[Camino-users] anomalous high FA values

[Ian Malone]

Hi,

Thanks for getting back to me. We've had a closer look at the images 
now. The high FA values are typically around the caudate and the rest of 
the ventricle boundary. This seems to correspond to voxels where the B=0 
intensity is LOWER than the weighted intensity in the source images. 
Obviously that's going to cause problems[1]. I don't think this can be 
genuinely anatomical or pathological, is there any mechanism where that 
can ever happen? I'm wondering if it might be an artefact caused by the 
high-contrast CSF-brain boundary around the ventricles in the B=0 
compared to weighted images, but there does seem to be a trail of low 
B=0 intensity up from the putamen in these images, so hard to be definite.

What we're considering doing is producing a mask image with a per-voxel 
criteria like mean b0 < mean weighted. Does this sound sensible? RESTORE 
would hopefully reject these too, but we have quite a lot of data which 
has already been processed.

[1] They just happen to be more apparent in the non-linear fit. These 
voxels do have exit code 0. Fitted diffusivity is very low (10^-5 mm^2/s 
axial, 10^-11 transverse).

Thanks again,
Ian

Philip A Cook wrote:
> Hi Ian,
>
> If the nonlinear fit fails to converge, you should get the linear result in the voxel. It sounds like it is converging (exit code 0) but it is converging on a bad solution. I expect that the ADC will be strange as well.
>
> You might have better luck with RESTORE. 
>
>
> Phil
>
> On Jan 28, 2013, at 1:30 PM, Ian Malone wrote:
>
>   
>> Hi,
>>
>> We're using Camino on some Siemens data, 42 directions, b=1000 s/mm^2, there are a few different protocols (different sites), but the problem seems to affect a few of them, one has 7 b=0 acquisitions. Fitting tensors using the non-linear least squares fit from modelfit we find a small number of voxels with FA>0.999, mostly in the caudate (sorry, don't have all the details yet on frequency). There are no obvious problems with the raw data and using FSL's FDT to do a simple least squares log signal fit doesn't produce this error. Looking at the error codes there are some 6's (bad data), but not corresponding to the problem voxels. I'd thought it might be a convergence problem, but we don't seem to be getting any error code = 2.
>>
>> Does anyone have any suggestions for things we could try to solve this problem? The Camino version is a fairly recent release (r1014).
>>
>> Thanks for your time,
>> Ian
>>
>> _______________________________________________
>> Camino-users mailing list
>> Camino-users at www.nitrc.org
>> http://www.nitrc.org/mailman/listinfo/camino-users
>>     
>
> _______________________________________________
> Camino-users mailing list
> Camino-users at www.nitrc.org
> http://www.nitrc.org/mailman/listinfo/camino-users
>
>   

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.nitrc.org/pipermail/camino-users/attachments/20130129/644901fc/attachment.html