Xiong Guanglei, Xing Lei
Biomedical Informatics Program, Stanford University, Stanford, CA 94305, USA.
Inf Process Med Imaging. 2011;22:486-97. doi: 10.1007/978-3-642-22092-0_40.
Junction structures, as the natural anatomical markers, are useful to study the organ or tumor motion. However, detection and tracking of the junctions in four-dimensional (4D) images are challenging. The paper presents a novel framework to automate this task. Detection of their centers and sizes is first achieved by an analysis of local shape profiles on one segmented reference image. Junctions are then separately tracked by simultaneously using neighboring intensity features from all images. Defined by a closed B-spline space curve, the individual trajectory is assumed to be cyclic and obtained by maximizing the metric of combined correlation coefficients. Local extrema are suppressed by improving the initial conditions using random walks from pair-wise optimizations. Our approach has been applied to analyze the vessel junctions in five real 4D respiration-gated computed tomography (CT) image datasets with promising results. More than 500 junctions in the lung are detected with an average accuracy of greater than 85% and the mean error between the automated and the manual tracking is sub-voxel.
连接结构作为自然解剖标志物，对于研究器官或肿瘤运动很有用。然而，在四维（4D）图像中检测和跟踪连接结构具有挑战性。本文提出了一个新颖的框架来自动完成这项任务。首先通过对一幅分割后的参考图像上的局部形状轮廓进行分析来实现对其中心和大小的检测。然后通过同时使用来自所有图像的相邻强度特征来分别跟踪连接结构。由一条封闭的B样条空间曲线定义，单个轨迹被假定为循环的，并通过最大化组合相关系数的度量来获得。通过使用成对优化中的随机游走改善初始条件来抑制局部极值。我们的方法已应用于分析五个真实的4D呼吸门控计算机断层扫描（CT）图像数据集的血管连接结构，结果很有前景。在肺部检测到了500多个连接结构，平均准确率大于85%，自动跟踪与手动跟踪之间的平均误差在亚体素级别。
