An off-the-grid approach to multi-compartment magnetic resonance fingerprinting

Abstract We propose a novel numerical approach to separate multiple tissue compartments in image voxels and estimate quantitatively their nuclear magnetic resonance (NMR) properties mixture fractions, given ?ngerprinting (MRF) measurements. The number of tissues, types or quantitative are not a-priori known, but the is assumed be composed sparse with linearly mixed Bloch magnetisation responses within voxels. Fine-grid discretisation multi-dimensional NMR creates large highly coherent MRF dictionaries that can challenge scalability precision methods for (discrete) approximation. To overcome these issues, we an o?-the-grid equipped extended notion group lasso regularisation approximation using continuous (non-discretised) response models. Further, nonlinear non-analytical approximated by neural network, enabling e?cient back-propagation gradients through proposed algorithm. Tested on simulated in-vivo healthy brain data, demonstrate e?ectiveness scheme compared baseline multi-compartment methods.