可解释性
老化
医学
斯科普斯
疾病
老年学
梅德林
健康老龄化
人工智能
系统回顾
黄斑变性
临床实习
生活质量(医疗保健)
纵向研究
科学网
人口老龄化
验光服务
考试(生物学)
风险评估
循证医学
荟萃分析
老年人
诊断试验
作者
Krithi Pushpanathan,Gabriel Dawei Yang,Min-kyu Kim,Qingsheng Peng,Zhi Da Soh,Siegfried K. Wagner,Yukun Zhou,Lisa Zhuoting Zhu,Can Can Xue,Yih-Chung Tham
标识
DOI:10.1016/j.apjo.2026.100391
摘要
Purpose Healthy ageing is a global health priority as populations age and chronic disease burden increases. Artificial intelligence (AI) applied to ocular imaging is a potential approach for deriving non-invasive markers of ageing. This review evaluates the development and validation of AI-derived ageing measures from ocular imaging, and their reported associations with health factors and outcomes. Design Systematic review Methods PubMed, Web of Science, and SCOPUS were searched from inception to February 1, 2026. Eligible studies applied AI to ocular imaging to derive age-related measures or evaluated their associations with health factors or outcomes. Two reviewers independently screened studies, extracted data, and assessed risk of bias using PROBAST+AI. Results Seventy-six studies (55 models) were included. Most models used chronological age as ground truth (n=53), reporting mean absolute errors of 2 to 6 years in internal test sets. External testing was limited, reported in only 16 studies. Reporting of outcomes and performance metrics was heterogeneous across studies. Interpretability analyses of fundus-based models consistently indicated attention to retinal vasculature and macular regions. Higher AI-derived ageing measures were generally associated with adverse health outcomes, although effect sizes varied across studies; most evidence was cross-sectional, with only one study evaluating longitudinal trajectories. Conclusions Ocular image-based AI-derived ageing measures show promise in capturing ageing signals and were associated with systemic health factors and outcomes. However, current reports are limited by heterogeneous reporting and limited external testing. Strengthening biological interpretability, standardisation of reporting metrics, validation across diverse populations, and temporal modelling is essential for future clinical translation.
科研通智能强力驱动
Strongly Powered by AbleSci AI