Re: [Freesurfer] FSFAST mkcontrast-sess with spmhrf 2 question

[Douglas N Greve]

On 03/24/2016 07:37 PM, Joseph Dien wrote:
> Doug, I see now what your concern was with just adjusting the beta 
> weights.  Since FSFAST is using a pseudo-mixed effects model, there is 
> also a need to pass the cesvar statistics up to the second level.  How 
> that might accommodate the derivative boost computation is not 
> straightforward to me.  I need to understand better what fsfast is 
> doing.  I was wondering if you could answer a couple questions about 
> how cesvar is computed;
>
> 1) The cesvar computations in the fast_fratio.m function provided a 
> good guide.  According to it:
> cescvm = inv(C*inv(X'*X)*X'*Sn*X*inv(X'*X)*C’);
> where Sn is the covariance matrix of the noise after the whitening 
> (which is on by default).
> cesvar=rvar./cescvm.
>
> How is Sn computed?
The autocorrelation function is computed at each voxel from the 
residual, these are fit to an AR1 model and adjusted for residual bias. 
They are spatially clustered based on AR1 value. The Sn is computed as a 
toeplitz matrix of the autocorrelation function derived from the 
best-fit AR1 model (that matrix is multipled by the residual variance to 
get the final Sn).
>
> 2) I need to understand what fsfast is doing conceptually but I 
> couldn’t reconcile the Thirion et al (2007) Equation #5 and the 
> equations above, specifically the inclusion of the rvar term.  Can you 
> throw some light on this?
It is really nothing fancy, just down-weighting each subject by the 
residual standard deviation.
>
> I greatly appreciate your helpful tips and this amazing software that 
> you have written.
>
> Respectfully,
>
> Joe
>
>
>> On Mar 17, 2016, at 15:48, Joseph Dien > > wrote:
>>
>> I did more digging around and came up with a procedure.  Please let 
>> me know if it would cause any problems.
>> Looking at the contents of the X.mat files (which contain the 
>> predictors), it appears that the betas are indeed arranged as c1 d1 
>> c2 d2…
>> I also found that isxconcat-sess requires the contrast output 
>> of mkcontrast-sess so I can’t just bypass it.
>>
>> So my thought is:
>>
>> 1) run mkanalysis-sess with -spmhrf 1
>> 2) run selxavg3-sess with the -no-con-ok flag to get the spmhrf1 
>> betas without bothering with the contrasts
>> 3) run mkanalysis-sess with -spmhrf 0
>> 4) run selxavg3-sess again but with the contrasts to get everything 
>> set up
>> 5) read in the spmhrf1 betas with MRIread and compute the Calhoun 
>> derivative boost with the beta values in mri.vol
>> 6) use these values to compute the desired contrasts.  From my 
>> examination of a sample ces.nii.gz file, a simple linear combination 
>> based on the contrast weights is all that is needed.
>> 7) replace the contents of the corresponding spmhrf0 ces.nii.gz 
>> files’ mri.vol with these new values
>> 8) write out the new ces.nii.gz files.
>> 9) proceed with the analysis stream, using isxconcat-sess to set up 
>> the second level analysis
>>
>> So the question is, would this work?  Am I neglecting files other 
>> than ces.nii.gz that would need to be modified?  or other fields in 
>> the mri data structure?
>> When you referred to writing “out a new volume” is this what you meant?
>>
>> Thanks again for this help!
>>
>> Joe
>>
>>
>>> On Mar 15, 2016, at 17:32, Joseph Dien >> > wrote:
>>>
>>> oh duh!  Sorry, wasn’t thinking clearly.
>>> Okay, I see how to generate the betas now.  I don’t even need to 
>>> mess with the mkcontrast-sess command.
>>> I just run selxavg3-sess with the -no-con-ok flag.
>>> With spmhrf 0 I generated a beta.nii.gz file with 85 betas in each 
>>> vertex.
>>> With spmhrf 1 I generated a beta.nii.gz file with 97 betas in each 
>>> vertex.
>>> With 12 conditions, 12 more betas is exactly right.
>>> So how do I know which of the betas is which?
>>> I looked through the files and couldn’t find any labels.
>>> I’m guessing the first twelve are the condition betas for "spmhrf 0".
>>> If so, for "spmhrf 1", is it arranged as:
>>>
>>> 1) c1 c2 c3…d1 d2 d3...
>>>
>>> or
>>>
>>> 2) c1 d1 c2 d2...
>>>
>>>  (where d is the first derivative term)
>>>
>>> also, when I asked you earlier about implementing this procedure, I 
>>> had suggested reading the betas, computing the Calhoun, then 
>>> generating new beta.nii.gz files with the new betas replacing the 
>>> original spmhrf0 betas and then continuing with the regular analysis 
>>> stream but you said it would result in invalid p-values.  Instead, 
>>> you suggested:
>>>
>>> "I was just thinking you could load the beta into matlab, make the
>>> Calhoun computations on each condition, then compute the contrasts, then
>>> write out the new volume”
>>>
>>> can you expand on how one might “compute the contrasts” and “write 
>>> out the new volume”?
>>>
>>> Thanks again for this help!
>>>
>>> Joe
>>>
>>>
 On Mar 15, 2016, at 12:43, Douglas N Greve 
 

Re: [Freesurfer] FSFAST mkcontrast-sess with spmhrf 2 question

[Joseph Dien]

Doug, I see now what your concern was with just adjusting the beta weights.  
Since FSFAST is using a pseudo-mixed effects model, there is also a need to 
pass the cesvar statistics up to the second level.  How that might accommodate 
the derivative boost computation is not straightforward to me.  I need to 
understand better what fsfast is doing.  I was wondering if you could answer a 
couple questions about how cesvar is computed;

1) The cesvar computations in the fast_fratio.m function provided a good guide. 
 According to it:
cescvm = inv(C*inv(X'*X)*X'*Sn*X*inv(X'*X)*C’);
where Sn is the covariance matrix of the noise after the whitening (which is on 
by default).
cesvar=rvar./cescvm.

How is Sn computed?

2) I need to understand what fsfast is doing conceptually but I couldn’t 
reconcile the Thirion et al (2007) Equation #5 and the equations above, 
specifically the inclusion of the rvar term.  Can you throw some light on this?

I greatly appreciate your helpful tips and this amazing software that you have 
written.

