Hi John,
A follow up question with DARTEL normalization of fMRI data. As you replied
previously (https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=spm;6d04fba6.1202),
fMRI DARTEL normalization will have holes because that voxels in the
original image are pushed to their new locations by scanning through the
original and adding their values to the appropriate locations of the
normalised version. The holes are pretty severe when normalize the fMRI
data to 2x2x2 voxels.
You mentioned a solution is to smooth the data. However, if we are
interested in some fine structures (e.g., amygdala) and need to avoid
smooth, what kind of other remedy you will suggest?
Thank you very much!
Best,
Chao-Gan
On Fri, Jul 29, 2016 at 1:54 AM, John Ashburner <[log in to unmask]>
wrote:
> Sorry. My mistake. I'm so used to answering VBM questions, and not used
> to people wanting to get better results from their fMRI.
>
> Yes, you are right. The way you originally had the batch is correct.
>
> (4). Normalize into MNI space
> DARTEL template: template_6.nii
> Few subjects
> Flow fields = u_rc1xx
> Images = rxxx.nii
> Preserve = Preserve Concentrations ("no modulation")
>
> Now it makes sense that you wanted 2mm voxel sizes and no "modulation".
>
> Best regards,
> -John
>
>
> On 28 July 2016 at 18:16, 陳志剛 <[log in to unmask]> wrote:
>
>> Hi John,
>>
>> Thanks but, my aim is to normalize my realigned EPI data (that's why I
>> use " Images = rxxx.nii").
>> Your suggestion (i.e., " Images = cxxx.nii") seems normalizing the gray
>> matter of structural imaging to MNI space, right?
>>
>> Or, maybe you mean that, using "Flow fields = u_rc1xx; Images =
>> cxxx.nii," it will produce a *xxx_sn.mat *file, and then I do the
>> following to normalize my EPI data?
>>
>> Normalize: Write
>>
>> Parameter files: *xxx_sn.mat *
>>
>> Images to write: rxxx.nii
>>
>> Voxel size: [2 2 2]
>>
>> Interpolation: 7th degree B-spline
>>
>>
>> Mike
>>
>>
>> 2016-07-29 0:23 GMT+08:00 John Ashburner <[log in to unmask]>:
>>
>>> I'd suggest:
>>>
>>> (4). Normalize into MNI space
>>> DARTEL template: template_6.nii
>>> Few subjects
>>> Flow fields = u_rc1xx
>>> Images = cxxx.nii
>>> Preserve = Preserve Amount ("modulation")
>>>
>>> It's been a while since I've done anything with SPM8, but I guess you
>>> could use a voxel size of 2 mm. I don't know if it makes any difference to
>>> the results.
>>>
>>> Best regards,
>>> -John
>>>
>>> On 28 July 2016 at 16:06, 陳志剛 <[log in to unmask]> wrote:
>>>
>>>> Hi John,
>>>>
>>>> Thanks. Let me clarify further, please.
>>>>
>>>> This is my original batch:
>>>>
>>>> (1). Segment
>>>> Images = Coregistered T1WI
>>>> Grey matter/white matter/CSF = Native space
>>>> [This process produced a xxx_sn.mat file as well as c1xx, c2xx, c3xx
>>>> files corresponging to GM, WM, and CSF]
>>>>
>>>> (2). DARTEL tool, Initial import
>>>> Parameter files = xxx_sn.mat
>>>> GM, WM, CSF: Yes
>>>> Voxel size: 1.5
>>>> [This process produced rc1xx, rc2xx, rc3xx file]
>>>>
>>>> (3). DARTEL tool, run DARTEL
>>>> Images: rc1xx, rc2xx, rc3xx
>>>> [This process produced u_rc1xx, u_rc2xx, u_rc3xx file, and a series
>>>> of template files, with the last one called template_6.nii]
>>>>
>>>> (4). Normalize into MNI space
>>>> DARTEL template: template_6.nii
>>>> Few subjects
>>>> Flow fields = u_rc1xx
>>>> Images = rxxx.nii
>>>> Preserve = Preserve Concentrations ("no modulation")
>>>> [This process produced swrxxx.nii but also Template_6_2mni.mat]
>>>>
>>>> By saying "I'd suggest normalising the native-space images (c*.nii)
>>>> rather than the imported versions (rc*.nii), as this can reduce aliasing
>>>> effects", do you mean:
>>>>
>>>> (4). Normalize into MNI space
>>>> DARTEL template: template_6.nii
>>>> Few subjects
>>>> Flow fields = c1xx
>>>> Images = rxxx.nii
>>>> Preserve = Preserve Concentrations ("no modulation")
>>>> [This process produced swrxxx.nii but also Template_6_2mni.mat]
>>>>
>>>> Also, for the voxel size issue, how about:
>>>>
>>>> (2). DARTEL tool, Initial import
>>>> Parameter files = xxx_sn.mat
>>>> GM, WM, CSF: Yes
>>>> Voxel size: 2.0
>>>>
>>>> Thanks in advance.
