Accelerated motion-robust non-Cartesian multi-shot diffusion-weighted imaging with reconstruction in the image space

Purpose: To achieve high-resolution diffusion-weighted imaging (DWI) with short duration acquisition and robustness to patient motion. In [1, 2], high resolution (HR) dense k-space sampling of a DW image was achieved with a series of anisotropically oversampled acquisitions (so-called "shots"), which amounts to sampling k-space in a non-Cartesian manner, and by reconstruction in the image space. It was shown to provide a theoretical 8x acceleration at equal SNR compared to conventional sampling [1]. However, in [1, 2], each DWI was reconstructed separately. First, an isotropic HR gradient image could not be reconstructed if one of its shots was corrupted by intra-scan motion, even if other shots for this gradient were successfully acquired. Second, the fact that the DWIs constitute different views of the same anatomy was ignored. DW images are coupled, and this correlation of information can be leveraged by introducing in the reconstruction the knowledge of the local tissue microstructure. We propose to describe the tissue microstructure at a voxel with a diffusion compartment imaging (DCI) tissue model (e.g., Multi-tensor model, NODDI, DIAMOND) that provides a model-based description of the signal attenuation for any diffusion gradient orientation and strength. This tissue model also enables model-based generation of the non acquired shots. We propose a novel non-Cartesian multi-shot DWI technique that simultaneously achieves HR reconstruction and DCI model estimation (Simultaneous multi-sHot highresOlution ReconsTruCtion and diffUsion comparTment imaging, SHORTCUT). It enables reconstruction from shots with different subsets of gradients, providing increased robustness to patient motion and potential for acceleration. Method. SHORTCUT is formalized as a joint probabilistic model synthesized in Fig.1. Specifically, the simultaneous estimation of HR images x and of DCI model parameters t (see Fig.1) is performed according to the maximum a posteriori principle, by maximizing: (1). The likelihood relates to the HR reconstruction (HRR) and, assuming conditional independence, decomposes into: . The term incorporates an