BRAIN TUMOR SEGMENTATION USING AUTOMATED REGION GROWING APPROACH

Abstract

The main objective of this dissertation work is to segment the brain pathological tissues (Tumor, Edema and Narcotic core) and to visualize it in 3D for their better understanding. For the same we have processed a novel approach which is a combination of thresholding based and region grow based algorithms. For the complete system, FLAIR and T2 modalities of MRI has been used due to their ability to detect the high contrast as well as low contrast lesions with great accuracy. In this approach, first the tumor is segmented from an image which is a combination of FLAIR and T2 image using a threshold value, selected automatically based on the intensity values of the tumor and normal tissues present in 3D MR images. Then on the basis of this output the tumor part is extracted from the actual brain's 3D MRI by selecting the largest connected volume from the. thresholding output. To correctly detect tumor 26 connected neighbors are used. The second step calculates a set of four seed points from the tumor portion extracted by the first threshold method in order to refine the tumor shape as the threshold output is not up to the mark. A 3D region grow algorithm is used for this. This algorithm is applied to both images one by one, the image which is the combination of FLAIR and T2 and FLAIR image as they have both the tumor and non-tumor information which can be visualizes clearly. The method is evaluated using a publically available BRAT dataset of 80 different patients having Gliomas tumors with great challenges. The accuracy in terms of detection is reached to 97.5% which is best compared to other existing methods in given time frame. The algorithm takes 4-5 minutes for generating the final. output. Other than segmentation this dissertation also aims to visualize the tumor in 3 dimensional volumes along with actual brain volume so that size and location of tumor can be realized. In order to identify tumor and brain, they are colored different. In order to make 3D visualization simple and easy ITK and VTK is used. Since the accuracy of detection is considerable, hence the proposed system can be used in hospitals and various healthcare centers for monitoring and for surgery operation. In medical studies the system can also play a good role for visualizing the concepts and better understandings.