>>>> Mike
>>>>
>>>>
>>>>
>>>>
>>>> 2016-07-28 20:59 GMT+08:00 John Ashburner <[log in to unmask]>:
>>>>
>>>>>
>>>>> I'd suggest normalising the native-space images (c*.nii) rather than
>>>>> the imported versions (rc*.nii), as this can reduce aliasing effects.
>>>>>
>>>>> As for whether or not you use modulation when you write out the warped
>>>>> images, this will depend on your actual question. Do you want to see where
>>>>> there are differences in the registration errors of one group versus the
>>>>> other, or do you want to see the differences in tissue volumes? I find it
>>>>> hard to put a biological interpretation on unmodulated VBM results,
>>>>> although they seem to lead to greater sensitivity to differences.
>>>>>
>>>>> There's no easy way to change the voxel sizes of the spatially
>>>>> normalised images. If you want to do this, you'll need to write MATLAB
>>>>> code.
>>>>>
>>>>> Smoothing is combined with writing out the warped images. If you
>>>>> don't want it to smooth, you can set the FWHM to zero.
>>>>>
>>>>> Best regards,
>>>>> -John
>>>>>
>>>>>
>>>>> On 27 July 2016 at 21:18, Mike <[log in to unmask]> wrote:
>>>>>
>>>>>> Hi DARTELers,
>>>>>>
>>>>>> Because my subjects are aged patients with possible brain structural
>>>>>> changes, I want to perform DARTEL normalization rather than default spatial
>>>>>> normalization in SPM8 so as to have a better spatial normalization.
>>>>>>
>>>>>> All EPI images were adjusted for timing differences between slices,
>>>>>> unwarped using field maps, realigned and re-sliced to correct for motion
>>>>>> artifacts (producing rxxx.nii). The resulting mean EPI was co-registered
>>>>>> with the subject's T2-weighted image, which in turn was aligned with the
>>>>>> T1-weighted image (that is, T1WI has been coregistered to EPI).
>>>>>>
>>>>>> Below is my procedure for DARTEL normalization and I hope someone
>>>>>> could help to confirm my procedures:
>>>>>>
>>>>>> (1). Segment
>>>>>> Images = Coregistered T1WI
>>>>>> Grey matter/white matter/CSF = Native space
>>>>>> [This process produced a xxx_sn.mat file as well as c1xx, c2xx,
>>>>>> c3xx files corresponging to GM, WM, and CSF]
>>>>>>
>>>>>> (2). DARTEL tool, Initial import
>>>>>> Parameter files = xxx_sn.mat
>>>>>> GM, WM, CSF: Yes
>>>>>> Voxel size: 1.5
>>>>>> [This process produced rc1xx, rc2xx, rc3xx file]
>>>>>>
>>>>>> (3). DARTEL tool, run DARTEL
>>>>>> Images: rc1xx, rc2xx, rc3xx
>>>>>> [This process produced u_rc1xx, u_rc2xx, u_rc3xx file, and a series
>>>>>> of template files, with the last one called template_6.nii]
>>>>>>
>>>>>> (4). Normalize into MNI space
>>>>>> DARTEL template: template_6.nii
>>>>>> Few subjects
>>>>>> Flow fields = u_rc1xx
>>>>>> Images = rxxx.nii
>>>>>> Preserve = Preserve Concentrations ("no modulation")
>>>>>> [This process produced swrxxx.nii but also Template_6_2mni.mat]
>>>>>>
>>>>>> Are my procedures correct?
>>>>>> However, the voxel size in swrxxx.nii is 1.5x1.5x1.5. If I want to
>>>>>> get a MNI standard space with 2x2x2 voxel size, how should I do?
>>>>>> Also, (4) performed not only normalization but automatically
>>>>>> smoothing process...
>>>>>>
>>>>>> Thanks in advance. Mike
>>>>>>
>>>>>
>>>>>
>>>>
>>>
>>
>
--
Chao-Gan YAN, Ph.D.
Principal Investigator
Deputy Director, MRI Research Center
Institute of Psychology, Chinese Academy of Sciences
16 Lincui Road, Chaoyang District, Beijing 100101, China
-
Initiator
<http://rfmri.org/DPARSF>DPABI <http://rfmri.org/DPABI>
<http://rfmri.org/DPARSF>, <http://dpabi.org>DPARSF
<http://rfmri.org/DPARSF>, PRN <http://rfmri.org/PRN> and The R-fMRI Network
<http://rfmri.org> (RFMRI.ORG <http://rfmri.org/>)
http://rfmri.org/yan
http://scholar.google.com/citations?user=lJQ9B58AAAAJ
