基于体素互信息的多模式和多时相图像中股骨近端配准

Registration for Multi-Modal and Multi-Phase Images of Proximal Femur Based on Voxel Mutual Information

在股骨近端骨质疏松进程以及股骨头坏死状况评估方法中,图像分析是常用的工具,通过不同时相以及不同模式的多组影像可以对病人病情进行更全面的综合评估。然而,在综合评估过程中,由于病人多次在不同系统中成像,体位的差异使不同图像组之间的解剖点位置无法一一对应,因此分析之前需要将多组图像对齐,才能观察同一感兴趣区在不同模式或不同时间骨组织状况的差异。针对这个问题,设计一种多模式、多时相图像配准的解决方案,通过图像的前处理、双阈值分类并结合贝叶斯分类的股骨分割得到股骨体素,然后通过基于归一化互信息的图像配准获得各组图像中股骨之间的三维空间刚性变换矩阵,其中CT与MR图像的配准误差在4 mm以下,CT与CT图像的配准误差在2 mm以下。利用矩阵传递关系,以CT-CT多时相的配准矩阵为基础,可获取任何两组图像间的变换矩阵。在此基础上,再进行任意两组图像的融合、点对点的分析以及骨质状况和血供状况的定量评估。通过该方案,可以对多时相、多模式图像分析中相同感兴趣的区域进行对比。

Image analysis is a common method for evaluation of proximal femur osteoporosis and femoral head necrosis. This method evaluates patient＇s condition based on images of proximal femur with different modes and phases. However,because the images are generated in different systems and the positions of the patient relative to systems are different,the positions of anatomical points in different images are not one-to-one consistent. We need to align points of images with different modes and phases before we can research on interesting area. To solve the problem,we proposed a solution to get spatial rigid transformation through image pre-processing,voxels segmentation of femur based on dual-threshold combined with Bayes decision rule, and femurs registration based on normalized mutual information. The error of CT-MR registration and CT-CT registration were below 4 mm and 2 mm respectively. Using the matrix transfer relationship,rigid transformation between any two images based on multi-phase CT-CT registration matrix was obtained. On this basis,image fusion of any two images,point-to-point analysis and quantitative evaluation of osseous and blood supply condition were processed. The interesting area in images with different modes and phases through the solution was compared.