The following is a summary of “Diagnostic accuracy of artificial intelligence based automated diabetic retinopathy screening in real-world settings: a systematic review and meta-analysis,” published in the March 2024 issue of Ophthalmology by Joseph et al.
Researchers conducted a retrospective study to evaluate the real-world effectiveness of artificial intelligence (AI)- driven automated screening for diabetic retinopathy (DR) and determine its diagnostic accuracy.
They systematically reviewed pertinent literature from January 2012 to August 2022, utilizing PubMed, Scopus, and Web of Science databases. The included studies assessed quality using the Quality Assessment for Diagnostic Accuracy Studies 2 (QUADAS-2) checklist. Summary measures were calculated, including pooled accuracy, sensitivity, specificity, and diagnostic odds ratio (DOR). The study protocol was registered with the International Prospective Register of Systematic Reviews (PROSPERO – CRD42022367034).
The results showed that of 34 studies employing AI algorithms for diagnosing DR with real-world fundus images, the quality assessment indicated a low risk of bias and low applicability concern. Among gradable images, the overall pooled accuracy, sensitivity, specificity, and DOR were 81%, 94% (95% CI: 92.0-96.0), 89% (95% CI: 85.0-92.0), and 128 (95% CI: 80-204). Sub-group analysis indicated that when acceptable quality imaging could be obtained, non-mydriatic fundus images had a better DOR of 143 (95% CI: 82-251), and studies utilizing two-field images had a better DOR of 161 (95% CI: 74-347). The meta-regression analysis revealed a statistically significant association between DOR and variables such as income status and the type of fundus camera.
Investigators concluded that AI algorithms demonstrate promising performance in DR screening using fundus images, potentially aiding ophthalmologists in reducing workload and improving diagnostic accuracy.
Source: ajo.com/article/S0002-9394(24)00066-7/abstract
