Meeting Banner
Abstract #2248

Improving Rank Constrained Reconstructions Using Prior Information with Reordering

Ganesh Adluru1, Liyong Chen1, 2, David Feinberg2, Jeffrey Anderson1, Edward V.R. DiBella1

1Radiology, University of Utah, Salt Lake City, UT, United States; 2Advanced MRI Technologies, Sebastopol, CA, United States

Image reconstruction using a rank penalty term is a promising way to remove undersampling artifacts in multi-image MRI. Exciting results have been reported in dynamic imaging situations where temporal signal changes are highly correlated. However, when the underlying true data have a lot of variation, a low rank constraint may not be the best choice. Here we propose a reordering technique to improve rank constrained reconstructions in such cases. Pixel intensities in the matrix of the multi-image estimate are reordered based on the sorting order of a prior. This results in a better match with the low rank model. Promising results are presented on undersampled multi-image diffusion data.

Keywords

acquisition advantages aliasing alternated applied applying artifacts axial better brain breast certain city coil coils column columns combined completion complex compressed constrained constraint constraints context convex correlated dataset density described determined developments diffusion directly dynamic earlier encoding error except existing exploiting explored fashion fewer fidelity find formed framework fully hence highly ideas imaginary improve improving incorporate independently individually injected instead intensities intensity lake leading loss match matrix minimization minimizing monotonic norm nuclear offer offers operator optimization overall overlapping parts patient peak penalty perfect perfusion pixel pixels plot plots practice prior projection promising propose proposed quality radiology random rank reached readout real recent reconstructed reconstruction reconstructions reduced reduces refocusing relaxed reordered reordering reorders reported represents salt sampled satisfy scanner sensing separated sets significantly simple simultaneous singular slice slices sorted sorting space spatial squared squares step stroke structures taking technologies temporally term threshold thresholding transforms translation true underlying update variable variation vectorized zeroed