部分容积效应下的神经纤维方向估计模型与算法

发布时间：2018-10-05 07:36

【摘要】：基于扩散磁共振信号的脑白质纤维成像是目前重构活体大脑组织微观结构的唯一一种非入侵性的方法。在多种扩散磁共振方法中,高角度分辨率扩散成像(HARDI)技术不仅克服了扩散张量模型难以重构交叉纤维的缺点,而且需要较少的扩散采样数量,成为脑纤维成像研究的热点。其中球面反卷积模型作为目前最高效的HARDI成像方法之一,大大提高了纤维重构角度的精确性。然而由于采样体素具有相对较大的体积,在脑组织之间存在不可忽略的部分容积效应,这将导致传统的纤维成像方法的不准确。针对上述问题,本文在经典球面反卷积模型的基础上,设计了新的脑纤维成像模型,同时解决了反卷积模型中球谐函数的高次谐波截断、最小二乘算法的不适定性等问题,本文的主要工作如下:(1)针对脑组织中的部分容积效应问题,提出了一种多响应核函数模型。该模型综合考虑组织中的各向同性扩散信号和各向异性扩散信号得到了新的纤维方向分布函数,从而可以从扩散加权信号中分离出各向同性扩散信号,克服了组织中的部分容积效应,提供了更加精确的纤维方向的估计。(2)基于字典基函数表示纤维方向分布函数,提出了一种新的量化各向同性信号强弱的指标。该指标基于多响应核函数模型,利用纤维方向分布函数的一部分分离出组织中的各向同性扩散信号。(3)提出一种求解多响应核函数球面反卷积模型的Richardson-Lucy(RL)算法。该算法利用额外的先验知识,在模型中增加了全变差正则化和稀疏正则化项解决了最小二乘算法的高度不适定性,保证了纤维方向的非负性。模拟数据和实际人脑数据的实验表明,本文方法可以得到更加精确的纤维重构方向,显著提高了纤维角度分辨率,并且新的量化指标相比较FA,GFA等表现出更好的特性。
[Abstract]:Diffusion magnetic resonance imaging (DMR) based white matter imaging is the only noninvasive method to reconstruct the microstructures of brain tissue in vivo. Among various diffusion-magnetic resonance methods, high-angle resolution diffusion imaging (HARDI) technique not only overcomes the disadvantage of diffusion-Zhang Liang model to reconstruct cross-fibers, but also requires a small number of diffusion samples, so it has become a hot spot in the field of brain fiber imaging. As one of the most efficient HARDI imaging methods, spherical deconvolution model greatly improves the accuracy of fiber reconstruction angle. However, due to the relatively large volume of sampled voxels, there is a partial volume effect between brain tissues, which will lead to the inaccuracy of traditional fiber imaging methods. Based on the classical spherical deconvolution model, a new model of cerebral fiber imaging is designed in this paper, and the problems of high-order harmonic truncation of spherical harmonic function and ill-posed least-squares algorithm in the deconvolution model are solved. The main work of this paper is as follows: (1) A multi-response kernel function model is proposed for the partial volume effect in brain tissue. Considering the isotropic diffusion signal and anisotropic diffusion signal in the tissue, the new fiber direction distribution function is obtained in this model, and the isotropic diffusion signal can be separated from the diffusion-weighted signal. It overcomes the partial volume effect in the tissue and provides a more accurate estimation of fiber direction. (2) based on the dictionary basis function to represent the fiber direction distribution function, a new quantitative isotropic signal strength index is proposed. Based on the multi-response kernel function model, the isotropic diffusion signals in the tissue are separated by a part of the fiber direction distribution function. (3) A Richardson-Lucy (RL) algorithm for solving the spherical deconvolution model of the multi-response kernel function is proposed. Using extra prior knowledge, the algorithm adds total variation regularization and sparse regularization to the model to solve the highly ill-posed problem of the least squares algorithm, which ensures the non-negative orientation of the fiber. The experimental results of simulated data and actual human brain data show that the proposed method can obtain a more accurate direction of fiber reconstruction and improve the resolution of fiber angle significantly. The new quantitative index shows better characteristics than FA,GFA and so on.
【学位授予单位】：浙江工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R741.044;TP391.41

【相似文献】