness data, and minimal patient-centered outcomes. Future studies should use standardized outcomes, larger and more diverse datasets, and longitudinal evaluations to demonstrate real-world impact. https://www.crd.york.ac.uk/PROSPERO/view/CRD420261324781 , CRD420261324781.
