Efficacy of artificial intelligence in detecting diabetic retinopathy from retinal fundus images. A systematic review

Abstract

Introduction Diabetic retinopathy (DR) is a complication of diabetes mellitus, which damages the blood vessels of retina. Early diagnosis and treatment are important for proper management and reduced risk of vision loss [1]. Artificial intelligence (AI) based algorithms have been used to detect DR [2]. Aim The purpose of this review is to evaluate scientific literature about the efficacy of AI in detecting diabetic retinopathy from retinal fundus images. Methods Systematic literature review was carried out following PRISMA guidelines. “PubMed” and “ScienceDirect” databases were used for the search of scientific literature. Keywords and combinations of possible synonyms used for search were selected using the Medical Subject Headings dictionary. Both databases were last searched on 15th of February 2022. Articles were selected according to inclusion and exclusion criteria, developed using PICO method. Eligible studies met the following inclusion criteria: used artificial intelligence to detect DR and evaluated the accuracy of it by sensitivity (Se), specificity (Sp) and AUROC (area under the receiver operating characteristic curve), the index test was compared with ophthalmologists’ diagnosis as a ground truth and the diagnosis was made solely based on retinal images captured by fundus photography. For inclusion the publication had to be published in the last 10 years, written in English and have open access to full text. Case reports, literature reviews, systematic reviews and meta-analyses were excluded from thisreview. Two reviewers independently screened the titles and the abstracts of the citations from the literature search. Risk of bias was evaluated according to QUADAS-2 criteria. […].