摘要： 针对独立成分分析(ICA)方法的独立性假设在实际脑图像分析中很难完全满足的问题,根据脑图像数据的特点,提出一种基于字典稀疏性的盲分离方法。结合字典学习和盲分离技术,利用源信号在所选字典中的稀疏性,将盲分离的过程转换到稀疏域中,通过在稀疏域中对源信 号施加稀疏性约束进行盲分解。采用脑激活区定位实验对该方法进行验证,并用线性相关方法进行结果比较。结果证明,与ICA方法相比,该方法得到的激活体素平均时间序列与实验任务更相关,激活体素空间分布也更集中,具有较高的求解质量和求解效率,可用于更好地分析 脑图像数据,在稀疏域中进行盲分解时提高盲分解的质量。

关键词: 盲分离, 功能核磁共振成像, 独立成分分析, 字典学习, 体素选择

Abstract: As the Independent Components Analysis(ICA) is difficult to be fully satisfied in real brain image analysis.According to the characteristics of function Magnetic Resonance Imaging(fMRI) data,this paper proposes a new blind separation method,which exploits sparsity of sources in a dictionary,for brain image data.The proposed method combines dictionary learning and blind source separation technique.By exploiting sparsity of source component in a signal dictionary,the blind separation process is converted to the transformed sparse domain.It uses experiments of brain activation localization to evaluate the proposed method.Experimental results show that,compared with ICA based method,voxels selected by the proposed method are more related to task function and more concentrated in spatial distribution.The proposed method has high quality solution and the solving efficiency can reliably process the brain image data.Estimating source component in the sparse domain is helpful for improving quality of blind separation equality.

Key words: blinds separation, function Magnetic Resonance Imaging(fMRI), Independent Component Analysis(ICA), dictionary learning, voxel selection

中图分类号: 

• TP393.41