Respectfully,

Joe


> On Mar 17, 2016, at 15:48, Joseph Dien  wrote:
> 
> I did more digging around and came up with a procedure.  Please let me know 
> if it would cause any problems.
> Looking at the contents of the X.mat files (which contain the predictors), it 
> appears that the betas are indeed arranged as c1 d1 c2 d2…
> I also found that isxconcat-sess requires the contrast output of 
> mkcontrast-sess so I can’t just bypass it.
> 
> So my thought is:
> 
> 1) run mkanalysis-sess with -spmhrf 1
> 2) run selxavg3-sess with the -no-con-ok flag to get the spmhrf1 betas 
> without bothering with the contrasts
> 3) run mkanalysis-sess with -spmhrf 0
> 4) run selxavg3-sess again but with the contrasts to get everything set up
> 5) read in the spmhrf1 betas with MRIread and compute the Calhoun derivative 
> boost with the beta values in mri.vol
> 6) use these values to compute the desired contrasts.  From my examination of 
> a sample ces.nii.gz file, a simple linear combination based on the contrast 
> weights is all that is needed.
> 7) replace the contents of the corresponding spmhrf0 ces.nii.gz files’ 
> mri.vol with these new values
> 8) write out the new ces.nii.gz files.
> 9) proceed with the analysis stream, using isxconcat-sess to set up the 
> second level analysis
> 
> So the question is, would this work?  Am I neglecting files other than 
> ces.nii.gz that would need to be modified?  or other fields in the mri data 
> structure?
> When you referred to writing “out a new volume” is this what you meant?
> 
> Thanks again for this help!
> 
> Joe
> 
> 
>> On Mar 15, 2016, at 17:32, Joseph Dien > > wrote:
>> 
>> oh duh!  Sorry, wasn’t thinking clearly.
>> Okay, I see how to generate the betas now.  I don’t even need to mess with 
>> the mkcontrast-sess command.
>> I just run selxavg3-sess with the -no-con-ok flag.
>> With spmhrf 0 I generated a beta.nii.gz file with 85 betas in each vertex.
>> With spmhrf 1 I generated a beta.nii.gz file with 97 betas in each vertex.
>> With 12 conditions, 12 more betas is exactly right.
>> So how do I know which of the betas is which?
>> I looked through the files and couldn’t find any labels.
>> I’m guessing the first twelve are the condition betas for "spmhrf 0".
>> If so, for "spmhrf 1", is it arranged as:
>> 
>> 1) c1 c2 c3…d1 d2 d3...
>> 
>> or
>> 
>> 2) c1 d1 c2 d2...
>> 
>>  (where d is the first derivative term)
>> 
>> also, when I asked you earlier about implementing this procedure, I had 
>> suggested reading the betas, computing the Calhoun, then generating new 
>> beta.nii.gz files with the new betas replacing the original spmhrf0 betas 
>> and then continuing with the regular analysis stream but you said it would 
>> result in invalid p-values.  Instead, you suggested:
>> 
>> "I was just thinking you could load the beta into matlab, make the
>> Calhoun computations on each condition, then compute the contrasts, then
>> write out the new volume”
>> 
>> can you expand on how one might “compute the contrasts” and “write out the 
>> new volume”?
>> 
>> Thanks again for this help!
>> 
>> Joe
>> 
>> 
>>> On Mar 15, 2016, at 12:43, Douglas N Greve >> > wrote:
>>> 
>>> The setwdelay is an option for mkcontrast-sess (not mkanalysis-sess)
>>> 
>>> On 03/13/2016 10:29 PM, Joseph Dien wrote:
 After a long break, back to this…
 
 My goal is still to get the betas for the first and maybe second spm 
 hrf so I can calculate a Calhoun derivative boost measure.
 
 As a first step I ran:
 
 mkanalysis-sess -fsd bold -analysis RPA.sm05.lh -surface fsaverage lh 
 -fwhm 5 -event-related  -paradigm RPA1fix.par -nconditions 12 -spmhrf 
 1 -TR 2 -refeventdur .25 -polyfit 2 -per-run -force -nuisreg 
 nuisreg.dat 6 -tpexclude tpexclude.dat -b0dc
 
 to use the SPM HRF with one 

Re: [Freesurfer] FSFAST mkcontrast-sess with spmhrf 2 question

[Joseph Dien]

I did more digging around and came up with a procedure.  Please let me know if 
it would cause any problems.
Looking at the contents of the X.mat files (which contain the predictors), it 
appears that the betas are indeed arranged as c1 d1 c2 d2…
I also found that isxconcat-sess requires the contrast output of 
mkcontrast-sess so I can’t just bypass it.

So my thought is:

1) run mkanalysis-sess with -spmhrf 1
2) run selxavg3-sess with the -no-con-ok flag to get the spmhrf1 betas without 
bothering with the contrasts
3) run mkanalysis-sess with -spmhrf 0
4) run selxavg3-sess again but with the contrasts to get everything set up
5) read in the spmhrf1 betas with MRIread and compute the Calhoun derivative 
boost with the beta values in mri.vol
6) use these values to compute the desired contrasts.  From my examination of a 
sample ces.nii.gz file, a simple linear combination based on the contrast 
weights is all that is needed.
7) replace the contents of the corresponding spmhrf0 ces.nii.gz files’ mri.vol 
with these new values
8) write out the new ces.nii.gz files.
9) proceed with the analysis stream, using isxconcat-sess to set up the second 
level analysis

So the question is, would this work?  Am I neglecting files other than 
ces.nii.gz that would need to be modified?  or other fields in the mri data 
structure?
When you referred to writing “out a new volume” is this what you meant?

Thanks again for this help!

Joe


> On Mar 15, 2016, at 17:32, Joseph Dien  wrote:
> 
> oh duh!  Sorry, wasn’t thinking clearly.
> Okay, I see how to generate the betas now.  I don’t even need to mess with 
> the mkcontrast-sess command.
> I just run selxavg3-sess with the -no-con-ok flag.
> With spmhrf 0 I generated a beta.nii.gz file with 85 betas in each vertex.
> With spmhrf 1 I generated a beta.nii.gz file with 97 betas in each vertex.
> With 12 conditions, 12 more betas is exactly right.
> So how do I know which of the betas is which?
> I looked through the files and couldn’t find any labels.
> I’m guessing the first twelve are the condition betas for "spmhrf 0".
> If so, for "spmhrf 1", is it arranged as:
> 
> 1) c1 c2 c3…d1 d2 d3...
> 
> or
> 
> 2) c1 d1 c2 d2...
> 
>  (where d is the first derivative term)
> 
> also, when I asked you earlier about implementing this procedure, I had 
> suggested reading the betas, computing the Calhoun, then generating new 
> beta.nii.gz files with the new betas replacing the original spmhrf0 betas and 
> then continuing with the regular analysis stream but you said it would result 
> in invalid p-values.  Instead, you suggested:
> 
> "I was just thinking you could load the beta into matlab, make the
> Calhoun computations on each condition, then compute the contrasts, then
> write out the new volume”
> 
> can you expand on how one might “compute the contrasts” and “write out the 
> new volume”?
> 
> Thanks again for this help!
> 
> Joe
> 
> 
>> On Mar 15, 2016, at 12:43, Douglas N Greve > > wrote:
>> 
>> The setwdelay is an option for mkcontrast-sess (not mkanalysis-sess)
>> 
>> On 03/13/2016 10:29 PM, Joseph Dien wrote:
>>> After a long break, back to this…
>>> 
>>> My goal is still to get the betas for the first and maybe second spm 
>>> hrf so I can calculate a Calhoun derivative boost measure.
>>> 
>>> As a first step I ran:
>>> 
>>> mkanalysis-sess -fsd bold -analysis RPA.sm05.lh -surface fsaverage lh 
>>> -fwhm 5 -event-related  -paradigm RPA1fix.par -nconditions 12 -spmhrf 
>>> 1 -TR 2 -refeventdur .25 -polyfit 2 -per-run -force -nuisreg 
>>> nuisreg.dat 6 -tpexclude tpexclude.dat -b0dc
>>> 
>>> to use the SPM HRF with one derivative.
>>> 
>>> Then:
>>> 
>>> selxavg3-sess -sf sessidlistALL.dat -analysis RPA.sm05.lh
>>> 
>>> I got the following error:
>>> 
>>> ERROR: flac_evconw(): conVinc, Condition01 evrw dim mismatch
>>> This condition has 2 regressors, but evrm has 1
>>> Struct contents reference from a non-struct array object.
>>> 
>>> Error in flac_conmat (line 37)
>>> if(nthcon > length(flac.con))
>>> 
>>> Error in flac_customize (line 369)
>>>  flacnew = flac_conmat(flacnew,nthcon);
>>> 
>>> Error in fast_selxavg3 (line 65)
>>> flac0 = flac_customize(flac0);
>>> 
> --
>>> ERROR: fast_selxavg3() failed\n
>>> 
>>> This ran fine with spmhrf 0
>>> 
>>> based on your prior response below that:
>>> 
>>> "This is what happens. If you want to use the derivatives,
>>> then you need to spec -setwdelay. When you run the command, it will 
>>> prompt you
>>> for 3 values to use. If you spec 1 0 0, then it will be the same as the
>>> default. If you want to test only the first derivative, then 
>>> you would spec 0 1 0. Note that the 3rd regressor is the 2nd 
>>> derivative wrt time, not the first derivative wrt the dispersion 
>>> parameter. You
>>> cannot get the Calhoun 2004 value using a contrast (it is non-linear).”
>>> 
>>> I tried:
>>> 
>>> 

