Reconstruction Algorithms

From Body Magnetic Resonance Group

Jump to: navigation, search

MR image reconstructions contain broad areas that include non-Cartesian image reconstruction (such as gridding), undersampled data reconstruction (such as parallel imaging), off-resonance correction, fat/water separation, gradient imperfections and segmentation and analysis tools. Depending on applications, not only excitation and acquisition schemes are properly designed but also efficient reconstruction techniques should be followed to achieve good image quality and provide diagnostic utility.

In body MRI, reliable fat suppression is important to show pathology or tissue morphology better and to increase dynamic ranges of images. Among different fat suppression methods, fat/water separation techniques are the least sensitive to field inhomogeneity. In addition, rapid imaging is necessary to reduce artifacts by cardiac or breath motions during a scan. To reduce scan time, using non-cartesian trajectory such as spiral and EPI, partial Fourier methods and undersampling k-space data is pretty common, which all require specific reconstructions to get artifacts-free images. Undersampled data reconstruction techniques using compressed sensing and parallel imaging are shown at the Rapid Imaging section. Here are some of our reconstruction techniques for fat/water separation.

Fat/Water Separation

Estimating field map correctly is important for reliable fat/water separation from multiecho sequences. Here is a example of fat/water separation by estimating field maps using golden section search combined with multiresolution pyramid structure [1]. Here, field map values can be estimated efficiently by coarse-to-fine propagation, exploiting smoothly varying nature of field maps.

Separated water/fat images and estimated field map of abdomen, where usually large field inhomogeneity is present. Fat/water images were acquired from a 42 second breath-hold scan with covering the entire liver (256 x144 x 28 matrix size).


Bipolar multiecho sequences replace fly-back gradients in unipolar multiecho sequences with alternating readout gradients and reduce echo spacing, thus can provide more robust field map estimation, shorter scan times, higher SNR efficiency, reduced motion-induced artifacts and less sensitivity to T2* decay. However, the alternating readout gradients cause problems including delay effects and image misregistrations. We solved these problems by using k-space water-fat separation, eliminating chemical shift-induced artifacts and correcting k-space echo misalignment [2].

Schematic diagram of bipolar multiecho GRE sequence.
Separated fat and water images of knee scan from bipolar multiecho sequence at 1.5T.

References

Error fetching PMID 24549782:
Error fetching PMID 24227703:
Error fetching PMID 23843341:
Error fetching PMID 23821305:
Error fetching PMID 24512800:
Error fetching PMID 23868425:
Error fetching PMID 23596017:
Error fetching PMID 23943610:
Error fetching PMID 23280540:
Error fetching PMID 23292822:
Error fetching PMID 23042658:
Error fetching PMID 22172536:
Error fetching PMID 22865658:
Error fetching PMID 22791572:
Error fetching PMID 23097185:
Error fetching PMID 22334505:
Error fetching PMID 22095672:
Error fetching PMID 22179942:
Error fetching PMID 22038883:
Error fetching PMID 22055852:
Error fetching PMID 21509870:
Error fetching PMID 21287596:
Error fetching PMID 21287593:
Error fetching PMID 21319219:
Error fetching PMID 20677276:
Error fetching PMID 20632411:
Error fetching PMID 20373445:
Error fetching PMID 19780156:
Error fetching PMID 19267347:
Error fetching PMID 19025951:
Error fetching PMID 18581397:
Error fetching PMID 18581362:
Error fetching PMID 17390355:
Error fetching PMID 17326087:
Error fetching PMID 17260387:
Error fetching PMID 17449772:
Error fetching PMID 16455400:
  1. Error fetching PMID 18581362: [lu2008a]
  2. Error fetching PMID 18581397: [lu2008b]
Personal tools