Classification of Eye Diseases using Machine Learning with Preprocessing

Abstract

The eye, as the sensory organ responsible for vision, is an essential part of our daily lives. However, various disorders can negatively impact eye health and visual acuity, potentially resulting in visual impairment. Machine learning holds immense potential in the diagnosis of eye diseases. This study investigates the utilization of machine learning models for the early detection of ocular diseases. After processing the retinal images with various image processing techniques, distinctive features are extracted using feature extraction algorithms such as histogram of oriented gradients (HOG), local binary patterns (LBP), and residual network-50 (ResNet-50). The obtained training data is applied to the classification algorithms like K-nearest neighbors (KNN), support vector machine (SVM), and extreme gradient boosting (XGBoost) to differentiate between normal eyes and those with diabetic retinopathy, cataracts, and glaucoma. These three different algorithms for classifying the eye diseases are compared according to key performancemetrics. The results indicate that the combination of XGBoost with ResNet-50 achieves the highest performance with 92% accuracy, followed by the combination of SVM with ResNet-50 at 90% accuracy. © 2024 IEEE.