Re: [Freesurfer] FSFAST mkcontrast-sess with spmhrf 2 question

[Joseph Dien]

oh duh!  Sorry, wasn’t thinking clearly.
Okay, I see how to generate the betas now.  I don’t even need to mess with the 
mkcontrast-sess command.
I just run selxavg3-sess with the -no-con-ok flag.
With spmhrf 0 I generated a beta.nii.gz file with 85 betas in each vertex.
With spmhrf 1 I generated a beta.nii.gz file with 97 betas in each vertex.
With 12 conditions, 12 more betas is exactly right.
So how do I know which of the betas is which?
I looked through the files and couldn’t find any labels.
I’m guessing the first twelve are the condition betas for "spmhrf 0".
If so, for "spmhrf 1", is it arranged as:

1) c1 c2 c3…d1 d2 d3...

or

2) c1 d1 c2 d2...

 (where d is the first derivative term)

also, when I asked you earlier about implementing this procedure, I had 
suggested reading the betas, computing the Calhoun, then generating new 
beta.nii.gz files with the new betas replacing the original spmhrf0 betas and 
then continuing with the regular analysis stream but you said it would result 
in invalid p-values.  Instead, you suggested:

"I was just thinking you could load the beta into matlab, make the
Calhoun computations on each condition, then compute the contrasts, then
write out the new volume”

can you expand on how one might “compute the contrasts” and “write out the new 
volume”?

Thanks again for this help!

Joe


> On Mar 15, 2016, at 12:43, Douglas N Greve  wrote:
> 
> The setwdelay is an option for mkcontrast-sess (not mkanalysis-sess)
> 
> On 03/13/2016 10:29 PM, Joseph Dien wrote:
>> After a long break, back to this…
>> 
>> My goal is still to get the betas for the first and maybe second spm 
>> hrf so I can calculate a Calhoun derivative boost measure.
>> 
>> As a first step I ran:
>> 
>> mkanalysis-sess -fsd bold -analysis RPA.sm05.lh -surface fsaverage lh 
>> -fwhm 5 -event-related  -paradigm RPA1fix.par -nconditions 12 -spmhrf 
>> 1 -TR 2 -refeventdur .25 -polyfit 2 -per-run -force -nuisreg 
>> nuisreg.dat 6 -tpexclude tpexclude.dat -b0dc
>> 
>> to use the SPM HRF with one derivative.
>> 
>> Then:
>> 
>> selxavg3-sess -sf sessidlistALL.dat -analysis RPA.sm05.lh
>> 
>> I got the following error:
>> 
>> ERROR: flac_evconw(): conVinc, Condition01 evrw dim mismatch
>> This condition has 2 regressors, but evrm has 1
>> Struct contents reference from a non-struct array object.
>> 
>> Error in flac_conmat (line 37)
>> if(nthcon > length(flac.con))
>> 
>> Error in flac_customize (line 369)
>>  flacnew = flac_conmat(flacnew,nthcon);
>> 
>> Error in fast_selxavg3 (line 65)
>> flac0 = flac_customize(flac0);
>> 
 --
>> ERROR: fast_selxavg3() failed\n
>> 
>> This ran fine with spmhrf 0
>> 
>> based on your prior response below that:
>> 
>> "This is what happens. If you want to use the derivatives,
>> then you need to spec -setwdelay. When you run the command, it will 
>> prompt you
>> for 3 values to use. If you spec 1 0 0, then it will be the same as the
>> default. If you want to test only the first derivative, then 
>> you would spec 0 1 0. Note that the 3rd regressor is the 2nd 
>> derivative wrt time, not the first derivative wrt the dispersion 
>> parameter. You
>> cannot get the Calhoun 2004 value using a contrast (it is non-linear).”
>> 
>> I tried:
>> 
>> mkanalysis-sess -fsd bold -analysis RPA.sm05.lh -surface fsaverage lh 
>> -fwhm 5 -event-related  -paradigm RPA1fix.par -nconditions 12 -spmhrf 
>> 1 -TR 2 -refeventdur .25 -polyfit 2 -per-run -force -nuisreg 
>> nuisreg.dat 6 -tpexclude tpexclude.dat -b0dc -setwdelay
>> 
>> but I just got the error:
>> 
>> ERROR: Flag -setwdelay unrecognized.
>> 
>> I also tried using the flag when setting up the contrasts but that 
>> didn’t work either.
>> 
>> so not quite sure what I’m doing wrong.  I had understood from your 
>> response below that the contrast weights didn’t need to be changed 
>> from the spmhrf 0 case but perhaps I misunderstood?
>> 
>> any help would be welcome!
>> 
>> Joe
>> 
>>> On Jul 10, 2013, at 13:58, Douglas N Greve >> >> 
>>> wrote:
>>> 
>>> 
>>> I was just thinking you could load the beta into matlab, make the
>>> Calhoun computations on each condition, then compute the contrasts, then
>>> write out the new volume
>>> 
>>> 
>>> On 07/10/2013 01:40 PM, Joseph Dien wrote:
 Sounds good!
 
 Regarding creating a new volume and computing contrasts from it, what
 do you mean?  I didn't follow that.
 
 Thanks!
 
 Joe
 
 
 On Jul 10, 2013, at 1:33 PM, Douglas N Greve
  
 >>> wrote:
 
> 
> On 07/10/2013 01:29 PM, Joseph Dien wrote:
>> Sorry, not following what you are suggesting?
>> 
>> 

Re: [Freesurfer] FSFAST mkcontrast-sess with spmhrf 2 question

[Douglas N Greve]

The setwdelay is an option for mkcontrast-sess (not mkanalysis-sess)

