Date of Publication :20th June 2018

Abstract: Earlier in the diagnosis of a disease; better is the rate of recovery. As far as the pestilent disease like brain tumor is concerned, its early identification may lead to improve the rate of care and thereby benefitting the survival of a patient. Typically, brain tumor detection and analysis starts from the process of brain MRI segmentation. This segmentation partitions the potentially overlapping parts in the internal structure of the brain into brain tissues such as White Matter (WM), Grey Matter (GM) and Cerebro Spinal Fluid (CSF). In this paper, automated brain tumor detection has been proposed for detecting the presence/ absence of brain tumor from brain MR images. Relevant pre-processing is applied to input brain MR images. Firstly, the brain input image is zoned using superpixel zoning and brain tissues are being segmented using discriminative clustering. Secondly, feature extraction is done using level 2 2-D discrete wavelet transform to generate the matrix vectors. AdaBoost with random forests algorithm (ADBRF) is used as its base classifier to classify the given brain MR image into normal or abnormal. Simulation results are compared with the existing methods on BrainWeb brain MRI dataset and it is observed that the proposed scheme outperforms other state of the art methods.

Reference :