Nonnegative Matrix Factorization Methods for Brain Tumor Segmentation in Magnetic Resonance Images

Abstract

Automatic segmentation of objects in magnetic resonance images (MRI) is very challenging to provide reliable and automated computerized methods for clinical usage because this is a time-consuming process which requires more and more attention. MR imaging is the most usable technique of images to segment the tumor region from brain. By follow the novel framework of nonnegative matrix factorization (NNMF) along with fuzzy clustering and region growing, authors are able to do the tumor segmentation easily and accurately in this research work. By applying these methods, this paper represents the successful extraction of the whole tumor along with narcotic and edema from brain MR, T2, and FLAIR images. This proposed methodology is tested on the BRATS 2012 database with low-grade glioma (LGG) and high-grade glioma (HGG) brain tumor images. Moreover, along with NNMF, fuzzy c-means clustering and region growing, the algorithms of Dice, sensitivity, specificity, and Hausdorff are also evaluated to check the accuracy of the results.