Automated Detection of Mild Glaucoma Stage Using Grayscale Features of Fundus Images

摘要

Glaucoma is caused due to increase in the pressure within the eye. The symptoms are not always obvious; hence patients may seek treatment only when the condition progressed significantly. Early treatment will decrease the chances of vision loss, and so frequent screening is necessary. Therefore, a decision support system can reduce the effort for glaucoma detection manually. In this study, an automated detection system to classify normal, mild glaucoma and severe glaucoma conditions is developed. Various entropies mean, skewness, kurtosis and Gini features were extracted from the various Discrete Wavelet Transform (DWT) coefficients. Two-level of wavelet decomposition is performed to obtain 8 different coefficient matrices. Fourteen extracted features coupled with Support Vector Machine (SVM) classifier were able to yield an accuracy of 86.7%, sensitivity of 93.3% and specificity of 95.1% using ten-fold cross validation.