Speaker
Priya Sharma
(All India Institute of Medical Sciences (AIIMS), New Delhi)
Description
Background: Diabetic retinopathy affects approximately 103 million people worldwide. Early detection is critical, yet many healthcare systems lack sufficient ophthalmologists.
Objective: To develop and validate a deep learning model for automated detection of referable DR from fundus photographs.
Methods: Trained EfficientNet-B4 on 128,457 images, validated on 15,832 from three independent centers in India, Nigeria, and Brazil.
Results: AUC-ROC 0.967 (95% CI: 0.961-0.973). Sensitivity 94.2%, specificity 91.8%. Consistent across ethnic groups (p=0.34). Inference time 0.8s per image.
Conclusion: AI-assisted DR screening achieves near-expert performance across diverse populations.
Author
Priya Sharma
(All India Institute of Medical Sciences (AIIMS), New Delhi)