Brain Tumor Localization and Segmentation Based on Pixel-Based Thresholding with Morphological Operation

Abstract

Brain tumor localization and segmentation from MRI of the brain is a significant task in medical image processing. Diagnosis of brain tumors at early stages play a vital role in successful treatment and raise the survival percentage of the patients. Manual separation of brain tumors from huge quantity of MRI images is a challenging and time taking task. There is need for an automatic efficient technique for brain tumor localization and segmentation from MRI images of brain. Some years ago, improper filtration and segmentation techniques was used for brain tumor detection, which gives almost inaccurate detection of tumor in MRI images. The proposed technique is mainly based on the preprocessing step for de-noising input MRI, thresholding, and morphological operation and calculating performance parameters for validation. Firstly, anisotropic diffusion filter is applied for removal of noise because input MRI images are mostly noisy and inhomogeneous contrast. Secondly, MRI pre-processed brain image is segmented into binary using thresholding technique. Thirdly, the region-based morphological operation is used for separation of tumorous part from segmented image. At the end, Root mean square error (RMSE), peak signal to noise ratio (PSNR), tumorous area in pixels and centimeters, system similarity index measurement (SSIM), area under curve (AUC), accuracy, sensitivity and specificity are the parameters used for evaluation of the proposed methodology. Visual and parametric results of proposed method are compared with the existing literature.