Multifractal feature-based abnormal tissues segmentation in brain MRI using modified adaboost classifier

Abstract

Brain tumour segmentation is an important task in medical imaging. In this work, image features-based process is proposed to segment brain tumour in MRI images. For the segmentation of brain tumour, MRI brain images should be free from artefacts because it causes unwanted variation in the image and affects the performance of image processing techniques used for brain image analysis. The proposed system consists of three phases: preprocessing, feature extraction and segmentation. In preprocessing, the motion artefacts are corrected by spatial transformations. Texture features are extracted from the estimation of multifractal dimension using curvelet transform. Along with this feature, texton and intensity features are also considered. The fusions of all the features are fed to the modified adaboost classifier. BRATS 2013 dataset is used in this work along with its ground truth. The performance of the method is analysed in terms of sensitivity, specificity and accuracy. The proposed work gives higher accuracy on segmenting abnormal tissues compared with wavelet-based existing methods.