基于3D稀疏表示的多模态神经导航关键技术研究

Intracranial tumor is one of the most common diseases in the nervous system. To assist the doctor to observe the fiber tacts distribution, it is urgent to make clear that how to fully exploit the high redundancy of the diffusion imaging data in 3D space to more rapidly and accurately reconstruct the nerve fiber tracts around the tumor area, and afterwards fuse the salient information in the multimodal medical images. The project studies on sparse 3D model of diffusion data, and synthesizes the underdetermined compressive sampling matrix in the form of the subsampling matrix tensor product and by this way, we proposes to design matrix reconstruction algorithms for subsampling matrices to improve the efficiency. The method breaks the restrictions of the existing models regardless of 3D spatial whole redundancy with low compressive sampling rate. Secondly, we aim to solve the problem of errors accumulation of the deterministic tracking algorithm vulnerable to noise, so propose the spatial correlation based noise suppression method using 3D sparse representation. In the complex fiber distribution area, such as intersection or bifurcation, the fiber tact orientation is described by the particles flocking motion. Local fiber deterministic tracking algorithm is designed via a global optimization algorithm to guide the multiple tensor models. Eventually, 3D/joint saliency models are constructed to compute fusion weight for 3D images, which fully visualizes the 3D anatomical structure of the tumor region and adjacent fiber tracts, moreover, lays the foundation to clinical applications for the neuro-navigation system with high accuracy.

颅内肿瘤是神经系统中常见的疾病之一。为辅助医生观察神经纤维的分布，如何充分利用弥散成像数据3D空间高度冗余性，更加快速、高精度的重建肿瘤区的神经纤维传导束，并融合多模态医学影像的显著性信息，是亟待解决的问题。项目首先研究弥散数据的3D稀疏表示模型，通过子采样矩阵张量积合成欠定压缩采样矩阵，提出一种对子采样矩阵设计重构算法来提高重构效率的方法，突破现有模型没有考虑3D空间整体冗余性、压缩采样率不高的限制；其次，针对确定性追踪算法易受噪声影响造成误差累计的问题，提出一种基于空间相关性的3D稀疏去噪方法，并在交叉、分岔等纤维复杂分布的区域，通过粒子群体运动描述纤维束走向，提出一种全局优化算法引导的多张量模型进行局部精准的纤维确定性追踪。最终，构造3D/联合显著性模型，设计3D图像融合权重，充分显示肿瘤区域的3D解剖结构及周围纤维束分布，从而为高精度的神经导航系统走向临床应用奠定基础。

颅内肿瘤是神经系统中常见的疾病之一。为辅助医生观察神经纤维的分布，如何充分利用弥散成像数据3D空间高度冗余性，更加快速、高精度的重建肿瘤区的神经纤维束，并融合多模态医学影像的显著性信息，对脑神经纤维分析进行可视分析是亟待解决的问题。项目首先研究弥散张量数据的重构与追踪，提出一种噪声强度估计方法降低重构及追踪误差；其次，针对纤维分布复杂度高的问题，提出基于散布矩阵的纤维筛选和密度峰值聚类算法，降低视觉混杂性；最终，融合纤维和MRI结构信息，充分显示肿瘤区域的3D解剖结构及周围纤维束分布，并设计三维空间交互方法，从而为高精度的神经导航系统走向临床应用奠定基础。