On 03/13/2016 10:29 PM, Joseph Dien wrote:
> After a long break, back to this…
>
> My goal is still to get the betas for the first and maybe second spm 
> hrf so I can calculate a Calhoun derivative boost measure.
>
> As a first step I ran:
>
> mkanalysis-sess -fsd bold -analysis RPA.sm05.lh -surface fsaverage lh 
> -fwhm 5 -event-related  -paradigm RPA1fix.par -nconditions 12 -spmhrf 
> 1 -TR 2 -refeventdur .25 -polyfit 2 -per-run -force -nuisreg 
> nuisreg.dat 6 -tpexclude tpexclude.dat -b0dc
>
> to use the SPM HRF with one derivative.
>
> Then:
>
> selxavg3-sess -sf sessidlistALL.dat -analysis RPA.sm05.lh
>
> I got the following error:
>
> ERROR: flac_evconw(): conVinc, Condition01 evrw dim mismatch
> This condition has 2 regressors, but evrm has 1
> Struct contents reference from a non-struct array object.
>
> Error in flac_conmat (line 37)
> if(nthcon > length(flac.con))
>
> Error in flac_customize (line 369)
>   flacnew = flac_conmat(flacnew,nthcon);
>
> Error in fast_selxavg3 (line 65)
> flac0 = flac_customize(flac0);
>
> >> --
> ERROR: fast_selxavg3() failed\n
>
> This ran fine with spmhrf 0
>
> based on your prior response below that:
>
> "This is what happens. If you want to use the derivatives,
> then you need to spec -setwdelay. When you run the command, it will 
> prompt you
> for 3 values to use. If you spec 1 0 0, then it will be the same as the
> default. If you want to test only the first derivative, then 
> you would spec 0 1 0. Note that the 3rd regressor is the 2nd 
> derivative wrt time, not the first derivative wrt the dispersion 
> parameter. You
> cannot get the Calhoun 2004 value using a contrast (it is non-linear).”
>
> I tried:
>
> mkanalysis-sess -fsd bold -analysis RPA.sm05.lh -surface fsaverage lh 
> -fwhm 5 -event-related  -paradigm RPA1fix.par -nconditions 12 -spmhrf 
> 1 -TR 2 -refeventdur .25 -polyfit 2 -per-run -force -nuisreg 
> nuisreg.dat 6 -tpexclude tpexclude.dat -b0dc -setwdelay
>
> but I just got the error:
>
> ERROR: Flag -setwdelay unrecognized.
>
> I also tried using the flag when setting up the contrasts but that 
> didn’t work either.
>
> so not quite sure what I’m doing wrong.  I had understood from your 
> response below that the contrast weights didn’t need to be changed 
> from the spmhrf 0 case but perhaps I misunderstood?
>
> any help would be welcome!
>
> Joe
>
>> On Jul 10, 2013, at 13:58, Douglas N Greve > > wrote:
>>
>>
>> I was just thinking you could load the beta into matlab, make the
>> Calhoun computations on each condition, then compute the contrasts, then
>> write out the new volume
>>
>>
>> On 07/10/2013 01:40 PM, Joseph Dien wrote:
>>> Sounds good!
>>>
>>> Regarding creating a new volume and computing contrasts from it, what
>>> do you mean?  I didn't follow that.
>>>
>>> Thanks!
>>>
>>> Joe
>>>
>>>
>>> On Jul 10, 2013, at 1:33 PM, Douglas N Greve
>>> >> > wrote:
>>>

 On 07/10/2013 01:29 PM, Joseph Dien wrote:
> Sorry, not following what you are suggesting?
>
> I want the second derivative for calculating the Calhoun et al 2004
> derivative boost measure.
> My understanding is that to the extent that the BOLD signal deviates
> from the canonical hrf, the amplitude of the primary regressor will be
> attenuated and the variance will instead end up in the first and
> second derivatives (to the extent that they are able to accommodate
> the divergence).  By using a Calhoun measure that incorporates both
> the first and second derivatives, in principle I'll have a BOLD
> measure that is more robust to deviations from the canonical hrf.
 Sorry, it had been a while since I read that paper. I did not know that
 they had a formulation that included the 2nd derivative.
>
> However, if the way FSFAST is calculating the second derivative
> regressor is resulting in loss of statistical power due to shared
> variance with the primary regressor, then it would be best to just not
> include it at all in the estimation step.
 I don't know how much it will hurt the power. You'd have to look at the
 efficiency.
 doug

>
>
> On Jul 10, 2013, at 1:21 PM, Douglas N Greve
>  
> > wrote:
>
>>
>> Why not just create a new volume and then compute contrasts from
>> the new
>> volume? What you are suggesting will work I think, but it leaves me a
>> little nervous. The p-values will be meaningless.
>>
>>
>> As for the 2nd derivative, I think it must be a numerical issue 
>> (it is
>> not computed analytically). Why 

Re: [Freesurfer] FSFAST mkcontrast-sess with spmhrf 2 question

[Joseph Dien]

After a long break, back to this…

My goal is still to get the betas for the first and maybe second spm hrf so I 
can calculate a Calhoun derivative boost measure.

As a first step I ran:

mkanalysis-sess -fsd bold -analysis RPA.sm05.lh -surface fsaverage lh -fwhm 5 
-event-related  -paradigm RPA1fix.par -nconditions 12 -spmhrf 1 -TR 2 
-refeventdur .25 -polyfit 2 -per-run -force -nuisreg nuisreg.dat 6 -tpexclude 
tpexclude.dat -b0dc

to use the SPM HRF with one derivative.

Then:

selxavg3-sess -sf sessidlistALL.dat -analysis RPA.sm05.lh

I got the following error:

ERROR: flac_evconw(): conVinc, Condition01 evrw dim mismatch
This condition has 2 regressors, but evrm has 1
Struct contents reference from a non-struct array object.

Error in flac_conmat (line 37)
if(nthcon > length(flac.con))

Error in flac_customize (line 369)
  flacnew = flac_conmat(flacnew,nthcon);

Error in fast_selxavg3 (line 65)
flac0 = flac_customize(flac0);
 
>> --
ERROR: fast_selxavg3() failed\n

This ran fine with spmhrf 0

based on your prior response below that:

"This is what happens. If you want to use the derivatives, 
then you need to spec -setwdelay. When you run the command, it will prompt you
for 3 values to use. If you spec 1 0 0, then it will be the same as the
default. If you want to test only the first derivative, then you would spec 0 1 
0. Note that the 3rd regressor is the 2nd derivative wrt time, not the first 
derivative wrt the dispersion parameter. You
cannot get the Calhoun 2004 value using a contrast (it is non-linear).”

I tried:

mkanalysis-sess -fsd bold -analysis RPA.sm05.lh -surface fsaverage lh -fwhm 5 
-event-related  -paradigm RPA1fix.par -nconditions 12 -spmhrf 1 -TR 2 
-refeventdur .25 -polyfit 2 -per-run -force -nuisreg nuisreg.dat 6 -tpexclude 
tpexclude.dat -b0dc -setwdelay

but I just got the error:

ERROR: Flag -setwdelay unrecognized.

I also tried using the flag when setting up the contrasts but that didn’t work 
either.

so not quite sure what I’m doing wrong.  I had understood from your response 
below that the contrast weights didn’t need to be changed from the spmhrf 0 
case but perhaps I misunderstood?

any help would be welcome!

Joe

> On Jul 10, 2013, at 13:58, Douglas N Greve  wrote:
> 
> 
> I was just thinking you could load the beta into matlab, make the 
> Calhoun computations on each condition, then compute the contrasts, then 
> write out the new volume
> 
> 
> On 07/10/2013 01:40 PM, Joseph Dien wrote:
>> Sounds good!
>> 
>> Regarding creating a new volume and computing contrasts from it, what 
>> do you mean?  I didn't follow that.
>> 
>> Thanks!
>> 
>> Joe
>> 
>> 
>> On Jul 10, 2013, at 1:33 PM, Douglas N Greve 
>>  
>> >> wrote:
>> 
>>> 
>>> On 07/10/2013 01:29 PM, Joseph Dien wrote:
 Sorry, not following what you are suggesting?
 
 I want the second derivative for calculating the Calhoun et al 2004
 derivative boost measure.
 My understanding is that to the extent that the BOLD signal deviates
 from the canonical hrf, the amplitude of the primary regressor will be
 attenuated and the variance will instead end up in the first and
 second derivatives (to the extent that they are able to accommodate
 the divergence).  By using a Calhoun measure that incorporates both
 the first and second derivatives, in principle I'll have a BOLD
 measure that is more robust to deviations from the canonical hrf.
>>> Sorry, it had been a while since I read that paper. I did not know that
>>> they had a formulation that included the 2nd derivative.
 
 However, if the way FSFAST is calculating the second derivative
 regressor is resulting in loss of statistical power due to shared
 variance with the primary regressor, then it would be best to just not
 include it at all in the estimation step.
>>> I don't know how much it will hurt the power. You'd have to look at the
>>> efficiency.
>>> doug
>>> 
 
 
 On Jul 10, 2013, at 1:21 PM, Douglas N Greve
  
 > 
 >> 
 wrote:
 
> 
> Why not just create a new volume and then compute contrasts from 
> the new
> volume? What you are suggesting will work I think, but it leaves me a
> little nervous. The p-values will be meaningless.
> 
> 
> As for the 2nd derivative, I think it must be a numerical issue (it is
> not computed analytically). Why do you need the 2nd derivative?
> 
> doug
> 
> 
> 
> On 07/10/2013 12:47 PM, Joseph Dien wrote:
>> I'm thinking of generating a modified beta.nii.gz file where the
>> 

Re: [Freesurfer] FSFAST mkcontrast-sess with spmhrf 2 question

[Joseph Dien]

I'm thinking of generating a modified beta.nii.gz file where the primary betas 
have been replaced with the Calhoun et al (2004) derivative boost measure.  
What do you think?  Also, please note my question below about the second 
derivative as it is causing me concern about my analysis.  Thanks!

