Integration of Local Features for Brain Tumour Segmentation

To visualize the brain anatomy and to understand its functioning in non-invasive manner, magnetic resonance imaging (MRI) is one of the widely used imaging modality. Computer aided diagnosis (CAD) of the brain is a challenging and complicated task. Among various tasks, brain tumour segmentation plays a vital role in the efficiency of the CAD systems. Accurate detection and localization of brain tumour is an open research area because of its ill-posed problems. In this paper, integration of pixel local neighbourhoods (PLN) and local binary patterns (LBP) is proposed to extract the robust local information followed by self-organizing map (SOM) to accurately segment the brain tumour from brain MRI scan. DICE index and Jaccard index are used to evaluate the robustness of the proposed approach. Further, performance of proposed local feature integration is compared with existing state-of-the-art feature descriptor and clustering algorithms. Publicly available BRATS 2015 database is used to evaluate the performance of proposed and existing algorithms for brain tumour segmentation. Experimental analysis shows that the PLN+LBP followed by SOM outperforms other compared algorithms.