Detection of diabetic retinopathy using artificial intelligence: an exploratory systematic review

Abstract

Diabetic retinopathy is a disease that can lead to vision loss and blindness in people with diabetes, so its early detection is important to prevent ocular complications. The aim of this study was to analyze the usefulness of artificial intelligence in the detection of diabetic retinopathy. For this purpose, an exploratory systematic review was performed, collecting 77 empirical articles from the Scopus, IEEE, ACM, SciELO and NIH databases. The results indicate that the most commonly used factors for the detection of diabetic retinopathy include changes in retinal vascularization, macular edema and microaneurysms. Among the most commonly applied algorithms for early detection are ResNet 101, CNN and IDx-DR. In addition, some artificial intelligence models are reported to have an accuracy ranging from 90% to 95%, although models with accuracies below 80% have also been identified. It is concluded that artificial intelligence, and in particular deep learning, has been shown to be effective in the early detection of diabetic retinopathy, facilitating timely treatment and improving clinical outcomes. However, ethical and legal concerns arise, such as privacy and security of patient data, liability in case of diagnostic errors, algorithmic bias, informed consent, and transparency in the use of artificial intelligence.