Joe


On Jul 10, 2013, at 12:39 PM, Douglas N Greve gr...@nmr.mgh.harvard.edu wrote:

 
 I think that would work. You would need to change the time stamps on the 
 .mat file in the analysis folder. When you re-run selxavg3-sess, it will 
 see that the .mat files are newer than the beta and regenerate the 
 contrasts. But what are you planning to do the the beta file? It sounds 
 like a potentially bad idea
 
 doug
 
 
 On 07/09/2013 10:26 PM, Joseph Dien wrote:
 It looks as though selxavg3-sess generates the contrast analyses at 
 the same time as the beta weights.  Would it be possible to 
 run selxavg3-sess once to obtain the beta weights, modify the 
 beta.nii.gz file, and then rerun selxavg3-sess to obtain the contrast 
 statistics using the modified beta weights?
 
 Joe
 
 On Jul 9, 2013, at 5:08 PM, Joseph Dien jdie...@mac.com 
 mailto:jdie...@mac.com wrote:
 
 I tried correlations and the 2nd derivative is definitely not 
 orthogonal.
 
 corrcoef([X(1:207,4) X(1:207,5) X(1:207,6)])
 
 ans =
 
1.   -0.   -0.5427
   -0.1.   -0.0298
   -0.5427   -0.02981.
 
 I looked at the regressors that SPM generates for the same data:
 
 ans =
 
1.0.04360.1740
0.04361.   -0.0226
0.1740   -0.02261.
 
 The first derivative is not as orthogonal but the second derivative 
 was much more orthogonal.
 Does this have to do with what you noted below about how the second 
 derivative is being calculated?
 So does this mean I should avoid the spmhrf 2 option entirely to 
 avoid loss of statistical power?
 
 Thanks for the help!
 
 Joe
 
 
 On Jul 9, 2013, at 4:34 PM, Joseph Dien jdie...@mac.com 
 mailto:jdie...@mac.com wrote:
 
 Thanks for the quick response!  So if I wanted to use the Calhoun 
 2004 approach, I should be able to use the Steffener 2010 correction 
 to address the violation of the assumption that the regressors were 
 standardized and generate a new beta.nii.gz file where the primary 
 beta values have been replaced with the Calhoun 2004 measure.  Can I 
 assume the three regressors are more or less orthogonal?  I got 
 non-zero numbers when I tried to test the assumption in the Xtmp.X 
 variable
 
 sum(X(1:207,4).*X(1:207,5))
 
 but not hugely non-zero so maybe just rounding errors?
 
 
 On Jul 9, 2013, at 4:16 PM, Douglas N Greve 
 gr...@nmr.mgh.harvard.edu mailto:gr...@nmr.mgh.harvard.edu wrote:
 
 
 On 07/09/2013 04:11 PM, Joseph Dien wrote:
 Hi,
  I have a question about how mkcontrast-sess works.  I ran an
 analysis using the mkanalysis-sess option spmhrf 2 so there are three
 regressors for each predictor, the primary, the latency, and the
 dispersion.  When specifying the contrast weights for 
 mkcontrast-sess,
 the documentation indicates that they are specified in terms of the
 conditions as numbered in the paradigm file, not the individual
 regressors.  Furthermore there only appears to be one contrast value
 output for each contrast, not three.
 
 How are the three regressors being handled?  I can think of several
 scenarios:
 
 1) the contrast weights are not actually in terms of conditions (the
 documentation is incorrect), they are actually in terms of the
 regressors (so contrasting conditions 1 and 2 could be specified 
 as -a
 1 -a 2 -a 3 -c 4 -c 5 -c 6).
 
 2) the latency and dispersion regressors are being ignored (a common
 practice).  The contrast weights should therefore be specified as 
 -a 1
 -c 2.
 This is what happens. If you want to use the derivatives, then you 
 need
 to spec -setwdelay. When you run the command, it will prompt you for 3
 values to use. If you spec 1 0 0, then it will be the same as the
 default. If you want to test only the first derivative, then you would
 spec 0 1 0. Note that the 3rd regressor is the 2nd derivative wrt 
 time,
 not the first derivative wrt the dispersion parameter. You cannot get
 the Calhoun 2004 value using a contrast (it is non-linear).
 doug
 
 3) The Calhoun et al (2004) approach is being used to combine the
 three regressors into a derivative boost amplitude
 measure.  The contrast weights should therefore be specified as -a 
 1 -c 2.
 
 Thanks for any help you can give me!
 
 Joe
 
 
 
 Joseph Dien,
 Senior Research Scientist
 University of Maryland
 
 E-mail: jdie...@mac.com mailto:jdie...@mac.com 
 mailto:jdie...@mac.com
 Phone: 202-297-8117
 http://joedien.com//
 
 
 
 
 
 
 
 
 
 
 
 
 
 ___
 Freesurfer mailing list
 Freesurfer@nmr.mgh.harvard.edu mailto:Freesurfer@nmr.mgh.harvard.edu
 https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer

Re: [Freesurfer] FSFAST mkcontrast-sess with spmhrf 2 question

[Joseph Dien]

Sorry, not following what you are suggesting?

I want the second derivative for calculating the Calhoun et al 2004 derivative 
boost measure.
My understanding is that to the extent that the BOLD signal deviates from the 
canonical hrf, the amplitude of the primary regressor will be attenuated and 
the variance will instead end up in the first and second derivatives (to the 
extent that they are able to accommodate the divergence).  By using a Calhoun 
measure that incorporates both the first and second derivatives, in principle 
I'll have a BOLD measure that is more robust to deviations from the canonical 
hrf.

However, if the way FSFAST is calculating the second derivative regressor is 
resulting in loss of statistical power due to shared variance with the primary 
regressor, then it would be best to just not include it at all in the 
estimation step.


On Jul 10, 2013, at 1:21 PM, Douglas N Greve gr...@nmr.mgh.harvard.edu wrote:

 
 Why not just create a new volume and then compute contrasts from the new 
 volume? What you are suggesting will work I think, but it leaves me a 
 little nervous. The p-values will be meaningless.
 
 
 As for the 2nd derivative, I think it must be a numerical issue (it is 
 not computed analytically). Why do you need the 2nd derivative?
 
 doug
 
 
 
 On 07/10/2013 12:47 PM, Joseph Dien wrote:
 I'm thinking of generating a modified beta.nii.gz file where the 
 primary betas have been replaced with the Calhoun et al (2004) 
 derivative boost measure.  What do you think?  Also, please note my 
 question below about the second derivative as it is causing me concern 
 about my analysis.  Thanks!
 
 Joe
 
 
 On Jul 10, 2013, at 12:39 PM, Douglas N Greve 
 gr...@nmr.mgh.harvard.edu mailto:gr...@nmr.mgh.harvard.edu wrote:
 
 
 I think that would work. You would need to change the time stamps on the
 .mat file in the analysis folder. When you re-run selxavg3-sess, it will
 see that the .mat files are newer than the beta and regenerate the
 contrasts. But what are you planning to do the the beta file? It sounds
 like a potentially bad idea
 
 doug
 
 
 On 07/09/2013 10:26 PM, Joseph Dien wrote:
 It looks as though selxavg3-sess generates the contrast analyses at
 the same time as the beta weights.  Would it be possible to
 run selxavg3-sess once to obtain the beta weights, modify the
 beta.nii.gz file, and then rerun selxavg3-sess to obtain the contrast
 statistics using the modified beta weights?
 
