Compressed Sensing for Motion Artifact Reduction

1
Compressed Sensing for Motion Artifact Reduction
4593
Presentation Wednesday _at_ 3pm
Joëlle K. Barral Dwight G. Nishimura
Electrical Engineering Stanford University
2
In a Nutshell

• Navigators are useful to correct motion of
  small amplitude.
• They can also be used to detect data that
  needs to be discarded.
• Discarding data provides an undersampling
  dataset

a Compressed Sensing approach can be used to
reduce motion artifacts in high-resolution MRI.
3
Motion Artifacts
4
Classical Navigators
kx ky kz
Zero-padding
Fast Large Angle Spin Echo 3D FLASE
FFT
kx
Projection along x
Projections
TR number
Cross-correlation
TR 80 ms
Navigators interleaved
Shifts
Phase-modulation
Ehman MRI 1989173255-263 -- Wang MRM
199636117-123 -- Song MRM 199941947-953
5
Rejecting Outliers
LARYNX -- Volunteer scan
Shift of small amplitude, well approximated by a
translation
Shift of large amplitude, that cannot be
corrected outlier
SI Superior/Inferior AP Anterior/Posterior LR
Left/Right
Resulting undersampled trajectory after outliers
rejection
kz
2 outliers
256x12 encodes
ky
6
Randomizing the Acquisition
"A man has made all his decisions at random. He
did not do worse than others who consider
carefully their choices" Paul Valéry
Phantom scans -- FLASE sequence
256x16 encodes, 11 outliers
Korin JMRI 19922687-693 -- Wilman MRM
199738793-802 -- Bernstein MRM 200350802-812
7
Compressed Sensing (1/2)
Simulation with in-vivo data
256x32 encodes, 30 outliers
Sequential acquisition
Pseudo-random acquisition
8
Compressed Sensing (2/2)
256x32 encodes, 30 outliers
Haacke JMR 199192126-145 -- Lustig MRM
2007581182-1195
9
Discussion

• A pseudo-random acquisition often avoids
  getting corrupted samples that are contiguous in
  k-space.
• If the undersampled trajectory (after outlier
  rejection) is incoherent, Compressed Sensing
  allows an accurate reconstruction.
• However, how can the acquisition be robust
  against the worst case scenario (since motion is
  truly random) where the undersampled trajectory
  that we first get is not incoherent?

10
Diminishing Variance Algorithm
Number of pixels
mode
Acquire encodes and navigators
Compute shifts
mm
Determine prioritized list of encodes to reacquire
Priority distance from histogram mode
(weighted by distance from k-space origin)
Scan time 6 min 12 s (38 s overhead to
reacquire the outliers)
Scan time 5 min 34 s
Sachs MRM 199534412-422
11
Future work Diminishing Variance Algorithm
Coherency
pseudo-randomly
Acquire encodes and navigators
Compute shifts
Determine prioritized list of encodes to reacquire
Determine prioritized list of encodes to reacquire
Priority incoherency of the underlying
undersampled trajectory
12
Conclusion

• Pseudo-random acquisition
• Outliers rejection

• Diminishing Coherency Algorithm

Incoherent undersampled trajectory
13
Thank you!
Acknowledgments Michael Lustig, Bob Schaffer,
Uygar Sümbül, Juan Santos
Contact jbarral_at_stanford.edu