Both these modulated versions (mwrp* or m0wrp*) allready accounts for different head sizes in the data; there is no need for further corrections for head size in the statistical model (which would be eg adding TIV as a covariate), am I right?

The answer depends on the question. See http://dbm.neuro.uni-jena.de/vbm/segmentation/modulation/. If you use modulation (affine+non-linear) then you are looking at absolute volume differences and if you include TIV, then you are looking at regional volume differences. The former will show all voxels in the brain if the whole brain shrinks or expands even if the non-linear warps are 0.

Best Regards, Donald McLaren
=================
D.G. McLaren, Ph.D.
Postdoctoral Research Fellow, GRECC, Bedford VA
Research Fellow, Department of Neurology, Massachusetts General Hospital and Harvard Medical School
Office: (773) 406-2464
=====================
This e-mail contains CONFIDENTIAL INFORMATION which may contain PROTECTED HEALTHCARE INFORMATION and may also be LEGALLY PRIVILEGED and which is intended only for the use of the individual or entity named above. If the reader of the e-mail is not the intended recipient or the employee or agent responsible for delivering it to the intended recipient, you are hereby notified that you are in possession of confidential and privileged information. Any unauthorized use, disclosure, copying or the taking of any action in reliance on the contents of this information is strictly prohibited and may be unlawful. If you have received this e-mail unintentionally, please immediately notify the sender via telephone at (773) 406-2464 or email.


On Tue, Dec 21, 2010 at 1:42 PM, michel grothe <[log in to unmask]> wrote:
Dear Joao,


Suppose we use smoothed modulated non-linear only segments (ie, smoothed_mwrp* (derived from affinely aligned DARTEL exported images in VBM8, as described above in this thread) or standard smoothed_m0wrp* images of the VBM8 toolbox (also earlier described in this thread) - note that smoothing is a separate step for all options). Both these modulated versions (mwrp* or m0wrp*) allready accounts for different head sizes in the data; there is no need for further corrections for head size in the statistical model (which would be eg adding TIV as a covariate), am I right?
Yes.


In other words, building a simple two-sample t-test with one smoothed modulated non-linear only segment per subject and defining a contrast like [1 -1] (NO COVARIATES) yields brain regions with significantly higher absolute amount of tissue corrected for individual brain sizes (TIV), is this correct?
Exactly.


If using the m0wrp* images, is the best option to overlay the results on the avgT1_Dartel_IXI152.nii template provided with the VBM8 toolbox (the IXI-template, in MNI space)?
Yes, this is a good option, although I don´t know if it´s the "best". Many people prefer to overlay their (GM-) results on a structural (e.g. T1-) template, and the IXI-template only comes in tissue-types. As you suggested you can do something like this:

use the customized DARTEL_template_6 to directly warp T1 images using "DARTEL tools --> Normalise to MNI Space"

leave Voxelsize and BB as default, preserve concentration (do not modulate T1-images) and 0 smoothing (eventually 0.5 or 1mm). The VBM8-toolbox will provide you with deformation-fields (y*), which can´t be used with "DARTEL tools --> Normalise to MNI Space". In this case you can use the deformation utility (which can also be used with DARTEL-flow-fields).

Hope this helps,
Michel



 
Date: Tue, 21 Dec 2010 16:11:43 +0000
From: [log in to unmask]
Subject: [SPM] VBM8 quesiont

To: [log in to unmask]


Dear Michel and all SPMers,

for the sake of clarity:

regardless of the discussions about the pros and cons of using estimates of total brain volume, total GM volume or head size (total intracranial volume, TIV), let's think about a design matrix with two groups (eg, patients vs controls) in VBM8 toolbox (which includes high-dimensional DARTEL normalization). Suppose we use smoothed modulated non-linear only segments (ie, smoothed_mwrp* (derived from affinely aligned DARTEL exported images in VBM8, as described above in this thread) or standard smoothed_m0wrp* images of the VBM8 toolbox (also earlier described in this thread) - note that smoothing is a separate step for all options). Both these modulated versions (mwrp* or m0wrp*) allready accounts for different head sizes in the data; there is no need for further corrections for head size in the statistical model (which would be eg adding TIV as a covariate), am I right?

In other words, building a simple two-sample t-test with one smoothed modulated non-linear only segment per subject and defining a contrast like [1 -1] (NO COVARIATES) yields brain regions with significantly higher absolute amount of tissue corrected for individual brain sizes (TIV), is this correct?

