Compressed dictionary learning for detecting activations in fMRI using double sparsity

This paper focuses on detecting activated voxels in fMRI data by exploiting the sparsity of the BOLD signal. Due to the large volume of the data, we propose to learn a dictionary from the compressed measurements of the BOLD signal. The solution to the inverse problem induced by the General Linear Model is then sought through sparse coding using the double sparsity model, where sparsity is imposed on both the learnt dictionary and the generated coefficients. The estimated sparse coefficients are then used to decide whether or not a stimulus is presented in the observed BOLD signal. Experimental results on real fMRI data demonstrate that the proposed method leads to similar activated regions as compared to those activated by the Statistical Parametric Mapping (SPM) software but with much less samples needed.