Development Of Medical Image Segmentation Algorithms Using Active Contour Models

Abstract

Medical image segmentation plays a very significant role in clinical and also in the field newlineof diagnostic. The process of segmentation of medical images is used for the detection newlineof region of interest from given images for the analysis of various diseases. It also play newlinea significant role in image guided surgery which requires precision in positioning of newlinetumor and calculation of size. In clinical diagnostics, brain tumor detection is the very newlinechallenging issue due to the complex nature of the human brain. The images obtained newlinefrom MRI scanners may not give sufficient information to doctors who is diagnosing newlinevarious brain anomalies. Therefore, developing an effective segmentation technique is newlinethe main goal of the proposed research work. In the first part of the research work, the newlinemultiresolution approach with Gradient Vector Flow (GVF) is proposed to segment newlinebrain tumour accurately from MRI data. In this method, pre-processed image is newlinerepresented as pyramid and every layer of pyramid is partitioned into different regions. newlineThese partitioned regions are then extended to higher level layer of resolution by the newlineapplications of DWT. Modulus values of these regions which are calculated by wavelet newlinetransform act as contours. The location of the tumor boundary and the corresponding newlinesegmentation is carried out by applying GVF model. The outcomes of this method newlineshows improvements in capture range of the contour towards the tumor boundary with newlinehigh accuracy of segmentation and also less computational complexity with respect to newlinethe available techniques. Due to large quantity of medical data, segmenting a brain newlinetumor is very complex tasks in the analysis of medical images. In second part of the newlineproposed work, hybrid model which incorporates K-means clustering algorithm and newlineLevel sets is applied to segment the tumor from MR images of the brain. The features newlineof non-cancerous and cancerous tumours are classified using Power LBP-Operator NB newlineclassifier. newline newlineThe standard database is used to collect the MR images for the experiment purpose. newlineVarious performance metrics were considered to evaluate proposed hybrid model. The newlineresults obtained from this method are encouraging. Generally, in medical images newlinetumours are having weak edges and inconsistent boundaries and traditional newlinesegmentation algorithms are incapable to face such situations. Incorporation of newlineContourlet transforms in medical image enhancement have shown prominent newlineimprovements for the MR images having very weak and inconsistent edges around the newlinetumor boundaries. In fact, Contourlet transforms are often applied for image newlinerepresentation and classification because these transforms represents images at the split newlineresolutions with different scale spaces. By applying transforms. MR images can be newlineanalysed at different levels of approximations. The third part of the proposed research newlineuses a novel methodology consisting of Contourlet transforms and deformable active newlinecontour models. This technique is based on the energy minimization of the active newlinecontour model by representing the images using Contourlet transform. The proposed newlinemethod performance is evaluated with the various performance metrics. The newlinecomparative analysis shows the improvements in the experimental results as compared newlineto the existing segmentation methods. newline