DEEP LEARNING–ENABLED SCREENING OF DIABETIC RETINOPATHY USING RETINAL IMAGING DATA FROM SOUTH ASIAN POPULATIONS: EVIDENCE FROM PAKISTAN

Abstract

Background: Diabetic retinopathy (DR) is a leading cause of preventable blindness among individuals with diabetes, particularly in low- and middle-income countries such as Pakistan. Deep learning has emerged as a promising approach for automated retinal image analysis and early disease detection.

Objective: This study evaluated the effectiveness of deep learning-enabled retinal image analysis for automated diabetic retinopathy screening using retinal imaging data from South Asian populations, with evidence from Pakistan.

Methods: A quantitative cross-sectional study was conducted using retinal fundus images from 600 patients with diabetes collected from selected hospitals in Pakistan. A deep learning model based on a Convolutional Neural Network (CNN) was developed and evaluated using standard performance metrics. Structural relationships were analyzed using PLS-SEM.

Results: The deep learning model achieved an overall accuracy of 95.4% and an AUC-ROC of 0.98, demonstrating excellent diagnostic performance. The findings revealed that deep learning-enabled retinal image analysis significantly improved automated detection of diabetic retinopathy, early diagnosis, clinical decision-making, and overall screening outcomes (p < .001).

Conclusion: Deep learning-enabled retinal screening is an effective and reliable approach for the early detection of diabetic retinopathy in Pakistan. Integrating AI-assisted retinal imaging into routine diabetic eye care can improve screening efficiency, facilitate timely intervention, and reduce preventable vision loss.