One more question would relate to the results of the 2 procedures described earlier: mwrp* with DARTEL customized templates and m0wrp* with standard VBM8 Estimate & Write.
If using the m0wrp* images, is the best option to overlay the results on the avgT1_Dartel_IXI152.nii template provided with the VBM8 toolbox (the IXI-template, in MNI space)?
Despite also being in MNI space, the template of the mwrp*-images reflect the average (local) anatomy of the group. So, which template should be used to overlay the results obtained using these images? An average of the T1 images of all subjects? And how to do this?
Can I use the customized DARTEL_template_6 to directly warp T1 images using "DARTEL tools --> Normalise to MNI Space"? If so, which options for "Voxel sizes", "Bounding box", "Preserve" and "Gaussian FWHM"?

Thank you very much for your time.

Best regards,
João


On Fri, Dec 3, 2010 at 12:14 PM, michel grothe <[log in to unmask]> wrote:
Dear Joao,


So, my question would be if following the next steps you get modulated normalized images, corresponding to the output m0wrp[123]*.nii tissue classes of VBM8 Estimate & Write:

1. VBM8 Estimate & Write --> INPUTS: T1 images (in native space)
                                        --> Spatial normalization: High-dimensional DARTEL (templates are provided in MNI space, as well as spm8\toolbox\Seg\TPM.nii)
                                        --> DARTEL export: affine ----- OUTPUTS: rp*.nii tissue classes (warped to DARTEL template in MNI space)
                                        --> Modulated normalized: non-linear only ----- OUTPUTS: m0wrp[0123]*.nii tissue classes (in MNI space, including correction for individual brain sizes)

2. Run DARTEL (create Templates) --> INPUTS: rp*.nii tissue classes
                                                    --> OUTPUTS: DARTEL_Templates_0-6, flow fields u_rp1*.nii

3. DARTEL Tools Create Warped --> INPUTS: flow fields u_rp1*.nii, rp*.nii tissue classes
                                                 --> Modulation: preserve amount (modulation)
                                                 --> OUTPUTS: mwrp*.nii tissue classes (in MNI space; DARTEL_Templates_0-6 were created from rp*.nii in MNI space, and so they are also linearly MNI aligned).

Are these mwrp*.nii tissue classes corresponding to the m0wrp*.nii from the first step? As both are modulated versions, all you need to do is smoothing and stats. You don't need to add brain volume as covariate (you can still include age)... am I correct?

Yes. The only difference between your mwrp* (derived from affinely aligned DARTEL exported images in VBM8) and standard m0wrp* images of the VBM8 toolbox should be the template. While both templates are in MNI space (global size and shape), the template of the mwrp*-images reflect the average (local) anatomy of your group and the template of the m0wrp*-images (the IXI-template of the VBM8 toolbox) is based on the average (local) anatomy of a healthy adult population (I don´t know what the age-average is; you can find information on the creation of the VBM8-IXI-template in the VBM8 manual). The latter should be very close to the MNI152 template, while the former, well, depends on your studied group. That is, if you deal with populations that differ significantly from the healthy adult population (e.g. geriatric populations) than the first approach (mwrp*-images) becomes favourable. If your population is not expected to differ largely from the healthy adult population than both approaches should reveal very similar results and the deviation steps 2. and 3. of your processing pipeline might not be worth the effort. 

In Tanja's procedure (using New segment), the DARTEL Tools: Normalise to MNI space also has the option to "preserve amount (modulation)". I suppose it was not used. If one would use it, would still there be the need to add brain volume as covariate?

In my opinion (and, more importantly, the one of John Ashburner), high-dimensionally normalised images should always be modulated. Thus, I assume that Tanja has used modulation. While the volumes (voxel-values) of your VBM8-m0wrp* images are allready 'normalised' to MNI space  and variations in global head/brain size have therefore allready been accounted for (no modulation is applied at this affine transform to MNI space), (Tanja´s ??) mwrp* images (based on the normal SPM8 procedure, or VBM8 modulate affine+non-linear) reflect the native tissue volumes and therefore the statistical model has still to account for differences in global head/brain size, if these want to be removed.
In the archives, you can find numerous discussions about the pros and cons of using estimates of total brain volume, total GM volume or head size (total intracranial volume, TIV). Also, the VBM8 manual as well as Christian Gasers Web-Site have separate sections dealing with removing the confound of different brain/head-sizes.

Given that including TIV or total GM volume in the statistical model ask quite different questions to the data (and often give different results), an interesting question would be which of the two approaches the m0-approach resembles more closely. If the initial affine normalisation is estimated from the GM segment, I would assume that not modulating the so affinely transformed images would be similar to accounting for total GM in the statistical model. However, although I haven´t investigated this systematically, in my studies the m0-approach seems to lead to results more comparable to the ones obtained when accounting for TIV in the statistical model.
Maybe someone else may want to comment on this.

