Strategies for undersampling and reconstructing MR DTI data

Diffusion Tensor Imaging (DTI) has emerged as a reliable, non-invasive method of characterizing tissue micro-structure using MRI, but is limited by long scan time and low SNR. In order to accelerate acquisition, different strategies for undersampling and a model-based reconstruction method are presented. The model-based approach estimates diffusion tensors directly from undersampled k-space data via minimizing an L2-norm cost function. Three different undersampling schemes are investigated: variable density, center-only and center offset. Each scheme was compared against a gold standard and found to perform better than the fully encoded case with equivalent scan time. Minor differences in the performance metrics and qualitative observations were seen among the schemes. These findings suggest the proposed strategy can be used to reduce DTI scan time while incurring little loss in parameter estimation accuracy.