提高白质纤维交叉重构能力是有效提高纤维跟踪技术的前提之一,目前大多纤维重构方法都是基于白质体素的独立重构,没有考虑到纤维的连续性特征,这就促使文章从全局范围考虑提高白质纤维重构能力.文章提出了一种基于全变差空间正则化的纤维方向分布估计方法,该方法首先利用字典基分布的球面反卷积策略拟合多壳采样信号,为了能够适用于单壳和多壳采样方案,文章重新定义了广义的纤维响应函数;进而在$q$空间中定义基函数系数的全变差正则化约束,旨在减少不必要的方向信息,降低因噪声引起的方向偏差,以获得纤维方向的空间局部一致性.实验分别在模拟数据和实际数据下进行,分别采用单壳和多壳数据验证了文章所提方法能够以更高效的性能实现纤维方向估计,相对于其他算法显著提高了纤维的连续性.

It is an important prerequisite of development of tractography to improve the accuracy of crossing fiber reconstruction. Most of the existing reconstructing methods are voxel-wise, which are sub-optimal for the disregard of the fiber consistency. Thus, in this work we propose a global fiber estimation method based on local sparsity and spatial total variation regularization to improve the accuracy of fiber orientation distribution. Firstly, spherical deconvolution based on a dictionary basis is built for the multi-shell signal fitting, in which the response function is re-estimated with different sensitive coefficients. Then a total variation in the $q$-space is forced on the basis function coefficients to guarantee the spatial consistency of brain fiber, which eliminates the spurious directions caused by noise. The results of experiments on simulated and in vivo data demonstrate the efficiency of the proposed methods in fiber orientation estimation when single-shell and multi-shell data involved, and compared with other methods, the present methods shows a prominent superiority in fiber consistency.