A robust semi-automatic approach for ROI segmentation in 3D CT images

In CT-based clinical applications, segmentation of regions of interest (ROIs) is a preliminary but vital step. The task is, however, quite challenging, especially for 3D objects, because suspicious ROIs are usually soft-tissue structures, which include a various organs and anatomical objects while sharing a small intensity dynamic range in CT images. Furthermore, the ROIs usually vary significantly in size, shape, and boundary conditions. Among considerable efforts contributed to addressing the problem, live wire, also known as intelligent scissors, has been recognized as an efficient and robust tool for dealing with a wide range of 2D ROIs. Such an approach provides full user control during the process while minimizing human interaction to optimally counterbalance automatic and manual approaches. In this work, we improve our previous live-wire-based segmentation of 3D objects and the experiment results show its efficiency and robustness.