AI-based DR Screening Device Shows Favorable Real-World Results

In order to address the growing prevalence of diabetic retinopathy (DR) and the underdiagnosis of the disease, Digital Diagnostics developed IDx-DR. This is an AI-based diagnostic system that shows promise for the detection of mild, moderate and severe DR, as long as high-quality retinal images are captured, research finds. Photo: Optomed/LumineticsCore. Click image to enlarge. There are a number of ocular diseases that are vision threatening and continue to increase in prevalence year after year. Additionally, there are areas nationwide where physicians lack the resources and technology to diagnose and manage their patients. This has been studied before for numerous conditions and various populations, and the research is intended to spread awareness for groups that need more attention. Now, during this “Age of AI,” patients who struggle to receive routine health care and thus go undiagnosed might be able to get what they need.In a recent study published in Scientific Reports, researchers from Germany observed the real-world performance of the FDA-approved AI-based diagnostic system IDx-DR, now known commercially as Aeye-DS (LumineticsCore). By analyzing fundus images, the program can estimate whether a patient has no, mild, moderate or severe diabetic retinopathy (DR). Research assistants captured images for investigation, added them to the diagnostic system and then the team of researchers compared the results to mydriatic fundus examination and professional image analysis.A total of 875 diabetic patients were selected for this study, but only 555 patients (63.4%) had images that could be classified by IDx-DR. Investigators were not able to obtain an image from 92 patients (10.5%), and 228 patients (26.1%) had images that could not be analyzed by the diagnostic system. Of the remaining 555 patients, 41.3% were not diagnosed with DR , 31.4% were diagnosed as mild, 12.4% were diagnosed as moderate and 15% were diagnosed as severe.Next, researchers calculated the sensitivity and specificity of IDx-DR and compared it to mydriatic fundus examination. They found that severe DR diagnostic comparisons achieved the highest sensitivity (94.4%) and specificity (90.5%). Sensitivity was second highest in cases of no DR (56%) and specificity was second highest in cases of moderate DR (90.2). Additionally, IDx-DR’s results exactly matched 54.2% of professionally analyzed images.Along with the analysis, researchers determined the confounders that may have deterred their image acquisition and IDx-DR’s ability to analyze certain data. After establishing potential confounders, the investigators calculated the proportion of missing images and missing analyses for each possible variable. Confounders included the examiners themselves, pupil sizes, patients’ ages and visual acuities.“Looking into the future, it will be crucial to conduct external validation studies across diverse clinical settings and populations to evaluate the generalizability and external validity of AI-based DR screening systems like IDx-DR,” said the authors in their paper. “Additionally, research should explore the impact of examiner training and standardization protocols on image quality and acquisition performance. Assessing the effectiveness of targeted training programs, workflow optimizations, and quality assurance measures is essential to enhance the consistency and reliability of DR screening results in real-world settings.”Click here for the journal source. Hunfeld E, Tayar A, Paul S, et al. Real-world performance of the AI diagnostic system IDx-DR in the diagnosis of diabetic retinopathy and its main confounders. Scientific Reports. January 29, 2026. [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.