International Journal of Medical Advances and Discoveries

International Scholars Journal of Medical Advances and Discoveries | Vol. 16, No. 8, August 2025 | pp. 49–56

DOI: 10.46882/2025/IJMAD/000107

Original Research Article

Title: Diagnostic Accuracy of Artificial Intelligence Algorithms in Detecting Early Diabetic Retinopathy from Fundus Photographs

Names of Authors: C. D. Garcia¹, D. E. Martinez², E. F. Kalu³

Authors’ Affiliations: ¹Department of Ophthalmology, San Jose Medical Center, Manila, Philippines; ²Department of Data Science, National Autonomous University, Mexico City, Mexico; ³Ophthalmology Unit, Lagos State University, Lagos, Nigeria

Abstract: Diabetic retinopathy (DR) is a leading cause of preventable blindness globally, yet screening programs are frequently hindered by a shortage of trained ophthalmologists. This diagnostic validation study assessed the performance of a deep convolutional neural network (CNN) in detecting referable DR (moderate non-proliferative DR or worse) and diabetic macular edema from un-mydriatic digital fundus photographs. A retrospective dataset of 12,500 retinal images obtained from diverse primary care screening sites was analyzed by the algorithm and benchmarked against a consensus reference standard established by three senior retinal specialists. The CNN achieved an overall sensitivity of 94.8% (95% CI, 93.1% - 96.2%) and a specificity of 91.5% (95% CI, 89.8% - 93.0%) for referable DR. Area under the receiver operating characteristic curve (AUC-ROC) reached 0.978. Processing time averaged 2.4 seconds per image. Subgroup analysis showed robust performance even in low-illumination or partially degraded image qualities, with sensitivity dropping by only 3.2% for images graded as suboptimal by human operators. Integration of this AI screening tool in community health centers demonstrated a 40% increase in throughput and earlier referral for sight-saving laser photocoagulation. These findings substantiate the clinical viability of automated deep learning systems as frontline triage mechanisms to expand screening coverage and prevent irreversible visual loss in high-risk diabetic populations.

Keywords: Artificial intelligence; Diabetic retinopathy; Deep learning; Fundus photography; Screening triage

Manuscript Timeline: Received: 10 April 2025; Revised: 05 July 2025; Accepted: 22 July 2025; Published: 05 August 2025

Citation: Garcia, C. D., Martinez, D. E., & Kalu, E. F. (2025). Diagnostic Accuracy of Artificial Intelligence Algorithms in Detecting Early Diabetic Retinopathy from Fundus Photographs. International Journal of Medical Advances and Discoveries, 16(8), 49–56.

Subscribe to read the full article: https://www.9journals.org/subscribe-to-read