BRAIN ACTIVATION LOCALIZATION ALGORITHM BASED ON ALTERNATING VOLUME MAXIMIZATION METHODS

Liushengqian Fengbao Zhanglieping and Sunxuan

ABSTRACT: Recent studies showed that the independence assumption for Independent components analysis (ICA) based method is sometime violated in practice for function Magnetic Resonance Imaging (fMRI) data analysis. In order to overcome this problem, we proposed a new blind separation method for brain image data analysis. Combined with the natural characteristics of fMRI data, the new method exploits sparsity and non-negativity of sources to blindly decompose fMRI data. The proposed method estimates the source components in a convex analysis framework. It is achieved in two steps. First, it shows that source components serve as extreme points of a convex set, which is constructed based on the observed fMRI data. Next, all the source components can be estimated by finding extreme points of the convex set obtained in the first step. For determination of extreme points, alternating volume maximization is exploited to obtain robustness of fMRI model errors to improve accuracy of blind separation accuracy of fMRI data. Numerical results showed that, compared with ICA based method, voxels selected by the pr oposed method are more related to task function. The proposed method has high quality solution and the solving efficiency and can reliably process the brain image data.

Keywords: Alternating Volume Maximization Methods, function Magnetic Resonance Imaging (fMRI), Independent Component Analysis, Convex Optimization, Voxel Selectio