 Joe
 
 On Jul 9, 2013, at 5:08 PM, Joseph Dien jdie...@mac.com 
 mailto:jdie...@mac.com
 mailto:jdie...@mac.com wrote:
 
 I tried correlations and the 2nd derivative is definitely not
 orthogonal.
 
 corrcoef([X(1:207,4) X(1:207,5) X(1:207,6)])
 
 ans =
 
   1.   -0.   -0.5427
  -0.1.   -0.0298
  -0.5427   -0.02981.
 
 I looked at the regressors that SPM generates for the same data:
 
 ans =
 
   1.0.04360.1740
   0.04361.   -0.0226
   0.1740   -0.02261.
 
 The first derivative is not as orthogonal but the second derivative
 was much more orthogonal.
 Does this have to do with what you noted below about how the second
 derivative is being calculated?
 So does this mean I should avoid the spmhrf 2 option entirely to
 avoid loss of statistical power?
 
 Thanks for the help!
 
 Joe
 
 
 On Jul 9, 2013, at 4:34 PM, Joseph Dien jdie...@mac.com 
 mailto:jdie...@mac.com
 mailto:jdie...@mac.com wrote:
 
 Thanks for the quick response!  So if I wanted to use the Calhoun
 2004 approach, I should be able to use the Steffener 2010 correction
 to address the violation of the assumption that the regressors were
 standardized and generate a new beta.nii.gz file where the primary
 beta values have been replaced with the Calhoun 2004 measure.  Can I
 assume the three regressors are more or less orthogonal?  I got
 non-zero numbers when I tried to test the assumption in the Xtmp.X
 variable
 
 sum(X(1:207,4).*X(1:207,5))
 
 but not hugely non-zero so maybe just rounding errors?
 
 
 On Jul 9, 2013, at 4:16 PM, Douglas N Greve
 gr...@nmr.mgh.harvard.edu mailto:gr...@nmr.mgh.harvard.edu 
 mailto:gr...@nmr.mgh.harvard.edu wrote:
 
 
 On 07/09/2013 04:11 PM, Joseph Dien wrote:
 Hi,
 I have a question about how mkcontrast-sess works.  I ran an
 analysis using the mkanalysis-sess option spmhrf 2 so there are 
 three
 regressors for each predictor, the primary, the latency, and the
 dispersion.  When specifying the contrast weights for
 mkcontrast-sess,
 the documentation indicates that they are specified in terms of the
 conditions as numbered in the paradigm file, not the individual
 regressors.  Furthermore there only appears to be one contrast value
 output for each contrast, not three.
 
 How are the three regressors being handled?  I can think of several
 scenarios:
 
 1) the contrast weights are not actually in terms of conditions (the
 documentation is incorrect), they are actually in terms of the
 regressors (so contrasting conditions 1 and 2 

Re: [Freesurfer] FSFAST mkcontrast-sess with spmhrf 2 question

[Joseph Dien]

Sounds good!

Regarding creating a new volume and computing contrasts from it, what do you 
mean?  I didn't follow that.

Thanks!

Joe


On Jul 10, 2013, at 1:33 PM, Douglas N Greve gr...@nmr.mgh.harvard.edu wrote:

 
 On 07/10/2013 01:29 PM, Joseph Dien wrote:
 Sorry, not following what you are suggesting?
 
 I want the second derivative for calculating the Calhoun et al 2004 
 derivative boost measure.
 My understanding is that to the extent that the BOLD signal deviates 
 from the canonical hrf, the amplitude of the primary regressor will be 
 attenuated and the variance will instead end up in the first and 
 second derivatives (to the extent that they are able to accommodate 
 the divergence).  By using a Calhoun measure that incorporates both 
 the first and second derivatives, in principle I'll have a BOLD 
 measure that is more robust to deviations from the canonical hrf.
 Sorry, it had been a while since I read that paper. I did not know that 
 they had a formulation that included the 2nd derivative.
 
 However, if the way FSFAST is calculating the second derivative 
 regressor is resulting in loss of statistical power due to shared 
 variance with the primary regressor, then it would be best to just not 
 include it at all in the estimation step.
 I don't know how much it will hurt the power. You'd have to look at the 
 efficiency.
 doug
 
 
 
 On Jul 10, 2013, at 1:21 PM, Douglas N Greve 
 gr...@nmr.mgh.harvard.edu mailto:gr...@nmr.mgh.harvard.edu wrote:
 
 
 Why not just create a new volume and then compute contrasts from the new
 volume? What you are suggesting will work I think, but it leaves me a
 little nervous. The p-values will be meaningless.
 
 
 As for the 2nd derivative, I think it must be a numerical issue (it is
 not computed analytically). Why do you need the 2nd derivative?
 
 doug
 
 
 
 On 07/10/2013 12:47 PM, Joseph Dien wrote:
 I'm thinking of generating a modified beta.nii.gz file where the
 primary betas have been replaced with the Calhoun et al (2004)
 derivative boost measure.  What do you think?  Also, please note my
 question below about the second derivative as it is causing me concern
 about my analysis.  Thanks!
 
 Joe
 
 
 On Jul 10, 2013, at 12:39 PM, Douglas N Greve
 gr...@nmr.mgh.harvard.edu mailto:gr...@nmr.mgh.harvard.edu 
 mailto:gr...@nmr.mgh.harvard.edu wrote:
 
 
 I think that would work. You would need to change the time stamps 
 on the
 .mat file in the analysis folder. When you re-run selxavg3-sess, it 
 will
 see that the .mat files are newer than the beta and regenerate the
 contrasts. But what are you planning to do the the beta file? It sounds
 like a potentially bad idea
 
 doug
 
 
 On 07/09/2013 10:26 PM, Joseph Dien wrote:
 It looks as though selxavg3-sess generates the contrast analyses at
 the same time as the beta weights.  Would it be possible to
 run selxavg3-sess once to obtain the beta weights, modify the
 beta.nii.gz file, and then rerun selxavg3-sess to obtain the contrast
 statistics using the modified beta weights?
 
 Joe
 
 On Jul 9, 2013, at 5:08 PM, Joseph Dien jdie...@mac.com 
 mailto:jdie...@mac.com
 mailto:jdie...@mac.com
 mailto:jdie...@mac.com wrote:
 
 I tried correlations and the 2nd derivative is definitely not
 orthogonal.
 
 corrcoef([X(1:207,4) X(1:207,5) X(1:207,6)])
 
 ans =
 
  1.   -0.   -0.5427
 -0.1.   -0.0298
 -0.5427   -0.02981.
 
 I looked at the regressors that SPM generates for the same data:
 
 ans =
 
  1.0.04360.1740
  0.04361.   -0.0226
  0.1740   -0.02261.
 
 The first derivative is not as orthogonal but the second derivative
 was much more orthogonal.
 Does this have to do with what you noted below about how the second
 derivative is being calculated?
 So does this mean I should avoid the spmhrf 2 option entirely to
 avoid loss of statistical power?
 
 Thanks for the help!
 
 Joe
 
 
 On Jul 9, 2013, at 4:34 PM, Joseph Dien jdie...@mac.com 
 mailto:jdie...@mac.com
 mailto:jdie...@mac.com
 mailto:jdie...@mac.com wrote:
 
 Thanks for the quick response!  So if I wanted to use the Calhoun
 2004 approach, I should be able to use the Steffener 2010 correction
 to address the violation of the assumption that the regressors were
 standardized and generate a new beta.nii.gz file where the primary
 beta values have been replaced with the Calhoun 2004 measure.  Can I
 assume the three regressors are more or less orthogonal?  I got
 non-zero numbers when I tried to test the assumption in the Xtmp.X
 variable
 
 sum(X(1:207,4).*X(1:207,5))
 
 but not hugely non-zero so maybe just rounding errors?
 
