Meeting Banner
Abstract #3537

Rotationally Invariant Gradient Schemes for Diffusion MRI

Carl-Fredrik Westin1, Ofer Pasternak2, Hans Knutsson3

1Department of Radiology , BWH, Harvard Medical School, Boston, MA, United States; 2Department of Radiology, BWH, Harvard Medical School, Boston, MA, United States; 3Department of Biomedical Engineering, Medical Informatics, Linkping University, Linkping, Sweden

Minimizing the error propagation that a diffusion MRI gradient scheme introduces is an important task in the design of robust and un-biased experiments. We propose two schemes for the construction of rotationally invariant multiple-shells. The first is a dual frame method that optimizes the rotation invariance of any set of samples. The second uses a subset of the icosahedral set that can intuitively be used for nested rotationally invariant schemes with pre-defined number of samples. Rotationally invariance produces orientationally unbiased estimates and reduces the correlation of the samples, which is an important feature for reconstruction methods such as multiple shells and compressed sensing.

Abstract PDF E-Poster Video

Keywords

achieved addition ally always among angular another balanced basis believe better bias biased black blue combine combined combining comp complementary compress condition constant construct corresponds curve define defines defining design deviation diffusion distance dodecahedron dual electro electrostatic ells energy error example existing extremes face faces feature frame geom geometric gradient green half hindrance icosahedral icosahedron include independent interleaved introduced intuitively invariance invariant iteratively linearly link make many mate medical metric middle minimized minimizing minutes modify necessarily neighboring nested newer noise normalized observe optimal optimally optimization optimize optimizes optimum particular pattern patterns perfectly ping predefine previous produce profile propagation propose proposed providing quadratic quality radiology reduce remain require required restriction ridge ridges rotation rotational rotationally sample samples sampling scanner scheme schemes selecting sensing shell shells sing sometimes spaced special static studies subset sufficient takes task thirds truncated usually various yields