融合SIFT高维特征的医学图像非刚性配准.DOC

中图法分类号：TP391 文献标识码：A 文章编号：1006-8961(2011) - - 文章索引信息: 融合SIFT特征的熵图估计医学图像非刚性配准本文提出一种配准算法。该算法首先k-最邻近图（kNNG）；最后，使用k-最邻近图来估计ααMI）。实验结果表明。医学图像配准k-最邻近图; α互信息 Entropic graph estimation integrated with SIFT features for medical image non-rigid registration ZHANG Shaomin1,2,3), ZHI Lijia1,2,3), ZHAO Dazhe1,2,3), LIN Shukuan1), ZHAO Hong1,3) 1) (College of Information Science and Engineering, Northeastern University, Shenyang, Liaoning, 110004) 2) (Key Laboratory of Medical Image Computing (Northeastern University), Ministry of Education, Shenyang, Liaoning, 110004) 3) (National Engineering Research Center of Digital Medical Imaging Equipment, Shenyang, Liaoning, 110004) Abstract: Accuracy is an important performance of the medical image registration. Pixel gray value is a widely used feature in image registration. However, pixel gray value comes from single source, and ignores the spatial information. In some cases, it will cause misalignment. To solve the problem, entropic graph estimation integrated with SIFT features is proposed as a medical image non-rigid registration algorithm. In the algorithm, mutual information based rigid registration is used to roughly register two images. Then the pixel gray value and the SIFT features are extracted to form k-nearest neighbor graph (kNNG), which is used to estimate α-mutual information (αMI). Comparison results of the images obtained from lung CT images and brain MRI images showed that the algorithm proposed can provide better accuracy than both the conventional rigid registration algorithm based on mutual information and the non-rigid registration algorithm based on entropic graph estimation and single pixel gray value. Keywords: medical image registration; SIFT (scale invariant feature transform) descriptor; k-nearest neighbor graph (kNNG); α-mutual information (αMI) 0 引 言 医学图像配准是指对不同时间、不同视场、不同模态的两幅或多幅图像进行空间几何变换，以使代表相同解剖结构的像素或体素在空间位置上能够对应起来[1]。许多重要的临床应用需要用非刚性变换来描述图像之间的空间关系。例如在腹部以及胸部脏器的中，内部器官和组织的位置、尺寸和形状由于不自主的生