利用多层受约束一范数优化检测功能磁共振成像中神经活动

Detection of multi-layer brain neural activities in functional MR images using constrained L1 optimization

本研究提出利用fMRI中神经信号内在的稀疏性,通过积分器转换,最大期望算法优化对脑fMRI中血流动力学变化建立多层神经信号模型,将检测脑fMRI中神经活动转化为受约束的一范数优化问题。利用空间自适应滤波器,优化结果可以准确地检测出fMRI中神经活动。通过与目前主流检测方法时间聚类分析、最大相关性方法及图模型推理法对比,本文提出的方法能够以较小的计算复杂度得出精确的结果 。

In this paper,a framework was proposed to utilize the sparsity within neural activities. Through integrators and EM approach,a multi-layer neural hymodynamic response model was established. By converting the neural activity detection problem into a finding the sparse solution in constrained L1 optimization problem,using adaptive spatial filtering,brain neural activities in multiple scales can be detected. The proposed method was compared with temporal cluster analysis （TCA）,the maximum correlation method （MCM）,and graphical model inference （GMI）. The experimental results demonstrated the computational efficiency and detection accuracy of the proposed approach.