Approximately three in ten diabetic patients develop diabetic retinopathy in the course of their lives. If this eye disease, caused by the excess sugars in the blood, is not treated early, they risk losing their sight. With the financial support of the Diabetes Liga Research Fund, a large-scale study by the UZ Leuven and VITO, and in collaboration with three Flemish hospitals, is investigating whether we can reliably diagnose diabetic retinopathy faster and more economically via artificial intelligence (AI).
There is a treatment that can inhibit the evolution of diabetic retinopathy and thus ensure that the patient retains his sight much longer. However, the condition must be discovered as soon as possible. Because diabetic retinopathy does not cause the patient any symptoms at an early stage, diabetic patients must have an annual check-up with their ophthalmologist. The examination is not difficult: an image is taken of the retina and the ophthalmologist can then see whether or not diabetic retinopathy is present and whether or not treatment needs to be started.
The examination and interpretation of retinal images by an ophthalmologist is expensive. The long waiting times at this specialist may in turn result in delayed care. The question is therefore whether this examination could not be more efficient and cost-effective. This is precisely why, at the request of the Diabetes Liga Research Fund set up by the King Baudouin Foundation, the E-CLAIR project has now been launched. The E-CLAIR project compares two new clinical methods that make use of artificial intelligence with the current way of working. 1,200 diabetic patients who come for consultations at ZNA Middelheim, the UZ Antwerpen and the AZ Turnhout are being examined in three ways. Either traditionally, directly by the ophthalmologist, or by taking retinal images that are only quoted by the computer, or by taking retinal images that are first analysed by the computer, after which they are also checked by a trained ophthalmologist in training. The results are finally checked by an ophthalmologist.
"In this way, we can evaluate the quality of the computer's grading using artificial intelligence in clinical practice," says VITO researcher Nele Gerrits. "We find that the AI models we use today are extremely strict. They are too quick to say that retinopathy is involved when it is not. A solution to this could be a manual check-up by an ophthalmologist in training. During this examination, all the images are then analysed at the UZ Leuven and divided into categories. This makes it possible to find out exactly how careful the AI model is".
This research will last until the end of 2022. "If the results of this study show that the computer analysis is sufficiently reliable, we will in future only be able to refer patients whose retina image shows injuries to the ophthalmologist for specialised examination at the annual check-up. In other words, the ophthalmologist will no longer have to carry out routine examinations, but will be able to concentrate entirely on those patients who really have an eye problem. It is precisely for this reason that we also want to include the cost of this examination, so that in the end, we also have a view on the cost-effectiveness of using artificial intelligence", concludes Nele Gerrits.
In this research project, AI models developed at VITO will be evaluated. These models have been trained on more than one million retinal photographs to detect signs of diabetic retinopathy. This completes the work of VITO and the software will soon be housed in the spin-off MONA under the supervision of Olivier Ménage.
About the Diabetes League Research Fund
In collaboration with the King Baudouin Foundation, Diabetes Liga recently set up the Diabetes Research Fund. Two groundbreaking scientific projects - the E-CLAIR project and the CRISTAL study in particular - can now count on support. The ambition is to be able to invest a total of 1 million euros in diabetes research from this fund by 2025.