Jun 2019–Zambia, a lower-middle income country, is ranked 159th by International Monetary Fund in terms of gross domestic product per capita in 2018. Like most countries which are resource constrained, most of its available healthcare capabilities are directed at tackling communicable diseases. Zambia has only 3 ophthalmologists per million population, compared to 80 per million for a developed country.

In collaboration with a Zambian team, a validation exercise of SELENA+, an artificial intelligence system to screen for retinal diseases developed jointly by Singapore Eye Research Institute (SERI) and National University Singapore – School of Computing (NUS-SOC) was conducted in Copperbelt province of Zambia. A sample size of 4,504 fundus images taken from 1,574 Zambians was used for this study.

This study sought to gauge the SELENA+ grading accuracy within a selected African population. The validation exercise showed similar outcomes for both SELENA+ and trained human graders(as shown in the table below) from Singapore Ocular Reading Centre (SORC).

AUC (95% CI) Sensitivity (95% CI) Specificity (95% CI)
Referable Diabetic Retinopathy 0·973
Vision Threatening Diabetic Retinopathy 0·934
Diabetic Macular Oedema 0·942

* The model was designed to identify referable diabetic retinopathy, hence specificities are not reported for vision-threatening diabetic retinopathy and diabetic macular oedema.

From this, we can interpret that our AI system shows clinically acceptable performance in detecting diabetic retinopathy, vision-threatening diabetic retinopathy and diabetic macular odema in population-based diabetic retinopathy screening. This shows the potential application and adoption of such AI technology in an under-resourced African population to reduce the incidence of preventable blindness, even when the model is trained in a different population.

This article was originally published on THE LANCET on 30 JAN 2019 and can be read at

Please direct all queries related to this article to Marketing at info@eyris.io