Hi, 

there is still an issue there, this high in-plane resolution is likely due to interpolation at the scanner. Usually this interpolation takes place in k-space and the original data cannot be recovered. Take a look at the list archive, this issue has been discussed a lot before.

Cheers,
Stam



On 4 Apr 2011, at 10:45, Luis Morís wrote:

> Thanks a lot for your answer.
> 
> I have checked all the points you mentioned on your previous e-mail
> and maybe it's not the best set of images for doing a comparison.
> 
> I have one more question regarding difference between images, I have a
> set of images which its voxel size is 1.25, 1.25 and 2.5. And the
> other one is 2.5 2.5 2.5. Both of them have similar protocols
> (gradients direction, ect) except on which refers to voxel size during
> acquisition. Does normalizing the voxel size of the first group to 2.5
> 2.5 2.5 represent a problem for stablishing a comparison with DTI
> tractography?
> 
> Cheers and thanks again!
> 
> Luis
> 
> On Tue, Mar 29, 2011 at 11:59 PM, Stamatios Sotiropoulos
> <[log in to unmask]> wrote:
>> Hi Luis,
>> 
>> there are many issues with all the approaches that you are considering. I do not think you can do a meaningless comparison, for the following reasons:
>> 
>> - It is a bit strange to eliminate data from a good dataset (64 directions) in order to match the poorer quality of a 16 direction dataset. I would try to keep the best quality data and work with those.
>> 
>> - The number of directions alone is not determining the quality of the data. The direction schemes are (or should be) optimized, so that they are uniformly distributed on the sphere. A dataset of 16 uniform directions and another of 16 random directions are not comparable, the latter is likely to give you biased estimates. If you insist eliminating directions from the 64 direction dataset, this should be done in a systematic way that preserves the optimality of the scheme.
>> 
>> - Averaging directions: Are your directions unique? If yes, then you cannot average, you need to have repeats of exactly the same directions to average them. Averaging will increase the SNR for the datasets that you can apply this averaging compared to the datasets that you cannot. This means better resolving power for the former.
>> 
>> - Are the datasets of different resolution? What about the sequence itself, is it the same? Eddy-current compensation schemes have been employed for all acquisitions? Differences in all these factors can cause significant differences in the quality of the data and make them even less comparable.
>> 
>> - Even if we hypothetically ignore all the above issues, and assume that you can reduce all datasets down to 16 directions, they will not be good for probabilistic tractography. Depending on the SNR and the b-value, you will hardly be able to resolve crossings and also the path distribution is likely to be very wide. If however you want to do tractography using such a low number of directions, I would suggest keeping 1 fibre compartment per voxel (i.e. no crossings). Some "easy" tracts (in terms of geometry) should be reconstructed, so it depends on your application and the tract you are after.
>> 
>> Hope this is clear,
>> Stam
>> 
>> 
>> On 29 Mar 2011, at 20:52, Luis Morís wrote:
>> 
>>> Hello everyone,
>>> 
>>> At the moment I'm working on tractography with DTI images. I'm trying
>>> to make an study on images which has been done in different machines,
>>> (I know this is highly discouraged) and some of those images has
>>> different sequences from 16 to 64 gradient directions. It's not the
>>> ideal scenario.
>>> 
>>> My doubt is if there is anyway of reducing the 64 directions images to
>>> 16 in order to make them comparable. I have thought of simply erasing
>>> gradients until I have 16, but that could lead to some unbalance. Also
>>> averaging some images etc...
>>> 
>>> Is there a known method for acomplishing this? Does averaging DTI
>>> images and gradients makes any sense?
>>> 
>>> Cheers and thanks for your help,
>>> 
>>> Luis.
>>> 
>> 
>