 
 On Jul 9, 2013, at 4:16 PM, Douglas N Greve
 gr...@nmr.mgh.harvard.edu mailto:gr...@nmr.mgh.harvard.edu 
 mailto:gr...@nmr.mgh.harvard.edu
 mailto:gr...@nmr.mgh.harvard.edu wrote:
 
 
 On 07/09/2013 04:11 PM, Joseph Dien wrote:
 Hi,
 I have a question about how mkcontrast-sess works.  I ran an
 analysis using the mkanalysis-sess option spmhrf 2 so there are
 three
 

Re: [Freesurfer] FSFAST mkcontrast-sess with spmhrf 2 question

[Douglas N Greve]

On 07/09/2013 04:11 PM, Joseph Dien wrote:
 Hi,
I have a question about how mkcontrast-sess works.  I ran an 
 analysis using the mkanalysis-sess option spmhrf 2 so there are three 
 regressors for each predictor, the primary, the latency, and the 
 dispersion.  When specifying the contrast weights for mkcontrast-sess, 
 the documentation indicates that they are specified in terms of the 
 conditions as numbered in the paradigm file, not the individual 
 regressors.  Furthermore there only appears to be one contrast value 
 output for each contrast, not three.

 How are the three regressors being handled?  I can think of several 
 scenarios:

 1) the contrast weights are not actually in terms of conditions (the 
 documentation is incorrect), they are actually in terms of the 
 regressors (so contrasting conditions 1 and 2 could be specified as -a 
 1 -a 2 -a 3 -c 4 -c 5 -c 6).

 2) the latency and dispersion regressors are being ignored (a common 
 practice).  The contrast weights should therefore be specified as -a 1 
 -c 2.
This is what happens. If you want to use the derivatives, then you need 
to spec -setwdelay. When you run the command, it will prompt you for 3 
values to use. If you spec 1 0 0, then it will be the same as the 
default. If you want to test only the first derivative, then you would 
spec 0 1 0. Note that the 3rd regressor is the 2nd derivative wrt time, 
not the first derivative wrt the dispersion parameter. You cannot get 
the Calhoun 2004 value using a contrast (it is non-linear).
doug

 3) The Calhoun et al (2004) approach is being used to combine the 
 three regressors into a derivative boost amplitude 
 measure.  The contrast weights should therefore be specified as -a 1 -c 2.

 Thanks for any help you can give me!

 Joe

 

 Joseph Dien,
 Senior Research Scientist
 University of Maryland

 E-mail: jdie...@mac.com mailto:jdie...@mac.com
 Phone: 202-297-8117
 http://joedien.com//













 ___
 Freesurfer mailing list
 Freesurfer@nmr.mgh.harvard.edu
 https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer

-- 
Douglas N. Greve, Ph.D.
MGH-NMR Center
gr...@nmr.mgh.harvard.edu
Phone Number: 617-724-2358
Fax: 617-726-7422

Bugs: surfer.nmr.mgh.harvard.edu/fswiki/BugReporting
FileDrop: https://gate.nmr.mgh.harvard.edu/filedrop2
www.nmr.mgh.harvard.edu/facility/filedrop/index.html
Outgoing: ftp://surfer.nmr.mgh.harvard.edu/transfer/outgoing/flat/greve/

___
Freesurfer mailing list
Freesurfer@nmr.mgh.harvard.edu
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer


The information in this e-mail is intended only for the person to whom it is
addressed. If you believe this e-mail was sent to you in error and the e-mail
contains patient information, please contact the Partners Compliance HelpLine at
http://www.partners.org/complianceline . If the e-mail was sent to you in error
but does not contain patient information, please contact the sender and properly
dispose of the e-mail.

[Freesurfer] FSFAST mkcontrast-sess with spmhrf 2 question

[Joseph Dien]

Hi,
   I have a question about how mkcontrast-sess works.  I ran an analysis using 
the mkanalysis-sess option spmhrf 2 so there are three regressors for each 
predictor, the primary, the latency, and the dispersion.  When specifying the 
contrast weights for mkcontrast-sess, the documentation indicates that they are 
specified in terms of the conditions as numbered in the paradigm file, not the 
individual regressors.  Furthermore there only appears to be one contrast value 
output for each contrast, not three.

How are the three regressors being handled?  I can think of several scenarios:

1) the contrast weights are not actually in terms of conditions (the 
documentation is incorrect), they are actually in terms of the regressors (so 
contrasting conditions 1 and 2 could be specified as -a 1 -a 2 -a 3 -c 4 -c 5 
-c 6).

2) the latency and dispersion regressors are being ignored (a common practice). 
 The contrast weights should therefore be specified as -a 1 -c 2.

3) The Calhoun et al (2004) approach is being used to combine the three 
regressors into a derivative boost amplitude measure.  The contrast weights 
should therefore be specified as -a 1 -c 2.

Thanks for any help you can give me!

Joe



Joseph Dien,
Senior Research Scientist
University of Maryland 

E-mail: jdie...@mac.com
Phone: 202-297-8117
http://joedien.com//











___
Freesurfer mailing list
Freesurfer@nmr.mgh.harvard.edu
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer


The information in this e-mail is intended only for the person to whom it is
addressed. If you believe this e-mail was sent to you in error and the e-mail
contains patient information, please contact the Partners Compliance HelpLine at
http://www.partners.org/complianceline . If the e-mail was sent to you in error
but does not contain patient information, please contact the sender and properly
dispose of the e-mail.

Re: [Freesurfer] FSFAST mkcontrast-sess with spmhrf 2 question

[Joseph Dien]

Thanks for the quick response!  So if I wanted to use the Calhoun 2004 
approach, I should be able to use the Steffener 2010 correction to address the 
violation of the assumption that the regressors were standardized and generate 
a new beta.nii.gz file where the primary beta values have been replaced with 
the Calhoun 2004 measure.  Can I assume the three regressors are more or less 
orthogonal?  I got non-zero numbers when I tried to test the assumption in the 
Xtmp.X variable 

sum(X(1:207,4).*X(1:207,5))

but not hugely non-zero so maybe just rounding errors?


On Jul 9, 2013, at 4:16 PM, Douglas N Greve gr...@nmr.mgh.harvard.edu wrote:

 
 On 07/09/2013 04:11 PM, Joseph Dien wrote:
 Hi,
   I have a question about how mkcontrast-sess works.  I ran an 
 analysis using the mkanalysis-sess option spmhrf 2 so there are three 
 regressors for each predictor, the primary, the latency, and the 
 dispersion.  When specifying the contrast weights for mkcontrast-sess, 
 the documentation indicates that they are specified in terms of the 
 conditions as numbered in the paradigm file, not the individual 
 regressors.  Furthermore there only appears to be one contrast value 
 output for each contrast, not three.
 
 How are the three regressors being handled?  I can think of several 
 scenarios:
 
 1) the contrast weights are not actually in terms of conditions (the 
 documentation is incorrect), they are actually in terms of the 
 regressors (so contrasting conditions 1 and 2 could be specified as -a 
 1 -a 2 -a 3 -c 4 -c 5 -c 6).
 
