Automatic Diagnosis and Classification of Glaucoma Using Hybrid Features and k-Nearest Neighbor

摘要

Glaucoma detection is the most challenging aspect in the medical image processing and analysis field. In this paper, glaucoma detection is assessed by employing High-Resolution Fundus (HRF) images and RIMONE databases. The proposed approach contains four major steps: segmentation, feature extraction, dimensionality reduction and classification. At first, segmentation was carried-out using correlation based template matching, it was a flexible high level machine learning technique to localize the object in complex template. Secondly, hybrid feature extraction (homogeneity and correlation) performed on the segmented optic disc image after performing Haar Discrete Wavelet Transform (HDWT) in order to achieve feature subsets. The respective feature values were given as the input for Principal Component Analysis (PCA) for the rejection of irrelevant and redundant features. After finding the optimal feature information, a multi-objective classifier: K-Nearest Neighbor (KNN) employed for classifying the normality and abnormality of glaucoma disease. In experimental analysis, the proposed approach distinguishes the normality and abnormality of glaucoma disease by means of specificity, sensitivity, accuracy. Positive Predictive Value (PPV), Negative Predictive Value (NPV) and Matthews Correlation Coefficient (MCC). The experimental outcome showed that the proposed methodology improved accuracy in glaucoma detection up to 3-30% compared to the existing methods: Neural Network (NN), Naive Bayes (NB), Random Forest (RF) and Support Vector Machine (SVM).