Skip to navigation Skip to main content Skip to footer

Approved research

Retinal fundus photographs to predict diabetes: A potential solution for community-based diabetes screening in low- and middle-income countries

Principal Investigator: Mr Richard Haarburger
Approved Research ID: 57136
Approval date: April 30th 2020

Lay summary

This project aims at determining if and/or how well Diabetes can be predicted based on Retinal Fundus images using image-based machine learning. For this purpose we train a convolutional neural network on deducting if an individual has Diabetes given this individuals' Retinal Fundus images. Once the model is trained and shows to be reliable, it can be used to provide information on individuals' Diabetes status based solely on Retinal Fundus imagery. Conventional methods used for diagnosing Diabetes in middle and low income countries, i.e. random blood glucose as well as questionnaire-based methods perform poorly. If we can show that the Retinal Fundus image-based prediction of Diabetes works well, the next step would be to collect smartphone Retinal Fundus images and check whether our results can be replicated. Predicting Diabetes using Retinal Fundus imagery bears great potential with regard to facilitating and enabling Diabetes screening in middle and low income countries, where prevalence keeps rising.