Skip to content

Take part

Home
Talks
AIiH 2025

DiabEye-Q: AI-driven longitudinal analysis of ophthalmoscopic images for early diabetes prediction in Qatari adults

Speaker: Dr. Zubair Shah
Event: International Conference on AI in Healthcare (AIiH 2025)
Date: 8–10 September 2025
Place: Cambridge, United Kingdom

Slide 1 of 27

Slide 2 of 27

Slide 3 of 27

Slide 4 of 27

Slide 5 of 27

Slide 6 of 27

Slide 7 of 27

Slide 8 of 27

Slide 9 of 27

Slide 10 of 27

Slide 11 of 27

Slide 12 of 27

Slide 13 of 27

Slide 14 of 27

Slide 15 of 27

Slide 16 of 27

Slide 17 of 27

Slide 18 of 27

Slide 19 of 27

Slide 20 of 27

Slide 21 of 27

Slide 22 of 27

Slide 23 of 27

Slide 24 of 27

Slide 25 of 27

Slide 26 of 27

Slide 27 of 27

1 / 27

Download slides (PDF)Read the paper

Summary

Can a photograph of the back of the eye, taken years earlier, tell us who will develop type 2 diabetes? We analysed longitudinal retinal images and clinical records from 2,043 Qatar Biobank participants.

Fractal features from retinal images, fused with clinical data in an XGBoost and Vision Transformer pipeline, predicted future diabetes with 88% accuracy and 93% AUROC. Performance differed by age group, and the analysis pointed to higher susceptibility in men and older adults.