Classification of Pathological Magnetic Resonance Images of Brain Using Data Mining Techniques

Medical image analysis is a pioneer research domain due to the challenges posed by different kinds of images and the complexities in attaining the accurate prediction of abnormalities presence. Brain MRI classification into normal and abnormal has received increasing attention because of the high level of difficulty in handling those huge numbers of images. Recently, many computational techniques are widely employed to segregate the normal images from pathological. Thus, this study has attempted to analyse the capability of various supervised data mining techniques in classifying the brain MR images. Initially, the images are pre-processed and the volumetric features are extracted. Then, these are fed into feature selection techniques viz. Principal Component Analysis, Runs, Fisher filtering and ReliefF feature selection to determine relevant features. The selected features are utilised for the supervised data mining techniques viz. Naive Bayes, Support Vector Machine, Random Tree and C4.5 to identify the abnormal images of brain. Among them, SVM has achieved highest accuracy of 71.33% with the features extracted through ReliefF feature selection with Leave-One-Out cross validation. Random Tree achieved accuracy of 82% with Runs filtered features. The classification will aid the segmentation of brain tumor from large set of MRI slices by eliminating the normal slices. This greatly reduces the computational time and memory required for the process of segmentation.