Yep, zebras are probably spike variants.
Best wishes-
Andreas

________________________________

Von: FSL - FMRIB's Software Library im Auftrag von Andy Alexander
Gesendet: Fr 20.10.2006 23:44
An: [log in to unmask]
Betreff: Re: [FSL] DTI quality assurance


This zebra artifact sounds like so-called "white pixel" or spike noise artifacts.  We had a lot of problems with this on our scanner (GE 3T SIGNA) for quite a while, although upgrading to the new HD EXCITE  platform seemed to clear this up (they put a filter that removes this sort of noise).  Essentially, this is caused by a transient (or spike) signal in k-space, which when reconstructed looks like stripes or cross-hatches if you have multiple versions.  This can be caused by jiggling cables, loose elements in the RF coils.  In our case, it was the passive metal shims, which had to be removed and readjusted (a fully day job).  DTI is particularly sensitive because of all the vibration from the gradients.  

If you see this consistently, you should have your field engineer troubleshoot this as soon as possible as data collected using a system with this problem will not be very good.

I also agree that a robust tensor fitting algorithm will help with occasional artifacts, although it sounds like this is a systematic problem.  you should get your hardware fixed asap.

Hope this helps.

- Andy Alexander

At 10:15 AM 10/19/2006 -0400, Chang, Lin-Ching (NIH/NICHD) [F] wrote:



	Hi,
	
	 
	
	If I may suggest, the robust tensor fitting is a good solution to your problem.
	
	The outliers will be detected and removed from the fitting automatically in a voxel by voxel basis.
	
	 
	
	http://dir2.nichd.nih.gov/nichd/stbb/restore_rob_est05.pdf
	
	 
	
	 
	
	Lin-Ching Chang
	
	 
	
	 
	
	 
	
________________________________

	From: Saad Jbabdi [mailto:[log in to unmask]] 
	Sent: Thursday, October 19, 2006 4:54 AM
	To: [log in to unmask]
	Subject: Re: [FSL] DTI quality assurance
	
	 
	
	 
	
	Hi,
	
	 
	
	It is hard to tell how good is good in DWI. Generally, it depends on the question you are addressing. You might want to optimise your sequence differently if your goal is to study FA and ADC or if you want to do tractography.
	
	 
	
	In your specific case, it is effectively strange to have slices with almost no signal, and could explain a higher FA in these slices if this drop occurs in one of the volumes and not in the others. Do you have any idea about why this happens ?
	
	 
	
	The mean intensity is not supposed to be the same over slices, but it is generally close across volumes within the same slice (except for b=0 of course). So you might be able to detect a "strange" slice by computing the mean and std across volumes of each slice (inside the brain) and checking for slices that diverge by n*std (depending on your experience). You might determine this by having a quick look at the distribution of mean signal and see if outliers are easy to detect. This is faster than looking at each slice separately.
	
	 
	
	cheers,
	
	saad
	
	 
	
	 
	
	On 19 Oct 2006, at 05:55, Hedok Lee wrote:
	
	
	
	Dear FSLers.
	
	 
	
	My apology for being slightly off the topic.
	
	 
	
	Im writing regarding a quality assurance of DWI images in DTI(b=0 and 
	
	b=1000) sequence.  Weve been collecting DTI along 25 directions+1 b0 with 
	
	26 slices(1mmx1mmx5mm) in each volume (676 images total) using GE 1.5T.  We 
	
	recently observed relatively high FA in one slice compare to the adjacent 
	
	slices, so I began looking at individual images.  It turns out this is due 
	
	to a few slices of almost no signal in DWI.  After checking 676 slices, I 
	
	was wondering if there is criteria, or diagnostic tool, to detect bad 
	
	slices.  Really high, or low, intensity DWIs are easy to detect.  Some of 
	
	the slices look suspicious but I dont have enough experience to tell 
	
	whether its good or bad.  I appreciate if someone has an experience, or a 
	
	tool, to set a criteria on this.  How do people trust what they get from 
	
	DTI is of decent quality?  
	
	 
	
	More specific questions are
	
	 
	
	Within the same volume, do mean intensity over slices supposed to be 
	
	close?  If so, whats the reasonable standard deviation.  How about the 
	
	mean intensities of volumes over different diffusion gradients.
	
	 
	
	Basically, I m trying to write a simple script to detect bad slices so 
	
	that I dont have to eyeball 676 images every time.  If there is a script 
	
	for this, please let me know.
	
	 
	
	Thanks,
	
	 
	
	Hedok
	
	 
	
	---------------------------------------------------------------------------
	
	Saad Jbabdi, 
	
	Postdoctoral Research Assistant,  
	
	Oxford University FMRIB Centre
	
	 
	
	FMRIB, JR Hospital, Headington, Oxford  OX3 9DU, UK
	
	+44 (0) 1865 222545  (fax 222717)
	
	[log in to unmask]    http://www.fmrib.ox.ac.uk/~saad
	
	---------------------------------------------------------------------------