Great. Thanks for clariyfing.
One last question...
In some cases, I want to correlate the PPI term with another time course.
Given that both the PPI term and the other time course are whitened by
spm_regions, can I assume that the autocorrelation has been removed and I
can safely use the standard p-value for that correlation?
Thanks again!
Alex
On 07/05/2011 04:07, "MCLAREN, Donald" <[log in to unmask]> wrote:
> The null-space of the contrast are any columns of the contrast (which
> must be an F-contrast) that are 0. Thus, you remove the effects of
> motion when you do the adjustment and don't need to worry about them
> being included in the deconvolved data. The code that I have supplied
> has N rows for N conditions of interest after I talked to Darren about
> the ideal contrast to use for adjustment.
>
> As for removing the frequencies, I think it is probably fine to remove
> them, I was just under the impression that they weren't being removed
> because of me commenting out that line during testing of spm_regions.
> I would go with the SPM defaults until their effect is tested. If you
> don't set a high-pass filter, then you won't remove the frequencies,
> so you could test it that way as well.
>
> 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 Thu, May 5, 2011 at 9:46 PM, Alex Fornito <[log in to unmask]> wrote:
>> Hi Donald, Torben and Darren,
>> Thanks for your responses.
>>
>> Form each of your responses, it seems like best practice would be to remove
>> motion (and/or other confounds) from the VOI time series prior to
>> deconvolution.
>>
>> It also seems like you are recommending NOT to frequency filter the VOI time
>> series.
>>
>> In my reading though, spm_regions does this by default with the call to
>> spm_filter (line 135). So, if spm_regions is used to extract a VOI time
>> series, the time series that is input to spm_peb_ppi will already frequency
>> filtered. So it seems like, in standard practice, the deconvolution is
>> applied to frequency-filtered data. In this case, if xX.iG is empty, the
>> only difference between Y and Yc in spm_peb_ppi is the additional removal of
>> block effects (xX.iB).
>>
>> Based on the current chain of emails, it seems you are recommending that a
>> confound-corrected, whitened, but non-frequency filtered time course should
>> be input to spm_peb_ppi. IS that correct, or am I missing something?
>>
>> Finally, can I clarify exactly what the null space of the contrast is? Most
>> of the time in my analyses I don't get the option to adjust for it, so xY.Ic
>> is empty.
>>
>> Thanks again for all your help,
>> A
>>
>>
>>
>> On 06/05/2011 02:18, "MCLAREN, Donald" <[log in to unmask]> wrote:
>>
>>> I haven't thoroughly tested this, but if you set iG to be the movement
>>> parameters columns. Then still adjust for the null space (motion
>>> parameters). You will get the best of both. If you don't adjust for
>>> the null space, then movement will contribute to the deconvolution. In
>>> summary, if you have iG be the motion parameter columns AND adjust for
>>> the null-space (motion columns), then:
>>>
>>> Y will have the effect of motion removed. Yc (removal of motion twice)
>>> will be minimally different since Y is orthogonal to the motion
>>> parameters. So, you have motion removed from both the autocorrelation
>>> estimate, the Y for deconvolution, and Yc.
>>>
>>> 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 Thu, May 5, 2011 at 9:48 AM, Torben Ellegaard Lund
>>> <[log in to unmask]> wrote:
>>>> Hi Alex
>>>>
>>>>
>>>> Den Uge:18 05/05/2011 kl. 14.15 skrev Alex Fornito:
>>>>
>>>>> Hi Torben,
>>>>> Sorry, I just wanted to clarify: do you mean leaving SPM.xX.iG empty? It
>>>>> seems like this is done by default by spm_fmri_design (?)
>>>>
>>>> leaving it empty will include all user defined regressors in the effects of
>>>> interest F-test which determines the mask where the serial correlation is
>>>> estimated. I think it would make sense if the motion regressors did not go
>>>> into this test, just like the DCS HP-filter does not go into the
>>>> F-contrast.
>>>>
>>>>> Are you suggesting manually entering motion parameters into this field?
>>>>
>>>> Yes, or their indices that is. It should be done after specification, but
>>>> before model estimation. You can check if SPM recognized your changes by
>>>> looking at the automatically generated effects of interest F-contrast.
>>>>
>>>>
>>>> Best
>>>> Torben
>>>>
>>>>
>>>>
>>>>
>>>>>
>>>>> On 05/05/2011 18:16, "Torben Ellegaard Lund" <[log in to unmask]> wrote:
>>>>>
>>>>>> Just as a kind reminder. Leaving SPM.xX.iG is a bit unfortunate, because
>>>>>> voxels where motion artefacts are significant then constitutes a large
>>>>>> part
>>>>>> of
>>>>>> the F-test derived mask used to estimate serial correlations. If you are
>>>>>> picky
>>>>>> about this you can change it yourselves, but it is not as easy as it
>>>>>> should
>>>>>> be.
>>>>>>
>>>>>>
>>>>>> Best
>>>>>> Torben
>>>>>>
>>>>>>
>>>>>>
>>>>>> Den Uge:18 05/05/2011 kl. 05.03 skrev Alex Fornito:
>>>>>>
>>>>>>> Ahh, I see. I was assuming that SPM.xX.iG contained the indices to the
>>>>>>> nuisance covariates in the design matrix, but you're right -
>>>>>>> spm_fMRI_design
>>>>>>> leaves it empty.
>>>>>>>
>>>>>>> Thanks for your help!
>>>>>>> A
>>>>>>>
>>>>>>>
>>>>>>> On 05/05/2011 12:54, "MCLAREN, Donald" <[log in to unmask]> wrote:
>>>>>>>
>>>>>>>> I agree that nuisance covariates, such as motion should be removed;
>>>>>>>> however, motion covariates are not generally stored in X0 in my
>>>>>>>> reading of the SPM code (spm_fMRI_design, spm_regions). They should be
>>>>>>>> removed in the spm_regions filtering based on the null-space of the
>>>>>>>> contrast.
