FMRI activations via low-complexity second-order inverse-sparse-transform blind separation

Abstract

Since functional Magnetic Resonance Imaging (fMRI) signals are a group of sparse signals, and its autocorrelation matrix contains limited information, it is difficult to accurately locate the brain activation area directly using traditional blind separation algorithms. For the issue, this paper proposes a method with inverse-sparse transform and second order blind identification (SOBI) for the separation of the activations. The contribution of this paper is to achieve the separation of sparse brain map signals and have lower computational complexity than higher-order statistical BSS. In experiments, we use both simulated and measured fMRI data to evaluate our method. The experimental results show that the proposed method's running time is only 1/30 of a higher-order statistical independent component analysis (ICA) algorithm, while its separation errors is close to ICA and less than half of a traditional SOBI algorithm.