Structured sparse deconvolution for paradigm free mapping of functional MRI data

Paradigm-free mapping enables to map the hæmodynamic response in space and time without prior knowledge of the timing of the underlying neuronal events (i.e., no stimulation paradigm). Such deconvolution approach can take advantage of modern sparsity-promoting regularization. Here we extend this concept using structured sparsity approaches in order to gain robustnesss against model mismatch. Specifically, we extend the hæmodynamic dictionary with the informed basis set (i.e., canonical HRF, and its temporal and dispersion derivatives) and we deploy state-of-the art structured sparsity functionals. In addition, we propose the group-weighted fusion penalty. We demonstrate the feasibility of the proposed approach for both synthetic and experimental data, showing superior abilities to characterize the single-trial BOLD response with no timing information.