Best regards,
Michel

   
Date: Fri, 3 Dec 2010 10:46:43 +0000

From: [log in to unmask]
Subject: Re: [SPM] Fwd: [SPM] VBM8 quesiont
To: [log in to unmask]

Dear Michel,

thank you for your clarification.

I also noticed that Tanja referred adding the tissue volume as covariate. Shouldn't that be acounted for by the modulation?

I mean, in VBM8 you can write a modulated normalized non-linear only version of each tissue class, which produces tissue class images (m0wrp[0123]*.nii) in alignment with the template (DARTEL template in MNI space if you choose high-dimensional DARTEL normalization), that allows comparing the absolute amount of tissue corrected for individual brain sizes. This option is similar to using “Affine+non‐linear” in combination with “Global normalization” --> “Overall grand mean scaling – No” --> “Normalization – Proportional”. It is also similar to using “Affine+non‐linear” and including the numeric brain volume (as given in the *_seg8.txt file for each subject) as covariate in the statistical model (I suppose this is the option Tanja used). Although all 3 approaches allow comparing tissue volumes while correcting for individual brain size, it is recommended to use the option “non‐linear only” as it applies the correction directly to the data, rather than to the statistical model.

So, my question would be if following the next steps you get modulated normalized images, corresponding to the output m0wrp[123]*.nii tissue classes of VBM8 Estimate & Write:

1. VBM8 Estimate & Write --> INPUTS: T1 images (in native space)
                                        --> Spatial normalization: High-dimensional DARTEL (templates are provided in MNI space, as well as spm8\toolbox\Seg\TPM.nii)
                                        --> DARTEL export: affine ----- OUTPUTS: rp*.nii tissue classes (warped to DARTEL template in MNI space)
                                        --> Modulated normalized: non-linear only ----- OUTPUTS: m0wrp[0123]*.nii tissue classes (in MNI space, including correction for individual brain sizes)

2. Run DARTEL (create Templates) --> INPUTS: rp*.nii tissue classes
                                                    --> OUTPUTS: DARTEL_Templates_0-6, flow fields u_rp1*.nii

3. DARTEL Tools Create Warped --> INPUTS: flow fields u_rp1*.nii, rp*.nii tissue classes
                                                 --> Modulation: preserve amount (modulation)
                                                 --> OUTPUTS: mwrp*.nii tissue classes (in MNI space; DARTEL_Templates_0-6 were created from rp*.nii in MNI space, and so they are also linearly MNI aligned).

Are these mwrp*.nii tissue classes corresponding to the m0wrp*.nii from the first step? As both are modulated versions, all you need to do is smoothing and stats. You don't need to add brain volume as covariate (you can still include age)... am I correct?

In Tanja's procedure (using New segment), the DARTEL Tools: Normalise to MNI space also has the option to "preserve amount (modulation)". I suppose it was not used. If one would use it, would still there be the need to add brain volume as covariate?

Thanks for any help.

João



On Thu, Dec 2, 2010 at 6:07 PM, michel grothe <[log in to unmask]> wrote:
Dear Joao,


>So, my questions are: At the end, you just need to smooth the "mwrp1*-affine.nii" images and do the stats? These "mwrp1*-affine.nii", as well as the smoothed versions and the results are in MNI space?

Yes, as long as the images are written out in rp*-affine space. DARTEL 'create Template' warps the images to their average anatomy (the DARTEL-templates 1-6), and the average anatomy of x images linearly (12-parameters, affine transform) aligned to MNI space will also be in MNI space.
If you write the images as rp*-rigidly aligned to MNI space (SPM default) in the VBM8 toolbox, then your images are just rotated and translated (3 +3 = 6 parameters) to match the --orientation-- of MNI space. They are not scaled and thus do not conform to MNI space and will not after DARTEL-registration using 'create Template' and 'create warped'. These images can be warped with 'create Template' and subsequent 'Normalise to MNI space', i.e. VBM8 segmented images rigidly aligned to MNI space can be used the same way as DARTEL imported images from SPM8-New Segment (or even 'old' segment + DARTEL import).

Best,
Michel 

Date: Thu, 2 Dec 2010 17:42:14 +0000
From: [log in to unmask]

Subject: Re: [SPM] Fwd: [SPM] VBM8 quesiont
To: [log in to unmask]

Dear Michel and Tanja,

I've been following your very useful discussion.

