An artificial intelligence-based decision support model for diabetic retinopathy diagnosis

Diabetic Retinopathy (DR) is a disease commonly observed in diabetic patients, which, if left untreated, can lead to advanced vision loss or blindness. DR is found in approximately 30% of diabetic patients, making it one of the leading causes of vision loss. Most symptoms can be noticed during blood sugar monitoring and can be intervened with early treatment. Early diagnosis and treatment of the disease are crucial for preventing vision loss in diabetic patients. The use of traditional methods in the early diagnosis of DR, involving manual techniques, can lead to time loss or inconclusive results. Recently, like in many other fields, artificial intelligence approaches are being utilized in healthcare as well. AI-based models can make objective decisions and contribute to the diagnosis and treatment of many patients. In this study, a hybrid artificial intelligence model is proposed, which combines deep learning models for the early diagnosis and grading of DR along with machine learning methods. The images in the dataset used in the study were created using fundus imaging technique and graded between 1-5 according to the severity of the disease. The dataset was obtained from an open-access website and processed accordingly. In the first stage, the data underwent preprocessing to remove unnecessary parts from the original images. In the second stage, these data were trained by a deep learning model to extract features, and efficient features were highlighted by feature selection algorithms. The aim here is to save time for the model and contribute to its performance success. In the third stage, the data obtained from efficient features were trained using machine learning methods for classification, and the best results were achieved. Experimental analysis resulted in the proposed artificial intelligence model achieving 100% overall accuracy.