A meta-analysis on diabetic retinopathy and deep learning applications

Diabetic retinopathy is one of the negative effects of diabetes on the eye. Early diagnosis of this disease, which can progress to blindness, is very important in this sense. There are many studies that detect and classify diabetic retinopathy, especially Machine Learning and Deep Learning methods. It is known that Deep Learning has been used more and more on disease detection and classification in recent years. There are three important reasons why deep learning is more successful in disease detection than methods such as image processing or machine learning. The first of these is that it achieves higher accuracies. Secondly, there is no need to develop an algorithm for each disease, that is, the algorithm learns the disease itself. Thirdly, faster results can be achieved with GPU (Graphics Processing Unit) support. For these reasons, in this study, articles written between 2015 and 2022 on the classification of diabetic retinopathy with deep learning were examined, and meta and statistical analysis was performed. Considering the work in the last two years the combined SEN value is 0.97 [95% CI, 0.92, 0.98], and the SPE value is 0.99 [95% CI, 0.98, 1.00]. The results obtained show how effective and necessary deep learning is in the early diagnosis of diabetic retinopathy.