计算机辅助识别下虚拟内窥镜系统研究

【作者】 姚德民；

【导师】 宋志坚；

【作者基本信息】 复旦大学 ， 生物医学工程， 2007， 博士

【摘要】 虚拟内窥镜技术(Virtual Endoscopy-VE)是虚拟现实技术在现代医学中的应用。它类似光纤维内窥镜,是利用病人的医学影像数据,结合计算机图形学的方法重建和生成一组反映空腔器官内部表面形态的灰度或彩色图像。它具有光纤维内窥镜所无法替代的优点,如:安全、无创,适合于体弱者、婴幼儿、空腔器官重度狭窄者等,被视为光纤维内窥镜的有效补充。因此虚拟内窥镜的研究具有很高的临床应用价值。论文首先对构成虚拟内窥镜系统的关键技术进行了研究,并对开发的系统进行了临床验证。在此基础上研究了虚拟内窥镜新的发展方向:虚拟内窥镜的展开效果和计算机辅助识别的虚拟内窥镜系统虚拟内窥镜技术的两个关键性问题:自动寻找中心线的方法和交互式虚拟内窥镜成像方面,我们针对虚拟内窥镜的中心线自动提取问题,提出了一种利用距离尺度和最大代价生成树来提取结肠中心线的方法。该方法不但能提取全局最优的三维管腔的中心路径,而且还能处理多分支结构的对象。经过改进后的加速算法,可以迅速,自动的完成整个路径规划。对虚拟内窥镜的交互式成像,提出了一个易于操控和计算简便的坐标变换,并利用距离变换后的数据场进行体绘制的加速,结合并行计算,可以达到实时交互的效果。临床验证了系统的有效性。在虚拟内窥镜展开效果的研究方面,主要介绍了各种虚拟内窥镜对空腔器官展开的方法。我们对虚拟内窥镜展开效果的研究结果是:展开效果对临床的应用而言,始终是一个三维影像变换为二维的效果,因此需要对变形进行处理,虽然拓扑的方式是确切意义上的三维形变,但对临床的意义并不如视觉矫正的效果好,因为展开效果的目的是为了快速观察肠腔等器官的内表面,医师所希望的仍然是人体三维状态下的病灶形态,而非在经过器官变形以后的状态;从另一方面来说,展开效果对因病理因素导致的管腔狭窄的观察效果由于平面化而不如三维效果清晰明了,因此在临床虚拟内窥镜观察过程中,这种空腔展开效果始终无法代替虚拟内窥镜的观察,其临床意义仅限于快速浏览,处于从属地位。相对于计算机辅助识别,虚拟内窥镜展开效果的临床实用价值偏低。但作为一种虚拟内窥镜观察的辅助手段,展开效果仍然具有研究价值。在虚拟内窥镜的计算机辅助识别(CAD)方面。我们研究了已有的虚拟内窥镜计算机辅助识别方法,主要是结肠息肉的计算机辅助识别的方法。我们从着眼于开发适应面广泛的计算机辅助识别虚拟内窥镜系统出发,不拘泥于设计识别单一的CAD系统,采用了模板匹配的方式来研发整个CAD系统,模板匹配的核心采用ICP配准,模板采集的方式直接来自虚拟内窥镜的实际观察,整个识别过程为遍历新病例的空腔内壁,对空腔内壁的点,逐一作为中心向四周沿内壁搜集属于内壁的体素点组成判别区域的点集来进行识别。实验证明其能够进行息肉这样的病灶的自动检测,具有进一步发展的潜力。最后对虚拟内窥镜研究未来的发展方向做了展望。更多还原

【Abstract】 Virtual Endoscopy is an application of virtual reality in modern medicine. It simulates fibre endoscopy by representing the inner surface of hollow organ employing 3D reconstructed CT or MRI data which incorporates the research about Computer Graphics. It has advantage like safety, non-invasion, adapt to vulnerable, infant and extremely narrowed organs which fibre endoscopy not compare to. The complementarity of fibre endoscopy and Virtual Endoscopy is obviously. So, the research of Virtual Endoscopy has high value in clinical diagnosis.The first part in this thesis is the development of the key technology in the VE system, and so is the clinic trial of the VE System. The second and third parts is the research of new direct of the VE: The unfold of the inner surface and the Computer-Aided Detection (CAD) for Virtual Endoscopy.The two key technology of the VE System are: The Auto-extracted centerline and the interaction volume rendering of the VE. For the first technology, we put forward a new method which making use of the distance trasform and maximum-cost spanning tree Algorithm to extract the centerline .This method not only can pick-up the best fly-through path of the 3-D organ, but also,it can deal with the multi-branch object. After impoving some aspect of this method, it can finish the layout of the path quickly and automatically. For the second technology, we put forward a new method of coordinate transformation, which is making use of the parallel processing for the DT field on the first technology to accelerate the image creation. So the interaction of the new VE system is real-time and easy to control, And our clinic trial has prove that.In the research on the inner surface of organ unfolding, we have tried many methods. Because the essence of this problem is a transformation from 3D to 2D, so how to deal with the distortion is the most important. The topological technology is the most correct way to finish this transformation, but the it is not better as the method of visual revise because the doctors want to see the images more like the natural shape in the organs of the patients, not as the flatten shape on a plane with some distortion. On other hand, The result of flatten surface can not suit those case of the narrow organ by pathologic factor. For these reason, the unfold technology can not replace the 3D effect to observe the pathological changes on the inner surface, the only reason to use the unfold technology is to fast look through the inner surface. So the unfold technology is not as important as the CAD technology of the VE, but it still has some meaning to be developed as an assistant way to find the pathological changes.In the research of the CAD, We have tried some classical methods, especially for the auto-detected of the colon polyp. But we want to develop a new way, which can be used into the auto-detected for more pathological changes. We have applied serval pathological templates to fit this condition. The template is picked up from some of clinic VE examine, and the key technology is ICP registration. The ICP Algorithm is used to picking up the new templates and recognizes the pathological changes. The scheme for Detection of Polyps is search all over the entire inner surface of the cavity organ (i.e. colon).For every voxel of the surface ,collect the neighbor voxels of it as a center and compare to the templates by ICP. The candidate region is chosen by the threshold. The clinic trial has proved that the new method can detect the colon polypus and other some pathological changes automatically. We trust that this method have much potential.The final part of this paper, we have indicated some expectation of the VE technology更多还原