I have a couple of related questions more:

If you use the VBM8 "Estimate & Write" module to do segmentation, you can get the DARTEL formatted "rp*-affine.nii" tissue classes. In the VBM8 manual (page 18) you can read that with the Estimate & Write module "The volumes “rp1*‐affine.nii” and “rp2*‐affine.nii” will be written. These are the grey (rp1) and white
(rp2) matter segments after an affine registration to the MNI template". After that it is only suggested to run only the Modules "Run DARTEL (create Templates)" with "rp*-affine.nii" input, and finally the "Create warped" with flow fields "u_*.nii" and "rp*-affine.nii" inputs. This "Create warped" module has the option to "Preserve Amount (Modulation)".
Also in the beggining of this page you can read "This option uses affine registered GM and WM segments and modulates the segmentations for nonlinear
effects only, which accounts for brain size. No additional normalization to MNI space is
necessary, because the DARTEL template is provided in that space."

So, my questions are: At the end, you just need to smooth the "mwrp1*-affine.nii" images and do the stats? These "mwrp1*-affine.nii", as well as the smoothed versions and the results are in MNI space?

Thanks for your help,

João Duarte



On Thu, Dec 2, 2010 at 3:31 PM, michel grothe <[log in to unmask]> wrote:
Dear Tanja,
1. An absolute Threshold of 0.2 is totally acceptable and much more common than 0.01. Alternatively you can apply a brainmask for analysis (e.g. see Ridgway et al, 2009; http://www.ncbi.nlm.nih.gov/pubmed/18848632).
2. DARTEL 'Create Template' high-dimensionally registers your subjects to their average-anatomy, represented by Template_6. This template will most probably differ from MNI space (Normal brains are on average smaller than the MNI-template). This is why DARTEL 'Normalise to MNI space' estimates an (12-parameter) affine transform from Template_6 to MNI space and combines this transform with the individual DARTEL flow-fields. The scheme is:
Native --DARTEL--> Template_6 --affine--> MNI Template, where the last transform is the same for each subject and aims to model global brain size (and shape) differences between the average anatomy of your specific group and the MNI Template.
Your Template_6 will not be transformed to MNI space, which is why you see the disagreement with your results, which are derived from your warped tissue-classes in MNI space.
 
The best would be to overlay your results on a template based on the images you used for the stats, that is an average of your warped tissue-classes (or warped structurals as explained in my previous mail). You can use imcalc for this: Expression: mean(X); Options: Yes, read images in data matrix (--> X). This template can be considered something like 'average anatomy of the group in MNI space' and best matches the space of your results.

BTW, the avgT1_Dartel_IXI550_MNI152.nii is specific to the VBM8-toolbox and is not an official MNI template (although given that it is based on an average of the normal population aligned to MNI space, it should be very similar to MNI152, which is the most widely used official template of MNI space).

Hope this helps,
Michel

> Date: Thu, 2 Dec 2010 14:42:28 +0100> Subject: Re: [SPM] Fwd: [SPM] VBM8 quesiont

