摘要： 传统稀疏表示方法选择出的脑激活体素空间分布过于稀疏,不具有空间团块特性,在分析真实数据时的性能较低。针对该问题,提出一种基于先验信息的脑激活体素选择算法。该算法结合脑图像数据的高维性特点,以张量分析为基础,建立脑图像数据与任务函数之间的回归模型。用凸优化技术将脑激活体素的空间团块特性以凸约束的形式整合到体素选择过程中,使得该算法更加适合脑图像数据的信息提取,并采用脑激活区定位和解码实验对算法进行验证。实验结果表明,与传统稀疏表示算法相比,该算法选择出的脑激活体素空间分布更集中,在解码分析中能获得较高的解码准确率,在脑图像数据分析时表现出较高的求解质量和求解效率,能有效分析脑图像数据。

关键词: 体素选择, 稀疏表示, 功能核磁共振成像, 张量表示, 脑激活区定位

Abstract: Voxels selected by the traditional sparse representation algorithms are too sparse in spatial distribution and hardly show cluster effect,and the analysis performance of the real data is low.To overcome this problem,this paper proposes a prior information based algorithm for useful information extraction from brain image data.Considering the high dimensionality property of brain image data,a regression algorithm between function Magnetic Resonance Imaging(fMRI) data and task function is established by using tenor formulation.By introducing cluster effect of activated voxels in spatial distribution as prior information into voxel selection methods based on convex optimization technique,the proposed algorithms are more suitable for useful information extraction from brain image data.It uses experiments of brain activation localization and neural decoding to evaluate the proposed algorithm.Numerical results show that,compared with traditional sparse representation algorithm,voxels selected by the proposed algorithm are more concentrated in spatial distribution.It achieves higher decoding accuracy for decoding analysis.The proposed algorithm has high quality solution and can reliably process the brain image data.

Key words: voxel selection, sparse representation, function Magnetic Resonance Imaging(fMRI), tensor representation, brain activation area location

中图分类号: 

• TP393.41