Penalized-likelihood Algorithms for Time-Domain MR Diffusion Measure Estimation

点击翻译按钮获取中文摘要

The research proposed herein aims at developing accurate and robust methods for the estimation of both voxel-wise diffusion representations (diffusivity or anisotropy maps, diffusion tensor or spectrum maps) and global pathway structure from diffusion-weighted magnetic resonance (DW-MR) data. Although the algorithms will be widely applicable to diffusion MRI, the application of Interest is the imaging of cerebral white-matter structures. The proposed approach is probabilistic and it models two types of uncertainty that are present in DW-MR data: uncertainty introduced by the imaging process in the form of distortions and noise, and inherent uncertainty In the structures to be reconstructed due to individual variability in the underlying anatomy. The former will be addressed by accurate modeling of diffusion MR physics, including the effects of magnetic field inhomogeneities, eddy currents, and noise. The latter will be addressed by rich models of white-matter pathway anatomy, obtained by training the model on a set of subjects where major pathways have been defined manually. Cun'ently estimation of diffusion measures is suboptimal in that it is based on distorted images that are reconstructed without consideration for the underiying MR physics and then corrected for the distortions approximately in a series of post-processing steps. In addition reconstruction of white-matter pathways is labor-intensive because of the need for manual intervention to constrain the solution space and guide the tractography with neuroanatomical expertise. By addressing these issues the proposed project will make estimates of diffusion measures more accurate. It will also automate the reconstruction of white-matter pathways, making such studies practical even for large numbers of subjects. The proposed methods are being developed primarily to address the artifacts present at the data quality that is typical of routine in vivo studies. Thus we will evaluate and optimize our approach on such data. In addition, we will validate our methods on ex vivo brain acquisitions, where results from high-resolution, high-SNR images acquired in long scans can be used as a gold standard for comparison to results from routine-quality images.