Eye diseases classification based on hybrid feature extraction methods

Abstract

The field of image classification in medical diagnosis presents several challenges and problems that motivated our research and led us to propose a novel method. These challenges include the need for accurate and efficient diagnosis, extracting relevant features from medical images, and integrating different classification algorithms for improved performance. Considering these challenges, we aimed to develop a robust and effective approach to address these issues and enhance the accuracy of medical image classification. This research proposes a hybrid feature extraction method for eye disease classification using a combination of Local Binary Pattern (LBP), Gray-Level Co-occurrence Matrix (GLCM), and Texture Feature Coding Method (TFCM). The dataset used for evaluation is obtained from Kaggle and consists of retinal images representing various eye diseases. Pre-processing involves ROI extraction and image resizing. Features are extracted using LBP, GLCM, and TFCM, totaling 37 features. These features are combined into a single vector. The classification task is performed using k-Nearest Neighbors (kNN) and Support Vector Machine (SVM) classifiers, with performance analysis conducted using five metrics. The experimental results demonstrated the effectiveness of the hybrid feature extraction method in accurately classifying eye diseases. The SVM and kNN classifiers achieved high accuracy, with SVM achieving an accuracy of 0.9988 and kNN achieving an accuracy of 0.9955.