[FieldTrip] SPM12 for segmentation and (inverse) normalization

Stephen Whitmarsh stephen.whitmarsh at gmail.com
Fri Sep 22 13:01:25 CEST 2017


Hi Arjen, Jan-Mathijs, et. al.,

It seems the rotation was caused by a bug my side. The segmentation using
SPM12 solved problems caused by low quality MRIS.
Thanks!
Stephen


On 22 September 2017 at 10:09, Stephen Whitmarsh <
stephen.whitmarsh at gmail.com> wrote:

> Hi Arjen,
>
> Indeed, I do not think there is a problem with segmentation of
> brain/skull, as can be seen on the image. Stripping some skin of two
> subjects (thresholding the 'soft_tissue' probability output of SPM12, then
> removing it) with a similar problem of solved the rotation, but resulted in
> too small grids... On this subject I attached this procedure has no effect.
>
> However, at least for those other subjects normalization on skullstripped,
> or rather, scalpstripped MRIs might do the trick. As I understand it,
> however, the spm8 procedure (and spm12 I think) is a two-stepped procedure,
> with (affine) transformation based on the scalp first, after which it
> optimizes it based on brain segmentation. I would not know how to therefor
> do normalization without scalp. In fact, it expects a full volumetric
> image, not a (pre-)segmented one.
>
> Cheers,
> Stephen
>
> On 21 September 2017 at 18:45, Arjen Stolk <a.stolk8 at gmail.com> wrote:
>
>> First thought is a registration of brain outline to skull (instead of
>> brain), although at closer inspection the shift seems overall just a bit
>> too large for that. You could try calculating the normalization parameters
>> on skullstripped volumes (unless you want to keep non-brain tissue).
>>
>> On Thu, Sep 21, 2017 at 9:20 AM, Stephen Whitmarsh <
>> stephen.whitmarsh at gmail.com> wrote:
>>
>>> Hi Arjen,
>>>
>>> Thanks, and good to hear you've not been let down yet. It might be the
>>> fact that I have some bad quality MRIs to deal with. However... does this
>>> problem (see attached) ring a bell for anyone?:
>>>
>>> Brain segmentation is proper, and co-registration with polhemus
>>>  head-shape as well, but inverse warp to MNI result in a tilted grid.
>>> Linear vs. non-linear transformation gives the same result. Other subjects
>>> going through the same procedure work fine, except two others wherein I
>>> identified it as a problem in segmenting the scalp and therefor the first
>>> step of the normalization. This one looks absolutely fine in every other
>>> regard, however.
>>>
>>> I'm stumped...
>>>
>>>     cfg                        = [];
>>>     cfg.spmversion          = 'spm12';
>>>     cfg.grid.warpmni        = 'yes';
>>>     cfg.grid.template       = template_grid;
>>>     cfg.grid.nonlinear      = 'yes';
>>>     cfg.mri                 = mri_realigned;
>>>     cfg.grid.unit           = 'mm';
>>>     subject_grid            = ft_prepare_sourcemodel(cfg);
>>>
>>> Cheers,
>>> Stephen
>>>
>>>
>>> On 21 September 2017 at 17:00, Arjen Stolk <a.stolk8 at gmail.com> wrote:
>>>
>>>> Hey Stephen,
>>>>
>>>> Look for discussions regarding spm12 and also dartel on bugzilla. It's
>>>> been a while but as far as I can remember ft_volumenormalize is the only
>>>> function now that has not been integrated. Reason being that it wasnt
>>>> straightforward to house the dartel procedure under a single function, so
>>>> this is ongoing work still. You can however use spm12's coregistration
>>>> function with ft_volumerealign (for rigid body transformations), which Im
>>>> using quite a bit and never let me down (and is much faster than before).
>>>> But that wouldnt work for normalization to template space though (use spm8).
>>>>
>>>> Best
>>>>
>>>> On Sep 21, 2017, at 6:56 AM, Herbert J Gould (hgould) <
>>>> hgould at memphis.edu> wrote:
>>>>
>>>> I have retired please remove me from the mail list
>>>>
>>>> Herbert Jay Gould
>>>> Professor Emeritus
>>>> The University of Memphis
>>>>
>>>>
>>>>
>>>> Sent from my Verizon Wireless 4G LTE smartphone
>>>>
>>>>
