基于B样条自由形变三维医学图像非刚性配准研究

B-spline Free-form Deformation Based 3-D Non-rigid Registration of Medical Images

医学超声成像具有成本低、实时成像等优势,基于超声的多模态配准在临床诊断、病情监测、外科手术等应用上具有较大的意义。三维医学图像能够清晰地显示病变大小、形态,提供相对完整的人体组织的三维结构信息。本文采用基于B样条自由形变模型的非刚性配准方法对三维超声图像和计算机断层扫描图像(Computed Tomography,CT)进行配准,利用薄板样条能量约束项解决三维图像配准过程中的图像交叉与重叠问题。此外,使用仿体数据以及临床数据验证算法性能,通过感兴趣区域的相对重叠率,互信息值和程序运行时间这三个指标对算法精度和速度进行评价。其中Demons方法平均耗时1896s,本文改进算法平均耗时195s,运行效率提高8.7倍;算法改进前后的感兴趣区域重叠率分别是89.58%和91.35%,精度提高2.0%。实验结果表明此算法能够对超声和CT图像进行配准并获得较好的结果。

Ultrasound imaging has been widely used because of its inherent advantages compared with others imaging modalities.It plays a positive role in clinical diagnosis,monitoring and surgery.Three-dimensional medical images could clearly show the size,shape of the lesion and provide relatively complete three-dimensional structure information of human tissue.In this paper,a non-rigid registration based on B-spline free form deformation for three-dimensional ultrasound （US） and Computed Tomography Compared （CT） was studied with thin-plate spline energy constraint solving image cross and overlap.In addition,we took relative overlap,mutual information and computation time as evaluation criteria to evaluate the performance of nonrigid registration algorithm for breast phantom images and clinical images.The demons method took an average of 1896 seconds,the improved algorithm took an average of 195 seconds,the efficiency increased by 8.7 times.The relative overlap was 89.58 percent of the original algorithm and the relative overlap was 91.35 percent of the improved algorithm,the accuracy was increased by 2.0 percent.The experiment results showed that the algorithm could achieve better registration results.