[Mrtrix-discussion] Slice collapse problems

[Watts, Richard]

Hi Dorian -

Yes, you are quite correct, that the halfscan factor is Philips terminology for partial Fourier.  I agree that it is painful to push the TE out a bit, but if you get rid of a serious confound, then it's probably worth it.

There is an interesting discussion about the choice of B0, gradient strength, TE, and signal strength for the Human Connectome Project in
Setsompop, K., R. Kimmlingen, E. Eberlein, T. Witzel, J. Cohen-Adad, J. A. McNab, B. Keil, M. D. Tisdall, P. Hoecht, P. Dietz, S. F. Cauley, V. Tountcheva, V. Matschl, V. H. Lenz, K. Heberlein, A. Potthast, H. Thein, J. Van Horn, A. Toga, F. Schmitt, D. Lehne, B. R. Rosen, V. Wedeen and L. L. Wald (2013). "Pushing the limits of in vivo diffusion MRI for the Human Connectome Project." Neuroimage.

With an estimate T2 for WM of 65ms, the relative signal strength should be approximately exp(-4ms/65ms) = 94%.  On the other hand, every little helps.  I've included some comments on the effects of halfscan below.

For the Human Connectome Project (UW St Louis), and they are getting b=3000 s/mm2 with TE=89.5ms using 3/4 "Phase partial Fourier" and a SENSE (GRAPPA) factor of two on a souped up Siemens Connectome Skyra.  On our Philips Achieva we were closer to TE=100ms on a similar protocol.  They get a big boost in throughput using multiband to get 3 slices at once.  We're hoping that Philips will implement multiband for our almost-upgraded 3T Achieva with dStream and 32-channel head coil (which will have similar hardware to the HCP).

There are methods out there to detect and correct for these dropouts.  For example, Susumu Mori's group just published
"Image Corruption Detection in Diffusion Tensor Imaging for Post-Processing and Real-Time Monitoring" in PLOS one
http://www.plosone.org/article/info%3Adoi%2F10.1371%2Fjournal.pone.0049764

On the other hand, it is usually better to avoid artifacts in the first place, rather than trying to correct for them later.  Just my 2c-worth!

Cheers -


Richard Watts
University of Vermont


On Nov 6, 2013, at 11:46 AM, Dorian P. <alb.net at gmail.com<mailto:alb.net at gmail.com>>
 wrote:

The last point, smaller k-space coverage, seem to be related to a parameter called "halfscan" in Philips:

Halfscan is a method in which approximately only one half of the acquisition matrix in the phase encoding direction is acquired.

Effects if Halfscan is set to `Yes'

Signal-to-Noise Ratio:  Reduced with ÷2
I would have expected sqrt(2) reduced based on signal averaging

Spatial resolution:  Not affected
Agreed

Scan time  Shorter (almost a reduction by a factor of 2)
I doubt that the factor of 2 is realistic for single-shot diffusion EPI, since halfscan will only reduce the length of the readout train, not the rest of the sequence, including the diffusion encoding.

Susceptibility effects:  More sensitive to field inhomogeneities and susceptibility
Maybe, but for susceptibility-induced distortion, I don't think that the degree of distortion will change with halfscan.  This is determined by the field of view and the SENSE factor.


Artifacts:  More prominent flow and motion artifacts
Probably.  The homodyne reconstruction is more complicated, and makes assumptions about the signal phase.


We have been advised to increase the halfscan factor (currently 0.702), but in doing so I have to loose another 4ms TE (going to 98ms). I tried this and effectively the scan with higher halfscan factor had no slice problems compared to our regular hardi on the same subject (2-3 bad slices). The problem is that the signal is weaker and I was not sure the benefit would be consistent in the future.

Is this what you were referring for fourier transform? Do you have any advise on this?


Thank you.

Dorian
TJU



2013/11/6 Watts, Richard <Richard.Watts at vtmednet.org<mailto:Richard.Watts at vtmednet.org>>
My explanation for whole-slice signal dropout on diffusion is that it's an interaction between head rotation, the diffusion-weighting gradients and the k-space coverage.

Warning... here comes the physics!
Head rotation produces a position-dependent velocity (zero at the center of rotation, negative on one side, positive on the other).  The diffusion gradients convert this into a linear variation of phase with position (same mechanism as phase contrast imaging).  Applying the Fourier shift theorum makes this equivalent to a shift of the center of k-space.  If the center moves out of your k-space coverage, then you will end up dropping the low spatial frequencies and signal dropout.

Incidentally, the phase variability due to very small head movements is why we can't generally do multishot DW-EPI.  Think of diffusion as phase-contrast MRI on steroids!

Whole-slice dropout is most likely to occur when:
1. Your subject moves a lot (obviously!)
2. You use a high b-value (bigger phase shift for a given motion), especially in directions away from the z-axis
3. You use a smaller k-space coverage (as Jesper/Romain noted 5/8 partial Fourier is worse than 3/4)

I believe that the manufacturers have become a little more conservative with partial Fourier to try to avoid these problems, at the expense of increased TE.

Cheers -


Richard Watts
University of Vermont



_______________________________________________
Mrtrix-discussion mailing list
Mrtrix-discussion at www.nitrc.org<mailto:Mrtrix-discussion at www.nitrc.org>
http://www.nitrc.org/mailman/listinfo/mrtrix-discussion