>>>> -------- Original message --------
>>>> From: Stephen Whitmarsh
>>>> Date:09/21/2017 7:43 AM (GMT-06:00)
>>>> To: FieldTrip discussion list
>>>> Subject: Re: [FieldTrip] SPM12 for segmentation and (inverse)
>>>> normalization
>>>>
>>>> Dear Sarang and Jan-Mathijs,
>>>>
>>>> Thanks a lot. I am now able (after updating FT, which now includes
>>>> SPM12 in /external), to use SPM12 for segmentation of my template and my
>>>> subject MRI, by using cfg.spmversion = 'spm12'. 12 is definitely is a big
>>>> improvement over 8 when it comes to brain-segmentation, which now does not
>>>> require individual treatments anymore. It also outputs more compartments
>>>> which gives me a little bit more to work with when dealing with scans that
>>>> have bad delineation of the scalp for normalization.
>>>>
>>>> Pleas note that defaults seems to differ - some FT functions default to
>>>> spm8, others to spm12.
>>>>
>>>> In fact, FT still reverts to spm8 in ft_volumenormalise when called in
>>>> ft_prepare_sourcemodel, even when calling the latter with cfg.spmversion =
>>>> 'spm12'. In other words the cfg.spmversion is not passed along.
>>>>
>>>> Best wishes and thanks again!
>>>> Stephen
>>>>
>>>>
>>>>
>>>> On 21 September 2017 at 09:09, Schoffelen, J.M. (Jan Mathijs) <
>>>> jan.schoffelen at donders.ru.nl> wrote:
>>>>
>>>>> Hi Stephen,
>>>>>
>>>>> Please note that FT now has full support for SPM12, both using the
>>>>> old-style segmentation, and the new one (the latter yielding 6 tissue
>>>>> types).
>>>>>
>>>>> Best,
>>>>> Jan-Mathijs
>>>>>
>>>>> On 20 Sep 2017, at 17:03, Stephen Whitmarsh <
>>>>> stephen.whitmarsh at gmail.com> wrote:
>>>>>
>>>>> Dear all,
>>>>>
>>>>> I having some problems in normalizing MRIs for my study. Some have
>>>>> improper segmentation for which changing individual brain/scalp thresholds
>>>>> works in many cases but not all, e.g. when the scalp 'bleeds' into some
>>>>> noise outside of the head. Also, changing parameters in spm8 for
>>>>> normalization, such as number of iterations (directly in in spm_normalize,
>>>>> since FT does not pass these parameters) improves the transformation.
>>>>>
>>>>> However, some scans I cannot deal with, either because they have noise
>>>>> from outsides of the head 'bleed' onto the scalp, thereby preventing
>>>>> optimal scalp-segmentation and thereby normalization. Others have an
>>>>> inappropriate contrast MRI sequence.
>>>>>
>>>>> Some fMRI researchers advised me to use SPM12, because of its improved
>>>>> preprocessing procedures. However, it does not seem supported in FT yet.
>>>>> Does anyone have experience with this, and can perhaps share how they
>>>>> extracted the transformation matrix from the resulting nifti's?
>>>>>
>>>>> Thanks,
>>>>> Stephen
>>>>> _______________________________________________
>>>>> fieldtrip mailing list
>>>>> fieldtrip at donders.ru.nl
>>>>> https://mailman.science.ru.nl/mailman/listinfo/fieldtrip
>>>>>
>>>>>
>>>>>
>>>>> _______________________________________________
>>>>> fieldtrip mailing list
>>>>> fieldtrip at donders.ru.nl
>>>>> https://mailman.science.ru.nl/mailman/listinfo/fieldtrip
>>>>>
>>>>
>>>> _______________________________________________
>>>> fieldtrip mailing list
>>>> fieldtrip at donders.ru.nl
>>>> https://mailman.science.ru.nl/mailman/listinfo/fieldtrip
>>>>
>>>>
>>>> _______________________________________________
>>>> fieldtrip mailing list
>>>> fieldtrip at donders.ru.nl
>>>> https://mailman.science.ru.nl/mailman/listinfo/fieldtrip
>>>>
>>>
>>>
>>> _______________________________________________
>>> fieldtrip mailing list
>>> fieldtrip at donders.ru.nl
>>> https://mailman.science.ru.nl/mailman/listinfo/fieldtrip
>>>
>>
>>
>> _______________________________________________
>> fieldtrip mailing list
>> fieldtrip at donders.ru.nl
>> https://mailman.science.ru.nl/mailman/listinfo/fieldtrip
>>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20170922/e69812e4/attachment-0002.html>


More information about the fieldtrip mailing list