Assessing robustness and quantifying uncertainty in level set segmentation of the bowel lumen

摘要: This thesis explores various methods to assess the robustness of an algorithm used determine severeness Crohn's disease, segmenting bowel lumen by minimizing a total variation based energy functional similar Local Gaussian Distribution Fitting functional. By comparing local solution this minimization problem obtained using gradient descent with global convex formulation optimization problem, conclusion is made that it definitely not searched for, which emphasizes strong dependence on its initial condition. The input parameters also tested slightly varying or artificially creating missing data and measuring influence output solution. Applications discussed in test specific mentioned comprehends justification measure less accurate, both cheaper faster, necessity centerline serve as condition, technique most important parts along centerline, have greatest