基于凸优化的脑图像数据盲信号分离算法

Blind Signal Separation Algorithm of Brain Image Data Based on Convex Optimization

在实际脑图像分析中,独立成分分析方法的独立性假设很难完全满足。为此,结合脑图像数据的特点,以凸优化为基础,提出利用源分量稀疏性和非负性的脑图像盲信号分离算法。相比于独立性假设,稀疏性和非负性数学假设更符合f MRI数据的自然特性。将源分量的估计过程转化为寻找由观测数据构成的凸集合端点的过程。实验结果证明,由该算法选择出的激活体素与实验任务更相关,更容易进行生理解释。

Independent Component Analysis（ICA） is widely used in function Magnetic Resonance Imaging（fMRI） data analysis.However,recent studies show that the independence assumption for ICA based method is sometime violated in practice.In order to overcome this problem,combined with the characteristics of fMRI data,this paper proposes a new blind separation method,which exploits sparsity and non-negativity of sources,for brain image data.Compared with independence assumption,sparsity and non-negativity assumptions are considered more realistic to fMRI data.Based on non-negativity and sparsity assumptions,the new method estimates the source components by finding the extreme points of the observed fMRI data constructed convex set.Numerical results show that voxels selected by the proposed method are more related to task function and easily interpretable.