Automatic Detection of Glaucoma in Fundus Images through Image Features

Abstract

The techniques of image processing and data mining find high applications in the field of medicine. Ophthalmologists analyze the fundus images of retina for finding the presence or absence of retinal diseases viz. Glaucoma, Diabetic Retinopathy etc. Glaucoma is the leading cause for blindness. Early detection of Glaucoma helps in providing necessary treatment. Computational techniques are sought for automatic detection of the disease. Researchers detect Glaucoma either though segmentation of optic disc and disease related structures and/or mining the extracted features. In this paper, Glaucoma is automatically detected through retinal image analysis and data mining techniques. The proposed work detect Glaucoma through color channel extraction, noise removal, contrast enhancement, overall image features viz., Statistical, Grey Level Co-occurrence Matrix (GLCM), Histogram based features extraction and classification. Retinal fundus images from High Resolution Fundus (HRF) Image Database are used for training the classifiers. Weka 3.7.6, an open source data mining tool, is used for implementation of this research. Various classification algorithms are attempted and comparison on results of outcome of these classifiers is given in this paper. It is observed that, among the collection of classification procedures, J48Graft classification algorithm provides the best performance achieving an accuracy of 100% with train-test of 70-30%. It is also noticeable that the classifier yields 86.67% with cross validation of folds 2, 3, 10 and 30.