Improved whole-brain multivariate hemodynamic deconvolution for multi-echo fMRI with stability selection

Blind estimation of neuronal-related activity from functional magnetic resonance imaging (fMRI) data of resting-state, naturalistic paradigms or clinical conditions can be performed with paradigm free analysis methods such as hemodynamic deconvolution. These methods usually employ a linear hemodynamic convolution model and sparsity-pursuing regularized estimation techniques to detect single-trial blood oxygenation level-dependent (BOLD) responses in the brain. Recently, a deconvolution algorithm tailored for multi-echo fMRI data, named multi-echo sparse paradigm free mapping (ME-SPFM), exploited the linear dependence of the BOLD percent signal change on the echo time (TE) to compute voxel-wise estimates of the changes in the apparent transverse relaxation ![Graphic][1] associated with brain activity, achieving superior performance than its single-echo counterparts. In this work, we propose a novel multivariate formulation that operates at the whole brain level, considering all the voxels as a group in a mixed-norm regularization term to add spatial information in the deconvolution. In addition, we introduce a stability selection procedure that operates voxelwise and avoids the critical selection of the regularization parameter for the deconvolution. This procedure allow us to define a novel metric based on the area under the curve (AUC) of the stability path of the estimation, which provides valuable and complementary information about the probability of having a neuronal-related event at each voxel and time-point. We demonstrate that these novel features yield more robust results compared with the original ME-SPFM algorithm, and show high spatial and temporal agreement with the activation maps and BOLD signals obtained with a standard model-based linear regression approach. In summary, the proposed multivariate multi-echo sparse paradigm free mapping (MvME-SPFM) approach with stability selection provides more reliable estimates of ![Graphic][2] for the study of the dynamics of brain activity when no information about the timings of the BOLD events is available. ### Competing Interest Statement The authors have declared no competing interest. [1]: /embed/inline-graphic-1.gif [2]: /embed/inline-graphic-2.gif