> To: [log in to unmask]
>
> Dear all,
>
> I also wanted to note that for Factorial design specification I use
> Threshold masking -> absolute threshold 0.01. I tried higher values,
> it gives better results if threshold is 0.2 (but only while
> overlapping the statistics on any MNI template, not Template_6.nii).
>
> Therefore, I have 2 questions.
>
> 1. Is threshold of 0.2 considered too high or acceptable?
> 2. The Template_6.nii is smaler than any MNI template, so is it or is
> it not normalized to MNI at step "Run DARTEL (Create Templates)"?
>
> Thank you,
> Tanja.
>
>
> On Thu, Dec 2, 2010 at 1:17 PM, Tetiana Dadakova <[log in to unmask]> wrote:
> > Dear Michel,
> >
> > thank you for your reply.
> >
> > In attachment you can see statistics overlaid on template
> > avgT1_Dartel_IXI550_MNI152.nii. It is better, but still not good
> > enough. (Is it the MNI template you mentioned or is there other
> > standard MNI template in SPM?)
> >
> > I am a bit confused about the Template6.nii. It is written in VBM
> > Tutorial by John Ashburner (this one
> > http://www.fil.ion.ucl.ac.uk/~john/misc/VBMclass10.pdf) that
> > "...Template 6.nii. This template is registered to MNI space (affine
> > transform), allowing the transformations to be combined so that all
> > the individual spatially normalised scans can also be brought into MNI
> > space."
> > Therefore, if I understand correctly, Template6.nii is in MNI space, right?
> >
> > Thanks again, I appreciate your help,
> > Tanja.
> >
> >
> >
> >
> > On Thu, Dec 2, 2010 at 12:30 PM, michel grothe <[log in to unmask]> wrote:
> >> Dear Tanja,
> >> I suppose that the template in the attached image is your DARTEL-Template6.
> >> This would explain the mismatch of the overlayed results, which are in MNI
> >> space (The MNI template brain is a bit bigger than average anatomy). Try
> >> projecting them on a standard MNI-template or create your
> >> 'group-specific'-MNI-template by warping and averaging the structurals using
> >> 'Normalize to MNI space' (input: u_rc1*, T1-structurals, Template6; no
> >> modulation, no or only small smoothing).
> >>
> >> Hope this helps,
> >> Michel
> >>
> >>> Date: Wed, 1 Dec 2010 20:38:58 +0100
> >>> From: [log in to unmask]
> >>> Subject: [SPM] Fwd: [SPM] VBM8 quesiont
> >>> To: [log in to unmask]
> >>>
> >>> Dear Michel,
> >>>
> >>> thank you for your answer. I tried to do just 3 steps, according to
> >>> your mail. It looks a little better, but still something is wrong with
> >>> template.
> >>>
> >>> In the attachment you can see the statistics overlaid on the
> >>> Template6.nii.
> >>>
> >>> Do you have any idea what can be the problem, at which step can I make
> >>> a mistake?
> >>>
> >>> The steps:
> >>> 0. Reorient all the images to AC-PC line (most of images are .nii, but
> >>> some are .img/.hdr)
> >>> 1. Segmentation: Tools -> New Segment
> >>> 2. DARTEL Tools -> Run DARTEL (Create Template) (input: rc1*.nii,
> >>> rc2*.nii, rc3*.nii)
> >>> 3. DARTEL Tools -> Normalize to MNI space (input: u_rc1*.nii,
> >>> Template6.nii)
> >>> 4. Factorial design specification (2 sample t-test, covariates: age,
> >>> tissue volume)
> >>>
> >>> Thank you,
> >>> Tanja.
> >>>
> >>>
> >>> On Fri, Nov 26, 2010 at 9:26 PM, michel grothe <[log in to unmask]>
> >>> wrote:
> >>> > Dear Tanja,
> >>> > there is a problem wit your processing steps. You´d normally directly
> >>> > warp
> >>> > the images to MNI space, i.e. omit steps 3 and 4. The "Normalize to MNI
> >>> > space" option in SPM8 estimates an affine transform from the DARTEL
> >>> > template
> >>> > (specify your DARTEL-template_6 here) to the MNI-template and combines
> >>> > it
> >>> > with the individual DARTEL flow-fields to warp each subject directly
> >>> > from
> >>> > native space to MNI-space (better "average anatomy of your sample in MNI
> >>> > space"). Step 3 and 4 had to be used in SPM5 but are not necessary
> >>> > anymore
> >>> > in SPM8. The "create warped"-option is only for the case you don´t want
> >>> > to
> >>> > normalise to MNI space (but to the pure average anatomy of your sample).
> >>> >
> >>> > Best,
> >>> > Michel
> >>> >
> >>> >> Date: Fri, 26 Nov 2010 16:33:18 +0100
> >>> >> From: [log in to unmask]
> >>> >> Subject: [SPM] VBM8 quesiont
> >>> >> To: [log in to unmask]
> >>> >>
> >>> >> Dear SPM list,
> >>> >>
> >>> >> I have a problem with VBM8 data processing. When I try to overlay
> >>> >> statistics on the template (wTemplate6.nii), it seems that the
> >>> >> template is either shifted or the size of the template doesn't
> >>> >> correspond (the significant clusters are supposed to be in gray matter
> >>> >> region, see attached image).
> >>> >>
> >>> >> 1. Could you please explain to me what can be the problem? Can it be
> >>> >> the problem of wrong template generation or of statistical
> >>> >> calculations?
> >>> >> 2. At which of the following step is the smoothing done?
> >>> >>
> >>> >> Here are the steps I take
> >>> >> 1. Segmentation with Tools -> New Segment
> >>> >> 2. Creating DARTEL template Template6.nii with Tools -> DARTEL Tools
> >>> >> -> Run DARTEL (create Templates)
> >>> >> 3. Normalizing Template6.nii to MNI with Spatial -> Normalise:
> >>> >> Estimate & Write (get wTemplate6.nii)
> >>> >> 4. Warping segmented images with Tools -> DARTEL Tools -> Create Warped
> >>> >> 5. Normalizing warped images to MNI with Tools -> DARTEL Tools ->
> >>> >> Normalize to MNI Space (using wTemplate6.nii here)
> >>> >>
> >>> >> Thank you for your time,
> >>> >> Tanja.
> >>> >
> >>
> >