基于贪婪策略的生物发光断层成像重建算法的对比研究

发布时间：2018-05-29 22:08

本文选题：生物发光断层成像 + 贪婪算法　；参考：《陕西师范大学》2016年硕士论文

【摘要】：分子影像能够在分子水平上对生物组织实现在体成像,为研究基因功能、疾病发病机理、疗效评估等方面提供了新方法,现已广泛应用于肿瘤检测、基因治疗和药物研发等领域。作为光学分子影像的一种重要模态,生物发光断层成像(Bioluminescence Tomography,BLT)是根据生物体表测量到的光子分布来反演体内光源(靶目标)的分布情况,与其他断层成像类似,这是典型的逆问题,特别是由于测量信息不足,加剧了重建问题的不适定性,使得准确地对光源进行三维重构成为挑战性难题。在现有基于有限元方法的生物发光断层成像中,由于靶向目标在生物组织中分布非常稀疏,光源所在区域包含的网格节点数量远远小于整个重建域中的节点数量。借鉴信号处理中的稀疏信号恢复和压缩感知重构的理论和方法,本文对能高效重建生物发光信号的稀疏重构算法进行了对比研究。已有的压缩感知重构算法包括凸优化算法、贪婪算法以及组合算法等。其中贪婪算法是通过迭代的方法,每次构造并求解一个局部最优解来逐步逼近全局最优解,具有计算代价小,效率高的优点。本文着重对基于贪婪思想的几种代表性算法进行了对比研究,包括正交匹配追踪(orthogonal matching pursuit,OMP)、分段正交匹配追踪(Stagewise Orthogonal Matching Pursuit,StOMP)、正则化正交匹配追踪(Regularized Orthogonal Matching Pursuit,ROMP),以及基于压缩感知的正交匹配追踪算法(Compressive Sampling Matching Pursuit,CoSaMP)等,结合生物组织的解剖结构先验,在有限元方法的基础上,分别将这些代表性算法结合到生物发光断层成像的稀疏光源重建中。为了评估和比较各算法的性能,在异质数字鼠模型上,设计了多组仿真实验对以上算法对单目标和多目标的重建能力进行了测试。实验结果表明:它们可以在含噪情况下准确重构出光源位置,特别是CoSaMP作为一种深度改进的匹配追踪算法,其表现出更好的定位稳定性以及对抗噪声的鲁棒性。本研究可为实际的生物发光断层成像应用给予算法选择指导。
[Abstract]:Molecular imaging can perform in vivo imaging of biological tissues at the molecular level, which provides a new method for the study of gene function, pathogenesis of disease, evaluation of therapeutic effect, etc., and has been widely used in tumor detection. Gene therapy and drug development and other areas. As an important mode of optical molecular imaging, Bioluminescence TomographyBLTs invert the distribution of light source (target) based on photon distribution measured by biological surface, which is similar to other tomographic imaging. This is a typical inverse problem, especially because of the lack of measurement information, which exacerbates the ill-posed problem of reconstruction, which makes the accurate 3D reconstruction of light source become a challenging problem. In the existing bioluminescence tomography based on finite element method the number of grid nodes in the region of light source is much smaller than that in the whole reconstructed domain because the target is very sparse in biological tissue. Based on the theory and method of sparse signal restoration and compressed sensing reconstruction in signal processing, this paper compares and studies the sparse reconstruction algorithm which can efficiently reconstruct bioluminescence signals. The existing compression perception reconstruction algorithms include convex optimization algorithm, greedy algorithm and combination algorithm. The greedy algorithm approximates the global optimal solution step by iterative method and constructs and solves a local optimal solution each time. It has the advantages of low computational cost and high efficiency. This paper focuses on the comparison of several representative algorithms based on greedy thought. These include orthogonal matching pursuit, piecewise Orthogonal Matching pursuit, regularized Orthogonal Matching pursuit, Compressed-Perception-based orthogonal Sampling Matching pursuit algorithm, Compressed-Sampling Matching pursuit, etc., combined with a priori anatomical structure of biological tissue. Based on the finite element method, these representative algorithms are applied to the sparse light source reconstruction of bioluminescence tomography. In order to evaluate and compare the performance of each algorithm, a number of simulation experiments were designed to test the reconstruction ability of the above algorithms on the heterogeneous digital rat model. The experimental results show that they can accurately reconstruct the position of the light source in the case of noise, especially CoSaMP, as a depth improved matching tracking algorithm, shows better localization stability and robustness against noise. This study can provide guidance for practical bioluminescence tomography applications.
【学位授予单位】：陕西师范大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41

【参考文献】