基于ROI及Clifford代数相对不变量的3D医学图像配准

发布时间：2018-03-29 17:02

本文选题：医学图像配准　切入点：Clifford代数　出处：《图学学报》2017年01期

【摘要】：在利用颅骨轮廓几何特征配准基础上,提出配准前颅部图像的感兴趣区(ROI)圈定,并在Clifford代数框架下给出一种全新的相对不变量构造方法。该方法以颅部ROI的轮廓数据作为配准点云集,根据颅骨刚体轮廓相似性特点,运用Clifford代数构造相对几何不变量的数学模型及计算模型,并计算配准几何运算需求的平移量和旋转算子,采用3D医学图像的相似性测度直接进行三维数据的配准。数据源及评估使用Brain Web数据库和美国Vanderbilt大学的"回顾性图像配准评估"项目数据。实验表明,新方法在颅部的ROI区域进行配准,能够精确的定位组织器官的三维位置,执行效率高,配准均值误差在2~4 mm内,达到亚像素级配准精度。
[Abstract]:On the basis of the geometric feature registration of the skull contour, the region of interest (ROI) delineation of the anterior cranial image is proposed. A new method of constructing relative invariants under the framework of Clifford algebra is presented, in which the contour data of cranial ROI are used as the matching point cluster, and according to the similarity of skull rigid body contour, The mathematical model and calculation model of relative geometric invariants are constructed by using Clifford algebra, and the translation and rotation operators of matching geometric operation requirements are calculated. The similarity measure of 3D medical image is used to register 3D data directly. The data source and evaluation use Brain Web database and "retrospective image registration evaluation" project data of Vanderbilt University in the United States. The experimental results show that, The new method can accurately locate the three-dimensional position of tissues and organs in the ROI region of the cranium. The efficiency of the new method is high. The mean error of registration is within 2mm and the accuracy of sub-pixel registration can be achieved.
【作者单位】：南通大学电气工程学院;
【基金】：国家自然科学基金项目(61273024,61305031) 江苏省自然科学基金项目(BY2016053-11) 江苏省“333”高层次人才培养工程项目(BRA2015366) 江苏省优势学科项目(PAPD)
【分类号】：R445;TP391.41

【相似文献】