Magnetic resonance diffusion tensor imaging (DTI) is useful for characterizing tissue microstructures such as the brain white matter and myocardial fiber orientation, but it is limited by long scan times. Because a DTI dataset consists of scans of the same volume, but with different diffusion sensitization directions, there is high degree of redundancy in the acquired data. The redundancy can be exploited to decrease scan time by reduced sampling in conjunction with appropriate constrained image reconstruction. In this study, we retrospectively applied "compressed sensing" technology to a 3D DTI dataset acquired on a fixed primate brain specimen, employing a model-based formulation of the DTI signal. The white matter orientation, fractional anisotropy (FA) and mean diffusivity (MD) maps were compared to those alternatively achievable in the same scan time by two control cases, either reducing the image resolution or the number of sensitization directions, and by two other reconstruction techniques, conventional compressed sensing and asymmetrical sampling. Results of the comparison indicate that the proposed approach significantly outperforms the other means to reduce scan time. The findings show compressed sensing to be promising for DTI, and can be used to improve its resolution, accuracy and/or reducing its scan time.
Diffusion Tensor Imaging (DTI) has become a preferred method to rapidly and noninvasively quantify the 3D myofiber structure of the heart, and atlases representing group averages of hearts in humans and some animal species have been constructed from DTI data using computational anatomical techniques and voxel-based statistics. Besides being sensitive to image noise, a key assumption in the latter is that fiber orientations in different voxels are independent of one another. Moreover, few studies have directly investigated the intra-species variability of the myocardial fiber structure. Given the known structural and functional interconnectivity of the myocardium, intuitively, a pattern-based analytical approach may better reflect the individual anatomy and capture the group similarity and variability. The present study developed a computational framework in which fiber angle fields, calculated in prolate coordinates, were parameterized by polynomials of various transcendental basis functions (e.g., ordinary polynomials, sinusoids or hyperboloids).The parameterization was then used for variability analysis of the mouse heart myofiber structure. Results show that parametric modeling offers the benefit of greatly reduced dimensionality of an otherwise highly complex function. The findings underscore the advantages of the proposed framework, including accuracy and simplicity of the description, which have implications for current and future constructions cardiac computational models.