Fast NSR: An Optimized Non-Linear Stochastic Deconvolution for Large Data Sets and Clinical Analyses

Denis Peruzzo¹, Danilo Benozzo¹, Gianluigi Pillonetto¹, Alessandra Bertoldo¹

We present fast NSR, an optimization of the original Non-linear Stochastic Regularization algorithm to quantify the residue function and the CBF in DSC-MRI. Fast NSR introduces a preliminary step that allows to overcome the limits in the original NSR implementation, such as the sensitivity to the starting points and the required computational time. In the preliminary step the optimal starting points and the stochastic component of the residue function are computed for each voxel. Fast NSR elaborates a whole subject in a couple of hours and is now suitable for the analysis of large data sets and in clinical context.