Sparse representation of parallel MRI reconstruction

Joshua M Speciale, Purdue University

Abstract

Parallel imaging techniques enable the acceleration of MR image acquisition. This is made possible by spatial sensitivity information provided through the use of multiple receiver coils. The k-space is undersampled during the scan and the resulting image is reconstructed out of aliasing using the coil sensitivity data. The SENSE algorithm provides a general approach to reconstruction, but as more receiver coils and higher acceleration factors are used, so does the complexity of SENSE reconstruction. This increased complexity will begin to limit experimentation—especially in time-sensitive functional MRI—as the reconstruction time begins to exceed the accelerated acquisition time. The sparse matrix transform (SMT) is a generalization of the fast Fourier transform. The SMT was used to find a sparse approximate decomposition of the SENSE unfolding matrix for both simulated and actual scan data. A range of transform rotation orders and quantization threshold levels were utilized to find representations of the unfolding matrix as a series of sparse matrices. Comparison of these sparse representations to the original approach found that the number of multiplies required to perform SENSE reconstruction can be reduced significantly—by fifty percent or better—in the actual scan data. Both reduction and error levels increase with increasing acceleration factors. Future improvement in the complexity reduction level and error is likely, given ongoing work on advanced SMT decomposition techniques.

Degree

M.S.E.C.E.

Advisors

Talavage, Purdue University.

Subject Area

Biomedical engineering|Electrical engineering|Medical imaging

Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server
.

Share

COinS