 2) the latency and dispersion regressors are being ignored (a common 
 practice).  The contrast weights should therefore be specified as -a 1 
 -c 2.
 This is what happens. If you want to use the derivatives, then you need 
 to spec -setwdelay. When you run the command, it will prompt you for 3 
 values to use. If you spec 1 0 0, then it will be the same as the 
 default. If you want to test only the first derivative, then you would 
 spec 0 1 0. Note that the 3rd regressor is the 2nd derivative wrt time, 
 not the first derivative wrt the dispersion parameter. You cannot get 
 the Calhoun 2004 value using a contrast (it is non-linear).
 doug
 
 3) The Calhoun et al (2004) approach is being used to combine the 
 three regressors into a derivative boost amplitude 
 measure.  The contrast weights should therefore be specified as -a 1 -c 2.
 
 Thanks for any help you can give me!
 
 Joe
 
 
 
 Joseph Dien,
 Senior Research Scientist
 University of Maryland
 
 E-mail: jdie...@mac.com mailto:jdie...@mac.com
 Phone: 202-297-8117
 http://joedien.com//
 
 
 
 
 
 
 
 
 
 
 
 
 
 ___
 Freesurfer mailing list
 Freesurfer@nmr.mgh.harvard.edu
 https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
 
 -- 
 Douglas N. Greve, Ph.D.
 MGH-NMR Center
 gr...@nmr.mgh.harvard.edu
 Phone Number: 617-724-2358
 Fax: 617-726-7422
 
 Bugs: surfer.nmr.mgh.harvard.edu/fswiki/BugReporting
 FileDrop: https://gate.nmr.mgh.harvard.edu/filedrop2
 www.nmr.mgh.harvard.edu/facility/filedrop/index.html
 Outgoing: ftp://surfer.nmr.mgh.harvard.edu/transfer/outgoing/flat/greve/
 
 ___
 Freesurfer mailing list
 Freesurfer@nmr.mgh.harvard.edu
 https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
 
 
 The information in this e-mail is intended only for the person to whom it is
 addressed. If you believe this e-mail was sent to you in error and the e-mail
 contains patient information, please contact the Partners Compliance HelpLine 
 at
 http://www.partners.org/complianceline . If the e-mail was sent to you in 
 error
 but does not contain patient information, please contact the sender and 
 properly
 dispose of the e-mail.
 




Joseph Dien,
Senior Research Scientist
University of Maryland 

E-mail: jdie...@mac.com
Phone: 202-297-8117
http://joedien.com//












___
Freesurfer mailing list
Freesurfer@nmr.mgh.harvard.edu
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer


The information in this e-mail is intended only for the person to whom it is
addressed. If you believe this e-mail was sent to you in error and the e-mail
contains patient information, please contact the Partners Compliance HelpLine at
http://www.partners.org/complianceline . If the e-mail was sent to you in error
but does not contain patient information, please contact the sender and properly
dispose of the e-mail.

Re: [Freesurfer] FSFAST mkcontrast-sess with spmhrf 2 question

[Joseph Dien]

I tried correlations and the 2nd derivative is definitely not orthogonal.  

 corrcoef([X(1:207,4) X(1:207,5) X(1:207,6)])

ans =

1.   -0.   -0.5427
   -0.1.   -0.0298
   -0.5427   -0.02981.

I looked at the regressors that SPM generates for the same data:

ans =

1.0.04360.1740
0.04361.   -0.0226
0.1740   -0.02261.

The first derivative is not as orthogonal but the second derivative was much 
more orthogonal.
Does this have to do with what you noted below about how the second derivative 
is being calculated?
So does this mean I should avoid the spmhrf 2 option entirely to avoid loss of 
statistical power?

Thanks for the help!

Joe


On Jul 9, 2013, at 4:34 PM, Joseph Dien jdie...@mac.com wrote:

 Thanks for the quick response!  So if I wanted to use the Calhoun 2004 
 approach, I should be able to use the Steffener 2010 correction to address 
 the violation of the assumption that the regressors were standardized and 
 generate a new beta.nii.gz file where the primary beta values have been 
 replaced with the Calhoun 2004 measure.  Can I assume the three regressors 
 are more or less orthogonal?  I got non-zero numbers when I tried to test the 
 assumption in the Xtmp.X variable 
 
 sum(X(1:207,4).*X(1:207,5))
 
 but not hugely non-zero so maybe just rounding errors?
 
 
 On Jul 9, 2013, at 4:16 PM, Douglas N Greve gr...@nmr.mgh.harvard.edu wrote:
 
 
 On 07/09/2013 04:11 PM, Joseph Dien wrote:
 Hi,
   I have a question about how mkcontrast-sess works.  I ran an 
 analysis using the mkanalysis-sess option spmhrf 2 so there are three 
 regressors for each predictor, the primary, the latency, and the 
 dispersion.  When specifying the contrast weights for mkcontrast-sess, 
 the documentation indicates that they are specified in terms of the 
 conditions as numbered in the paradigm file, not the individual 
 regressors.  Furthermore there only appears to be one contrast value 
 output for each contrast, not three.
 
 How are the three regressors being handled?  I can think of several 
 scenarios:
 
 1) the contrast weights are not actually in terms of conditions (the 
 documentation is incorrect), they are actually in terms of the 
 regressors (so contrasting conditions 1 and 2 could be specified as -a 
 1 -a 2 -a 3 -c 4 -c 5 -c 6).
 
 2) the latency and dispersion regressors are being ignored (a common 
 practice).  The contrast weights should therefore be specified as -a 1 
 -c 2.
 This is what happens. If you want to use the derivatives, then you need 
 to spec -setwdelay. When you run the command, it will prompt you for 3 
 values to use. If you spec 1 0 0, then it will be the same as the 
 default. If you want to test only the first derivative, then you would 
 spec 0 1 0. Note that the 3rd regressor is the 2nd derivative wrt time, 
 not the first derivative wrt the dispersion parameter. You cannot get 
 the Calhoun 2004 value using a contrast (it is non-linear).
 doug
 
 3) The Calhoun et al (2004) approach is being used to combine the 
 three regressors into a derivative boost amplitude 
 measure.  The contrast weights should therefore be specified as -a 1 -c 2.
 
 Thanks for any help you can give me!
 
 Joe
 
 
 
 Joseph Dien,
 Senior Research Scientist
 University of Maryland
 
 E-mail: jdie...@mac.com mailto:jdie...@mac.com
 Phone: 202-297-8117
 http://joedien.com//
 
 
 
 
 
 
 
 
 
 
 
 
 
 ___
 Freesurfer mailing list
 Freesurfer@nmr.mgh.harvard.edu
 https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
 
 -- 
 Douglas N. Greve, Ph.D.
 MGH-NMR Center
 gr...@nmr.mgh.harvard.edu
 Phone Number: 617-724-2358
 Fax: 617-726-7422
 
 Bugs: surfer.nmr.mgh.harvard.edu/fswiki/BugReporting
 FileDrop: https://gate.nmr.mgh.harvard.edu/filedrop2
 www.nmr.mgh.harvard.edu/facility/filedrop/index.html
 Outgoing: ftp://surfer.nmr.mgh.harvard.edu/transfer/outgoing/flat/greve/
 
 ___
 Freesurfer mailing list
 Freesurfer@nmr.mgh.harvard.edu
 https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
 
 
 The information in this e-mail is intended only for the person to whom it is
 addressed. If you believe this e-mail was sent to you in error and the e-mail
 contains patient information, please contact the Partners Compliance 
 HelpLine at
 http://www.partners.org/complianceline . If the e-mail was sent to you in 
 error
 but does not contain patient information, please contact the sender and 
 properly
 dispose of the e-mail.
 
 
 
 
 
 Joseph Dien,
 Senior Research Scientist
 University of Maryland 
 
 E-mail: jdie...@mac.com
 Phone: 202-297-8117
 http://joedien.com//
 
 
 
 
 
 
 
 
 
 
 
 
 ___
 Freesurfer mailing list