Abstract:
Brain tumor segmentation involves detecting the presence of tumor in the available brain
scan image which is obtained from any neuroimaging machine. The main goal of any tumor
segmentation algorithm is to automatically segment the tumor tissues, in whole or
components (tumor core and edema), with maximum accuracy despite the nature of the tumor
present. This involves studying the differences between normal brain tissue in various
imaging modalities and choosing the most effective modality followed by applying a suitable
algorithm to segment the tumor. The diagnostic machines used these days provide complete
3-dimensional data of the subject’s brain. Hence a 3-dimensional algorithm is more suitable
and effective in detecting the tumor tissues. The proposed algorithm combines the concepts
of region split & merge and co-occurrence matrix to segment the tumor from the FLAIR
modality slices. It also uses region growing method to further improve the result obtained in
the first phase. The implementation involves the use of BRATS 2014 dataset. This is an
automatic method which detects the tumor of various shapes, sizes and locations with a
considerably high accuracy. The dice coefficient value to evaluate the test cases has an
average value of 0.89 in case of High Grade Tumors and 0.85 in case of Low Grade Tumors.
