Magnetic Resonance Imaging (MRI) is one of the prominent imaging techniques for assessment of brain tumor progression. Intensity inhomogeneity, partial volume effect (PVE) and diverse nature of tumor render a challenging task for automatic segmentation of brain tumors from MR images. Existing MRI brain tumor segmentation methods focus one or two of the above mentioned challenges. We aimed to present a framework for automatic brain tumor segmentation that effectively tackles all major challenges. In the proposed framework, first the intensity inhomogeneities in the MRI images are corrected using an Enhanced Homomorphic Unsharp Masking algorithm. Following intensity inhomogeneity correction, features are extracted. Finally, the extracted features are fused and clustered using Multiple Kernel FCM (MKFCM) clustering algorithm. The MKFCM clustering algorithm employed in the proposed framework overcomes the PVE and form more generalized clusters, thus proved to be effective for images with diverse tumor shape. To demonstrate the effectiveness of the proposed framework, it is compared with four other clustering algorithms using different validation measures.


Magnetic Resonance Imaging (MRI), Brain tumor, Segmentation, Fuzzy, Clustering, Kernels