Artificial Intelligence Using Deep Learning to Screen for Referable and Vision-Threatening Diabetic Retinopathy in Africa

Background: Radical measures are required to identify and reduce blindness due to diabetes to achieve Sustainable Development Goals by 2030. Therefore, we evaluated the accuracy of artificial intelligence (AI) using deep learning in a population based diabetic retinopathy (DR) screening program in Zambia, a lower middle-income country, ranked 159th in terms of gross domestic product.   Methods: A total of 4504 retinal fundus images from 3093 eyes of 1574 Zambians with diabetes were prospectively recruited. Referable DR was defined as moderate non-proliferative DR (NPDR) or worse, diabetic macular edema (DME) and ungradable images. Vision-threatening DR (VTDR) included severe NPDR and proliferative DR. We calculated the area under curve (AUC), sensitivity and specificity for referable DR, and sensitivities of VTDR and DME, using an Ensemble convolutional neural network compared to the grading by retinal specialists. A multi-variate analysis was performed for systemic risk factors and referable DR between AI and human graders. Findings: The prevalence of referable DR, VTDR and DME was 22·5%, 5·5% and 8·1% respectively. The AUC of the AI system for referable DR was 0·973, with corresponding sensitivity of 92·25% and specificity of 89·04%. VTDR sensitivity was 99·42% and DME sensitivity was 97·19%. AI model and human graders demonstrated similar outcomes in referable DR prevalence detection and systemic risk factors associations. Both AI model and human graders identified longer duration of diabetes, higher level of HbA1c, and increased systolic blood pressure as risk factors associated with referable DR. Interpretation: An AI system shows clinically acceptable performance in detecting referable DR, VTDR, and DME in population-based DR screening, demonstrating the potential application and adoption of such AI technology in the under-resourced African population to reduce the incidence of preventable blindness. Funding Statement: This project received funding from National Medical Research Council (NMRC) Health Service Research Grant, Ministry of Health (MOH), Singapore (National Health Innovation Center, Innovation to Develop Grant (NHIC-I2D-1409022); SingHealth Foundation Research Grant (SHF/FG648S/2015), and the Tanoto Foundation. The Singapore Diabetic Retinopathy Program (SiDRP) received funding from the MOH, Singapore (grants AIC/RPDD/SIDRP/SERI/FY2013/0018 & AIC/HPD/FY2016/0912). Declaration of Interests: Drs Ting, Lim, Lee, Hsu and Wong are co-inventors of a patent on the deep learning system in this paper; potential conflicts of interests are managed according to institutional policies of the Singapore Health System (SingHealth) and the National University of Singapore (NUS). Ethics Approval Statement: Lewis et al. DR screening program was previously approved by Topical Diseases Research Centre (TDRC), Ndola, Republic of Zambia. Subsequently, this specific study was also approved by the Centralized Institutional Review Board (IRB) of SingHealth, Singapore (protocol number SHF/FG648S/2015) and conducted in accordance with the Declaration of Helsinki.