3D 의료데이터를 이용한 가상혈관경 시스템 개발에 관한 연구

Abstract

The accurate diagnosis of vascular diseases, including coronary artery disease and cerebrovascular disease, requires complicated procedures that usually involve a great deal of time and cost. In an attempt to address this issue, virtual endoscopy and angioscopy based on the processing and utilization of 3D medical data have been researched and developed as one of tools required for noninvasive diagnosis in medical fields. In this dissertation, we propose three methods that can be used as key components of a virtual angioscopy system, each of which makes good use of 3D blood vessel models generated from medical images such as Computed Tomography (CT) and Magnetic Resonance Imaging (MRI). The methods include those for generating 3D curve-skeletons of human blood vessels, for determining navigation paths and operating a virtual camera to explore inside the 3D blood vessel models, and for detecting the suspicious region of vascular diseases. For the generation of the 3D models of blood vessels, we extract specific regions of interest from the medical images of human organs, express the regions as a voxel model, and construct a 3D mesh model by applying the marching cube algorithm to the voxel model. For the generation of the 3D curve-skeletons of blood vessels, we take two main steps: the generation of initial skeleton polygons and the refinement of the skeleton polygons. In the generation of the initial skeleton polygons, we obtain a set of thinned voxels by applying a voxel thinning algorithm to the voxel model. We then identify the vascular structure by detecting the junction and end voxels and dividing the thinned voxels into a set of branches. Next, we generate the initial skeleton polygons by approximating the voxels in each branch as a polygon. When refining the skeleton polygons, we move the points of all the branches closer to the centerlines of the 3D blood vessel model. For each branch point and end point, we compute its normal contour by intersecting the 3D mesh model and a normal plane at the branch or end point concerned, and then move the point to the inner center of the normal contour. For each junction, we modify its location using the actual branch points to which it is most adjacent. For navigation path determination and virtual camera operation, we make use of normal contours and 3D curve-skeletons. For a navigation zone specified by two points on the 3D curve-skeletons, the shortest path between the two points is computed and the positions of a virtual camera are estimated in the navigation zone. Then the positions are interpolated to ensure smooth movement of the camera along the path. In addition to keyboard and mouse input, intuitive hand gestures (determined by Leap Motion Software Development Kit (SDK)) are used as a user interface for virtual navigation of the 3D blood vessels. For the detection of the suspicious regions of vascular disease, we divide the 3D curve-skeletons into a set of branches, and compute the areas of normal contours of nodes located in each branch. By taking into account of the average area, the maximum and minimum areas, and the area difference between the adjacent normal contours, we detect and visualize the suspicious regions. We have implemented and developed a virtual angioscopy system based on the proposed methods, and tested it using several actual data sets of human blood vessels with artificially created data sets. We expect that this virtual angioscopy system can complement the weaknesses of conventional virtual angioscopy systems, and that it can be used as one of useful tools that are required to replace current invasive vascular endoscopy systems.