Predicting Diabetic Retinopathy using AI Models: An Experimental Study
Abstract
Diabetic retinopathy, a complication of diabetes that affects the eyes, can cause damage to the blood vessels in the retina. Artificial intelligence (AI) techniques play a crucial role in computer-aided diagnosis and have proven successful in identifying various diseases. This study aims to predict diabetic retinopathy by implementing feature extraction techniques to identify relevant factors. The dataset used in this study is sourced from the UCI Machine Learning Repository. Three machine learning algorithms, namely Support Vector Machine (SVM), Multilayer Perceptron (MLP) and Naïve Bayes classifiers, are employed to analyze the dataset and determine the most effective performance and accuracy. Among these classifiers, the SVM algorithm demonstrates the highest performance with an accuracy of 96.56%.
Keywords
Download Options
Introduction
Diabetic retinopathy is a complication of diabetes characterized by the damaging effects of elevated blood glucose on the retina, the surface of the eye. If left undetected and untreated, it can lead to vision loss and is a leading cause of new-onset blindness in adults. Diabetes affects the tiny blood vessels in various organs, including the retina [1]. Diabetic retinopathy is a result of diabetes mellitus and can lead to the formation of scars on the back of the eye. In severe cases, these scars can cause the retina to detach, a condition known as retinal traction detachment. Fortunately, there are several treatment options available that can help prevent or slow down the progression of the disease. Early detection and regular follow-up with a healthcare professional can significantly reduce the risk of blindness, up to 95%. The duration of diabetes is directly related to the likelihood of developing diabetic retinopathy, but maintaining well-controlled blood glucose levels can effectively delay its progression [3][4].
In recent literature, various AI algorithms have been employed for the detection of diabetic retinopathy. Diabetes mellitus is the leading cause of blindness among a significant age group in Western countries and its prevalence is also increasing in developing nations. Individuals with diabetes are at a significantly higher risk of developing blindness compared to those without diabetes. Moderate diabetic retinopathy and clinically significant macular edema can result in severe vision loss. Early detection through regular screening is crucial as it can be effectively treated in its initial stages. However, the cost and manual effort involved in screening are significant, making automated screening highly desirable. In diabetic retinopathy, the blood vessels that nourish the retina start leaking fluid and blood, leading to characteristic visual features such as microaneurysms, hemorrhages, hard exudates, cotton wool spots, and vein occlusion [9].
Conclusion
Based on the experimental results, it can be observed that all three machine learning algorithms achieved high accuracy in predicting diabetic retinopathy. The Support Vector Machine algorithm showed the highest overall performance, with an accuracy of 96.56%. The Multilayer Perceptron algorithm also performed well, with an accuracy of 95.86%. Naïve Bayes demonstrated good performance, albeit slightly lower than the other two algorithms, with an accuracy of 93.47%.
These results indicate that machine learning models have the potential to effectively predict diabetic retinopathy using the Diabetes Retinopathy Debrecen dataset. The high accuracy, precision, and recall values achieved by the algorithms highlight their ability to accurately classify instances of diabetic retinopathy. The findings of this study contribute to the development of AI-based models for the early detection and diagnosis of diabetic retinopathy, which can aid in timely interventions and prevent vision loss in patients with diabetes. Further research and refinement of these algorithms can potentially improve their performance and expand their applications in the field of diabetic retinopathy prediction.