IMPROVED AUTOMATED DETECTION OF DIABETIC RETINOPATHY ON A PUBLICLY AVAILABLE DATASET THROUGH DIFFERENT MACHINE LEARNING ALGORITHMS

Abstract

Diabetic retinopathy is a common complication of diabetes that can lead to vision loss if left untreated. Early detection and treatment are therefore essential to prevent irreversible visual impairment. In this study, we explore the use of machine learning algorithms for detecting diabetic retinopathy. Specifically, we compare the performance of five popular algorithms Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Decision Trees (DT), Random Forest (RF), and Gradient Boosting Classifier (GB) using hyperparameter tuning via Grid Search Cross-Validation.

Our experiments were conducted on a dataset of retinal images obtained from diabetic patients. The images were preprocessed to extract relevant features, including blood vessel density and lesion characteristics. These features were then combined with clinical patient information to train and test the models.

The results demonstrate that all five algorithms were capable of accurately detecting diabetic retinopathy from retinal images. However, the SVM algorithm outperformed the other four in terms of both accuracy and computational efficiency. Overall, our findings suggest that machine learning algorithms can serve as valuable tools for the early detection of diabetic retinopathy. Notably, the Random Forest algorithm also showed strong performance and may be particularly well suited for this task.