Please use this identifier to cite or link to this item:
|Title:||Modified distance regularized level set method to segment hepatic tumor|
|Keywords:||Computed Tomography;Level Set Function;Distance Regularized Level Set Method;Dice Similarity Coefficient;Modified Distance Regularized Level Set Method;Two-Dimensional|
|Abstract:||Severity of liver disease assessment, surgical planning and follow-up of liver cancer treatment essentially require an accurate segmentation of tumor region. However, Intratumoral heterogeneity and minor variations in tumor tissue intensities with respect to surrounding liver tissues are the major challenges in computerized hepatic tumor segmentation and reconstruction of its three-dimensional (3D) volume using computed tomography (CT) images. Level set method is an efficient segmentation method to segment hepatic tumor with the advantage of its ability to (i) model complex shapes of tumor even no prior information about the topology available (ii) split and merge efficiently to represent tumor region. Conventional level set methods face the problem of irregularities during the evolution of level set function (LSF) which leads to distortion of stability and numerical errors. To overcome these problems, curve is stopped and reshaping of degraded LSF is done after fixed interval of time termed as reinitialization. But it is unpredictable that when and how it is to be applied on LSF.A number of modifications was done in recent years for better convergence and to handle the problem of reinitialization in conventional level set methods. One approach was introduced by Li et al. to eliminate the need of reinitialization by introducing double well potential term for regularization and they used edge-based energy functional to converge or stop the contour at the boundary of interfaces having distinct intensity characteristics. However it faces the problem of oversegmentation or boundary leakage during the evolution of LSF. In the present work, a new method has been proposed to significantly overcome the boundary problem encountered during implementation of upwind scheme inspite of central difference scheme used in DRLSE method in external energy function to stop the contour to converge at exact boundary of the tumor. Further, the present work proposed modifications in the template of upwind scheme to capture accurate definition of tumor boundary using higher number of neighborhood pixels. This way the proposed method avoids the problem of leakage during the segmentation of boundary. The results of modified DRLSE method clearly demonstrate that the proposed method outperforms the DRLSE method in terms of Dice similarity coefficient and Relative volume difference. Further, Tumor volumetry has become an essential surgical tool to know the extent of liver cancer and staging, guidance during surgery and follow-up the treatment. Thus, 3D volume is constructed using segmented two-dimensional (2D) slices and this structure can demonstrate the extent of tumor and its volume for further treatment and planning using CT image slices.|
|Appears in Collections:||Masters Theses@EIED|
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.