Automatic classification of brain diseases in mr images using genetic algorithm and support vector machine

This study presents a method to improve the classification accuracy of brain disease that consist of Alzheimer's disease and the brain tumor in magnetic resonance (MR) images. For this purpose, 71 MR images that consist of 4 normal, 14 Alzheimer's disease, and 53 brain tumor images were acquired from 12 patients. A total of 42 features were extracted from MR images using first order statistics, gray level co-occurrence matrix, and Laws' texture energy measures. Then, the optimized feature set was selected by using genetic algorithm (GA) and support vector machine classified the brain MR images into normal, Alzheimer's disease, and brain tumor. GA method selected 10 different feature sets, and the classification accuracy of each feature set were compared to find the best set. Finally, the performance of the classification was evaluated using sensitivity, specificity, accuracy, and receiver operating characteristic curve. The results of this study showed that all evaluation parameters were improved for classification of brain disease through the application of GA selection for all classes. In particular, the specificity greatly increased from 87.8% to 95.0% in classification between Alzheimer's disease and brain tumor. In addition, the radial basis function kernel showed the highest classification accuracy of 96.2% among all kernel conditions examined. These experimental results demonstrated that the proposed method improve performance to classify the brain MR images.