Automatic Brain MRI Mining Using Support Vector Machine and Decision Tree

This paper presents a texture based classification method using Support vector machine and decision tree for diagnosis of dementia. Support Vector Machine has been proved to be an effective classifier in several applications. In this work, a comparison of linear and non-linear kernels of SVM with BPN is investigated. Rules are extracted from a trained SVM which is compared with rules extracted from BPN and C5.0. OASIS dataset is utilized for training and testing of the classifiers. Wavelet based textural features from the brain MRI images are given as input feature vectors for classification. From the analysis it is found that SVM outperforms other classifiers. Rules extracted from the trained SVM improve the comprehensibility of the classifier.