3D Liver Segmentation and Reconstrucion from CT Images by Interactive-Cooperative Active Contour

Abstract

This thesis aims at developing software to segment livers from computed tomography images and to build up the three-dimension model for the segmented liver. Recently, active contour approaches are popularly used in computer vision and image processing to address object segmentation. However, active contours still remain many weaknesses such as being traps with local optimization, spikes at the boundaries detected by the snake approach, and the disadvantages of high computation costs with the level set functions. This project introduces a new way to enhance active contour models for the tasks of segmentation. We combine the active contour approach with snake (ACS) [1] and active contour approach proposed by Chan-Vese (ACCV) [2] to improve the efficiency of object segmentation and to overcome the drawbacks of each active contour model when they are performed independently. Our developed method is named Interactive-Cooperative Active Contour (ICAC). We tested our approach with a large number of real CT images to show that it is superior to the two conventional active contour models including ACCV and ACS. Besides, for the task of 3D reconstruction of the liver, we utilize Poisson surface reconstruction implemented in Meshlab with our own developed filters to remove the artifacts caused by the segmentation process. The reconstruction results are acceptable when compared with the ground truth data provided by experts. Finally, all developed functions of this thesis are integrated in a graphical user interface that is easy and convenient to use. Therefore, our developed software and methods can be applicable to help doctors in medical diagnosis and treatment process. Key words: Interactive-cooperative active contour (ICAC), active contour model with Snake, active contour without edge, computed tomography.