[MINC-users] Problem with mritotal

[Ed Gronenschild]

Andrew (and Claude),

Here are the requested results of xfm2param.
For the original data (voxel size is 1.0 * 1.0 * 1.0 mm^3):
-center         0.00000    0.00000    0.00000
-translation  -70.45679  142.58327  101.26605
-rotation     -14.28202    6.44659   -0.85951
-scale          0.85468    0.87173    0.87334
-shear          0.00000    0.00000    0.00000

Total scale factor = 0.65068

For the data with the voxel size set to 1.2 * 1.0 * 1.2 mm^3:
-center         0.00000    0.00000    0.00000
-translation  -99.12584  188.75531   93.02094
-rotation     -24.77424    0.41362   -1.06088
-scale          0.90102    0.94179    0.95061
-shear         -0.00000   -0.00000    0.00000

Total scale factor = 0.80666
Converted to the original voxel size this would be 1.16159.

Notice also the difference in rotation angles!!

The head size of the offendig scans is the same as that of the
successfully processed scans: width left->right = 130 mm and
width AP = 172 mm. So, that cannot be the error source. I think
that Claude may be right to use a mask in order to have a more
robust registration. The fact that the position of the head in the
field of view changes from scan to scan (especially in the Z
direction) could that be a reason for the bad performance?

By the way: the non-uniformities in our 3T scans are almost absent,
qualitatively I don't see much differences between the original and
N3-corrected scans.

Ed

On 17 Aug 2007, at 13:25, minc-users-request at bic.mni.mcgill.ca wrote:

> Message: 6
> Date: Fri, 17 Aug 2007 21:25:17 +1000
> From: "Andrew Janke" <a.janke at gmail.com>
> Subject: Re: [MINC-users] Problem with mritotal
> To: "MINC users mailing list" <minc-users at bic.mni.mcgill.ca>
> Message-ID:
> 	<d747ed50708170425q70c795cdn2b8ee99d4a9fe230 at mail.gmail.com>
> Content-Type: text/plain; charset=ISO-8859-1
>
> On 8/17/07, Claude LEPAGE <claude at bic.mni.mcgill.ca> wrote:
>
>>> Currently we are acquiring MRI volumes on a 3T machine. After
>>> conversion to minc format I used the command 'insect' to preprocess
>>> the scans. I noticed that for some scans this resulted in too small
>>> brains which, in addition, were also not correctly orientated in
>>> stereotaxic orientation. Most probably 'mritotal' is not able to
>>> derive a correct (linear) transformation. To check this I  
>>> artificially
>>> multiplied the pixel size in X and Z by a factor 1.2 in the  
>>> conversion
>>> to minc format (so without changing the voxel data). This  
>>> resulted in
>>> a correctly sized and orientated volume.
>
> Ed, I what does xfm2param return for your final xfm? If it doesnt have
> a scale of 0.8 or so in both X and Y I would suspect that you have an
> error in your conversion to MINC or that you are scanning people with
> small heads.
>
> As Claude said, the MNI tools are (inadvertently) best at 1.5T data of
> about "adult head" size. This is not to say that they will not work on
> other sized data, just that this is how they have been most tested.
>
>> These new tools are used internally at the BIC and haven't been
>> officially packaged. Andrew, what do you think?
>
> By all means I think that bestlinreg.pl is ready for prime time.  My
> plan was to include it (and nlpfit) in the next release of mni-autoreg
> if that is OK by you. For now you can always put it on packages for
> others to have a go with it.  You can always use the scripts directory
> for now.
>
>
> a
>