rjgenik
commented
3 years ago Hi Rob,
I have seen something similar but in a completely different context, outside of 3-d imaging, when considering NN vibration simulations in 3D solid state modelling.
On the phase image, you see a smooth demarcation at the 2pi to 0 wrap coming down the left side. This bounces up one voxel right at the minimum, and seems to put an exact zero at the voxel neighbor down and left of the VOI. This leads the NN smoothing algorithm to a discontinuity when centered on the VOI.
What I tracked down in my situation was an incorrect boundary transition, something like < should have been <=, or vice versa, or an average was being calculated with raw numbers instead of considering special cases at the boundary. The offending elseif block was either in the phase algorithm or the NN averaging, but the issue precipitated to taking the mean including the 0-2pi transition. The algorithm was treating these terms as averaging to pi, when the correct answer should be also near the 0-2pi boundary. I solved this with something like a treating the zero, or near-zero, as 2pi + epsilon in the mean calculation in certain cases.
Hope that helps some.
Sincerely,
Rich
From: Robert Smith @.*** Sent: Thursday, October 14, 2021 5:07 AM To: CMRR-C2P/MB Cc: Subscribed Subject: [CMRR-C2P/MB] Dark voxels with MB + SENSE1 (#278)
[EXTERNAL]
Hi all,
I've not been able to find any reference to the particular artifact I've observed, so hoping that this isn't a duplicate and that someone may be able to provide some insight.
I had noticed incidentally in my in vivo DWIs "dark voxels" where the mean signal was lower than that expected given their respective neighbourhoods. I subsequently then observed the same in fMRI experimentation on a simple water bottle phantom. What appears to be happening is some form of local phase cancellation arising from the multi-slice unaliasing. Here is an example of what the magnitude and phase images look like:
Magnitude, stack of slices: [SENSE1_dropout]https://user-images.githubusercontent.com/5637955/137283854-ce127e8a-711c-4124-8cf6-8db36a0c6471.PNG
Magnitide, within-slice: [SENSE1_dropout_2]https://user-images.githubusercontent.com/5637955/137283907-9e6b96d9-432e-449a-b323-1283848304f3.PNG
Phase, within-slice: [SENSE1_dropout_4]https://user-images.githubusercontent.com/5637955/137283986-a80fbfc7-9a22-4cf5-9b3c-ca26e9843d76.PNG
This is not merely a matter of phase wrap: there's a 2pi cycle observed when tracing a circle centred at the highlighted voxel, and that couldn't possibly be corrected by any unwrapping algorithm. This makes the precise phase in that voxel ill-defined, which presumably leads to unwanted destructive interference across coils (or something like that).
Testing has shown that this artifact only appears if all of the following are true:
- SENSE1 is used (I have not seen the same effect under any circumstance using SoS)
- A specific head coil is used (here it appears with our 32ch coil but not our 64)
- MB of a certain factor is used (I see it at our chosen MB4, not at MB2 or using single-band, though I have not exhaustively tested all possible MB factors)
I'm guessing that this is some intrinsic interaction between the geometry of the slice stack and the geometry of the head coil, and thus can't be trivially addressed. I am however curious to know if anybody else has witnessed the same effect, and whether any thought has gone into how such issues could conceivably be corrected at the image reconstruction stage.
Cheers Rob
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHubhttps://github.com/CMRR-C2P/MB/issues/278, or unsubscribehttps://github.com/notifications/unsubscribe-auth/ABZYTIA7LDTAWYY45Z6AX23UG2MTBANCNFSM5F7EL5GA. Triage notifications on the go with GitHub Mobile for iOShttps://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Androidhttps://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub.
eauerbach
whitecraine
Lestropie
adamnarai
Hi all,
I've not been able to find any reference to the particular artifact I've observed, so hoping that this isn't a duplicate and that someone may be able to provide some insight.
I had noticed incidentally in my in vivo DWIs "dark voxels" where the mean signal was lower than that expected given their respective neighbourhoods. I subsequently then observed the same in fMRI experimentation on a simple water bottle phantom. What appears to be happening is some form of local phase cancellation arising from the multi-slice unaliasing. Here is an example of what the magnitude and phase images look like:
Magnitude, stack of slices:
Magnitide, within-slice:
Phase, within-slice:
This is not merely a matter of phase wrap: there's a 2pi cycle observed when tracing a circle centred at the highlighted voxel, and that couldn't possibly be corrected by any unwrapping algorithm. This makes the precise phase in that voxel ill-defined, which presumably leads to unwanted destructive interference across coils (or something like that).
Testing has shown that this artifact only appears if all of the following are true:
I'm guessing that this is some intrinsic interaction between the geometry of the slice stack and the geometry of the head coil, and thus can't be trivially addressed. I am however curious to know if anybody else has witnessed the same effect, and whether any thought has gone into how such issues could conceivably be corrected at the image reconstruction stage.
Cheers Rob