AI-based DR Screening System Exhibits High Diagnostic Accuracy but Low Usage

Diabetic retinopathy (DR) remains the leading cause of avoidable blindness in working-age individuals globally. Despite the efficacy of early detection through routine retinal exams, less than half of people with diabetes receive necessary screenings, partly due to healthcare limitations and a shortage of ophthalmologists. In response to this challenge, various artificial intelligence (AI)-based tools have been developed to automate DR screenings, enabling point-of-care retinal analysis and providing immediate triage without requiring ophthalmologist oversight. While multiple studies have showcased the ability of such systems to achieve diagnostic accuracy comparable to human graders, barriers such as reimbursement and workflow integration still hinder widespread adoption, leaving less than 5% of eligible patients screened in the US.A recent investigation concentrated on one AI-based DR screening tool, EyeArt (from manufacturer Eyenuk), which was approved by the FDA in 2020. The study conducted a thorough literature search across PubMed, Embase and ClinicalTrials.gov to assess the diagnostic accuracy of this autonomous AI system in detecting referable DR from color fundus photographs. Seventeen studies involving 162,695 examinations met eligibility criteria and were included in the analysis. In a review of 17 real-world studies encompassing 162,695 examinations, the FDA-approved AI screening device called EyeArt demonstrated strong diagnostic accuracy in identifying referable diabetic retinopathy, achieving a pooled sensitivity of 95% and specificity of 81%, with high confidence in its sensitivity and moderate confidence in its specificity. Photo: EyeNuk. Click image to enlarge. The results showed that EyeArt demonstrates high diagnostic accuracy for detecting referable DR, with a pooled sensitivity of 95% and specificity of 81%. The findings indicate strong certainty regarding its sensitivity, supporting the case for autonomous screening in primary care settings. However, variability in specificity and inconsistent reporting for ungradable images necessitate careful attention and the establishment of standardized quality assurance measures.“Many investigations lacked standardized definitions or quantification of ungradable images and often failed to detail how such images were processed by the AI system—whether auto-flagged as positive or excluded—impeding granular subgroup analyses,” the researchers explained in their paper. Adoption of standardized reporting frameworks in future studies “would substantially enhance comparability and interpretability,” they noted.Despite the system’s strong diagnostic performance, the study authors pointed out that US claims data show that the billing code for EyeArt was used in only 3,440 cases out of over 154,000 diabetic-eye imaging encounters from 2021 to 2023, accounting for just 2.2% of encounters. One factor contributing to the low real-world uptake of EyeArt in clinical practice is its modest reimbursement rate of approximately $40.28, which falls far below that of other AI applications (e.g., stroke CT). The initial costs for equipment and IT integration also deter primary care adoption. Competitors like LumineticsCore (Digital Diagnostics)—formerly known as IDx-DR—and AEye-DS (AEye Health) present further challenges, with comparable or superior performance and workflow flexibility; thus, the authors argue the need for more comprehensive cost-effectiveness studies to reinforce EyeArt’s position in clinical settings.Taking all of these findings into consideration, the authors summarized, “These data support EyeArt as a safe rule-out tool for population screening, provided programs anticipate and manage the referral burden from false positives through standardized image-quality protocols, repeat/confirmatory pathways and seamless workflow integration.” The successful adoption of the AI screening tool will require not only effective integration into clinical workflows and electronic health records but also sustainable reimbursement strategies and focused efforts to reach underserved populations to maximize its public health impact, they concluded.Click here for the journal source. Wang TW, Luo WT, Tu YK, et al. Diagnostic accuracy of EyeArt for fundus-based detection of diabetic retinopathy: a systematic review and meta-analysis. Am J Ophthalmol. October 3, 2025. [Epub ahead of print].  This article was developed by the editorial staff in conjunction with experts in the field. In the process, AI may have been among the editorial tools used to meet the goals of human editors, who approved all content.