Numerical results for an accelerated projected gradient method with sparsity constraints for neuroimaging

The purpose of neuroimaging is to investigate the brain functionality through the localization of the regions where bioelectric current flows, starting from the measurements of the magnetic field produced in the outer space. Assuming that each component of the current density vector possesses the same sparse representation with respect to a pre-assigned multiscale basis, regularization techniques to the magnetic inverse problem are applied. The linear inverse problem arising can be approximated by iterative algorithms based on gradient steps intertwined with thresholding operations with joint-sparsity constraints. We propose some numerical tests in order to show the features of the numerical algorithm, also regarding the performance in terms of CPU occupancy.