Macular edema in retinal fundus images by a computational algorithm

Background: Diabetes is a metabolic disease highly prevalent in our country that ends in disabling complications such as diabetic retinopathy and macular edema. As a high-impact socioeconomic issue, it is important to look for an early diagnostic test that also allows us to introduce a telemedicine service for the population with little access to specialized health services.

Objective: To describe the performance in discrimination of macular edema of a feature detection algorithm on retinal fundus images from diabetic patients.

Material and methods: We use a database of 266 retinal fundus images of diabetic patients and were classified in Macular Edema or Without Macular Edema by ophthalmologists' assessment and we test if the algorithm was capable of establish the presence or not of macular edema.

Results: We made 3 tests in which the standards of sensitivity, specificity and efficiency of the algorithm were increasing according to the amount of retinal fundus images in the training phase, reaching specificity values of 100%, sensitivity 84% and efficiency 91.30%.

Conclusions: Our study lays the foundation of a reliable screening method due to its high specificity value and allows not only a binary reply in the presence or not of macular edema but the clinical and topographic classification favoring the onset of treatment.