This subproject is one of many research subprojects utilizing the resources provided by a Center grant funded by NIH/NCRR. The subproject and investigator (PI) may have received primary funding from another NIH source, and thus could be represented in other CRISP entries. The institution listed is for the Center, which is not necessarily the institution for the investigator. Project Description:
From Specific Aims of P41 RR005959 IV.A.1c: We will optimize a 3D (dual-channel) radial acquisition sequence to limit the effects of susceptibility. The radial acquisition sequence will be implemented with two channels. We will use a modular approach for radial encoding and reconstruction through all of Core A, i.e. the acquisition and reconstruction elements will be built in such a fashion that they can be readily adapted to their use in Specific Aims A.1?A.3. The 3D sequence will uniformly sample a spherical volume of k-space. Trajectories are defined by wave tables generated in MATLAB or more recently using an analystical expression that can calculate trajectories within the psd, allowing larger image arrays. End-points of the trajectories are equidistant on the surface defining the edge of the sampled space. This approach allows us to define arbitrary volumes, e.g. a prolate spheroid for anisotropic acquisition in which spatial resolution along one axis can be traded for sensitivity. The radial data acquired will be reconstructed using a novel new automation architecture to facilitate integration of new sampling strategies used throughout Core A. We have previously developed a sophisticated software platform for automated reconstruction of very large multi-dimensional image arrays. New requirements are imposed on the reconstruction workflow by the imaging methods proposed in this Core A, e.g. separation of multi-channel data, pre-filtering, and view sorting. These tasks are in addition to the responsibilities of the current automated reconstruction (recon) tool, e.g. data combining, and baseline correction. This added complexity, and a decision to embrace both production imaging (for our Collaborators), and new acquisition strategies (Cores A and B), demand a new structure for the recon tool. Our current recon tool (see Resources, Appendix J) and Appendix I.A3-Johnson (Johnson et al., 2007) grew both by evolution and design, as an application on top of the underlying k-space-transforming reconstruction algorithms. In 1991, we developed a large array 3DFT reconstruction program. As time passed, this reconstruction program was driven by a command line of increasing power. For flexibility, we found it useful to separate image scaling from reconstruction. Perl scripting allowed us to simplify the user interface. The script was extended to handle more steps: data splitting for multi-echo and cine acquisitions, data combining for segments of k-space, baseline removal, byte-swapping and accommodation of data formats from multiple scanner versions. The new recon tool will be modeled as a series of data transformations. k-space data is transformed into spatial data, which is converted into images. The Annotated Store class transmits original acquisition parameters and added reconstruction parameters through the transformation process. Since the inception of our recon tool, the Computer Science community has developed a programming methodology called object-oriented programming (OOP), which addresses many of our needs in a more elegant fashion. We will use OOP Perl as demonstrated in Figure IV.A.5 to support the variety of radial reconstructions required for this core. By basing the new tool on OOP, we can naturally extend the tool to all protocols used in the lab and transform the Center's workflow into a much more efficient framework. The data transformations chaperoned by the new tool are controlled by the Store Processor. Object-oriented methods simplify coding of new classes of processing. New processing classes inherit their functionality from the general Store Processor, and specific adaptations differentiate them. Shared functionality is passed to specialized processors. For example, keyhole 2D radial is a variation derived from general 2D radial. The same principles hold for the k to Space Converter class, from which more specialized converters inherit functionality. To support radial acquisitions, such as those described here, the recon tool will transform k-space data to spatial data using the 3D Non-uniform fast Fourier transform (NUFFT). The NUFFT algorithm, written in C and currently implemented on the SGI workstation, is described in detail in the preprint included in Appendix I.A1-Song. NUFFT has been developed as an alternative to conventional methods that pre-weight the data, convolve it with a shift invariant kernel, and execute an inverse Fourier Transform (Jackson et al., 1991) (Beatty et al., 2005). NUFFT calculates complex kernels based on the given sampling pattern to minimize interpolation errors. The method provides increased SNR and flexibility relative to the more traditional regridding technique we have previously used (Chen et al., 2005;Johnson et al., 2001).
Showing the most recent 10 out of 239 publications
