Improved SNR in Diffusion Spectrum Imaging with Statistical Reconstruction

INTRODUCTION The diffusion spectrum imaging (DSI) technique [1] can map the probability density of the average relative spin displacement in a voxel, and has a unique capability to resolve complex intravoxel diffusion heterogeneity. This capacity is important because the voxel sizes in typical MRI experiments are large with respect to the diffusion scale, and the signal within each voxel can represent many distinct diffusional environments. There are two important practical limitations of DSI. First, because a large number of data samples need to be collected to sufficiently cover (k, q)space, the imaging time is inherently long. In standard implementations of DSI, a total of 515 different diffusion-weighted images are acquired; even with fast EPI pulse sequences, a full DSI experiment can take more than 25 minutes. Second, DSI data has very low signal-to-noise ratio (SNR), particularly for images encoded with heavy diffusion weightings. This is especially problematic because low SNR further limits our ability to improve resolution and imaging speed. As a result, in vivo DSI studies are typically acquired with relatively large voxel sizes in the range of 3-4 mm along each dimension, and still suffer from limited SNR.