An Ensemble Based Approach for Detection and Classification of Tumor in Human Brain

Abstract

A collection of cells which are abnormal in brain leads to brain tumor. Brain tumors can be malignant or benign. The most common type of brain tumor in humans is Glioma. It contributes to about 30% to brain and 80% to central nervous system of all tumors that are malignant. Magnetic Resonance Imaging (MRI) is the most sought tool for analysis of brain tumor. Traditional approach of detecting brain tumors does not provide accurate results, so we adopt an ensemble approach to improve the accuracy using proven training algorithms. In medical analysis automation of classifying these MRI accurately and interpreting the data plays a very important role. In our work carried out, to classify the MR images into normal and abnormal classes we have presented a hybrid method. In the proposed technique, all the possible feature to the minute level is extracted from the pre-processed image, after which the dimensions are reduced for the image. These reduced features are finally fed to kernel support vector machines on which k-fold cross validation is applied to enhance the generalization constant.