Reliability and Validity of Ophthal-360: A Novel Artificial Intelligence Tool to Diagnose Non-Proliferative Diabetic Retinopathy

Abstract

Background: Diabetic retinopathy (DR) is a leading cause of preventable vision loss among working-age adults, with rising global and national prevalence, particularly in high-burden countries such as Pakistan. Artificial intelligence (AI)–based fundus image analysis offers a scalable solution for early detection and triage, yet local validation is essential before clinical integration. Objective: To evaluate the diagnostic validity and grading agreement of the Ophthal-360 AI tool for detecting and classifying non-proliferative diabetic retinopathy (NPDR) compared with ophthalmologist grading. Methods: This cross-sectional diagnostic accuracy study was conducted at a tertiary-care center in Lahore between June and December 2024. Adults aged 40–85 years with type II diabetes underwent standardized fundus photography. Images were independently graded by a fellowship-trained ophthalmologist (reference standard) using the International Clinical Classification of DR and by Ophthal-360. Sensitivity, specificity, predictive values, overall accuracy, area under the ROC curve (AUC), and weighted kappa were calculated. Results: Among 134 participants (mean age 57.8 ± 10.8 years; 53.7% male), DR prevalence was 77.6%. Ophthal-360 achieved sensitivity of 97.0% (95% CI: 91.8–99.4), specificity of 82.4% (95% CI: 65.5–93.2), PPV of 94.2%, NPV of 90.3%, overall accuracy of 93.3%, and AUC of 0.897. Agreement in NPDR severity grading was strong (κ = 0.86). Conclusion: Ophthal-360 demonstrates high sensitivity and strong grading concordance, supporting its potential role as an assistive screening tool for NPDR in resource-constrained settings.