>>>>>>>>
>>>>>>>> As for the motion parameters being in the model, if they were in the
>>>>>>>> standard analysis, they should be included in the PPI analysis. My
>>>>>>>> gPPI code will do this automatically as well.
>>>>>>>>
>>>>>>>> X0 is made of SPM.xX.iB (block effects -- so removing the mean) and
>>>>>>>> SPM.xX.iG (nuisance variable that is empty as far as I can tell (line
>>>>>>>> 369 of spm_fMRI_design)) and the high-pass filtering (K.X0).
>>>>>>>>
>>>>>>>> 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 Wed, May 4, 2011 at 10:16 PM, Alex Fornito <[log in to unmask]>
>>>>>>>> wrote:
>>>>>>>>> Hi Donald,
>>>>>>>>> Thanks for your response. When you say "you don't want
>>>>>>>>> to filter out anything related to neural activity", I'm presuming that
>>>>>>>>> you
>>>>>>>>> are referring to the deconvolved neural signal?
>>>>>>>>>
>>>>>>>>> If so, that makes sense. However, X0 also contains user-specified
>>>>>>>>> nuisance
>>>>>>>>> covariates. I would have thought it would make sense to remove those,
>>>>>>>>> depending on what they were (e.g., movement parameters)? I guess they
>>>>>>>>> can
>>>>>>>>> always be entered into the PPI regression model...
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> On 05/05/2011 11:57, "MCLAREN, Donald" <[log in to unmask]>
>>>>>>>>> wrote:
>>>>>>>>>
>>>>>>>>>> I could be wrong, Karl or Darren please correct me if I am.
>>>>>>>>>>
>>>>>>>>>> X0 is composed of the high-pass filter and constant term. Thus, you
>>>>>>>>>> want to filter these out of the Yc regressor. However, you don't want
>>>>>>>>>> to filter out anything related to neural activity, so you wouldn't
>>>>>>>>>> want to filter your signal and then predict the neural activity from
>>>>>>>>>> the filtered signal, especially filtering frequencies.
>>>>>>>>>>
>>>>>>>>>> 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 Sat, Apr 30, 2011 at 8:03 PM, Alex Fornito
>>>>>>>>>> <[log in to unmask]>
>>>>>>>>>> wrote:
>>>>>>>>>>> Hi Donald,
>>>>>>>>>>> As far as I can tell, spm_regions filters and whitens the VOI time
>>>>>>>>>>> series,
>>>>>>>>>>> and adjusts for the null space of the contrast if an adjustment is
>>>>>>>>>>> specified
>>>>>>>>>>> in xY.Ic. It defines the confound matrix, X0, but does not correct
>>>>>>>>>>> for
>>>>>>>>>>> it.
>>>>>>>>>>> The following line in spm_peb_ppi corrects for the confounds.
>>>>>>>>>>>
>>>>>>>>>>> Yc = Y - X0*inv(X0'*X0)*X0'*Y;
>>>>>>>>>>> PPI.Y = Yc(:,1);
>>>>>>>>>>>
>>>>>>>>>>> So the above correction does seem to be doing something different to
>>>>>>>>>>> spm_regions. I'm wondering why the uncorrected data, Y, is used when
>>>>>>>>>>> generating the ppi interaction term, while the corrected data Yc is
>>>>>>>>>>> to
>>>>>>>>>>> be
>>>>>>>>>>> used as a covariate of no interest in the PPI model. I would has
>>>>>>>>>>> thought
>>>>>>>>>>> that Yc should be used to generate the ppi interaction term as well.
>>>>>>>>>>>
>>>>>>>>>>> In my data, my X0 is a 195 x 7 matrix (I have 195 volumes per
>>>>>>>>>>> session).
>>>>>>>>>>>
>>>>>>>>>>> Regards,
>>>>>>>>>>> Alex
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> On 30/04/2011 02:22, "MCLAREN, Donald" <[log in to unmask]>
>>>>>>>>>>> wrote:
>>>>>>>>>>>
>>>>>>>>>>>> I'm actually travelling at the moment. However, I know Y has
>>>>>>>>>>>> already
>>>>>>>>>>>> been adjusted as part of the VOI processing.
>>>>>>>>>>>>
>>>>>>>>>>>> Can you check what X0 looks like?
>>>>>>>>>>>>
>>>>>>>>>>>> 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 Fri, Apr 29, 2011 at 2:55 AM, Alex Fornito
>>>>>>>>>>>> <[log in to unmask]>
>>>>>>>>>>>> wrote:
>>>>>>>>>>>> Hi all,
>>>>>>>>>>>> I have a question concerning the code used to implement a PPI
>>>>>>>>>>>> analysis.
>>>>>>>>>>>>
>>>>>>>>>>>> In the spm_peb_ppi.m function packaged with SPM8, lines 329-332
>>>>>>>>>>>> remove
>>>>>>>>>>>> confounds from the input seed region¹s time course:
>>>>>>>>>>>>
>>>>>>>>>>>> % Remove confounds and save Y in ouput structure
>>>>>>>>>>>>
%------------------------------------------------------------------>>>>>>>>>>>>
-
>>>>>>>>>>>> ---
>>>>>>>>>>>> --
>>>>>>>>>>>> --
>>>>>>>>>>>> Yc = Y - X0*inv(X0'*X0)*X0'*Y;
>>>>>>>>>>>> PPI.Y = Yc(:,1);
>>>>>>>>>>>>
>>>>>>>>>>>> PPI.Y is then to be used as a covariate of no interest when
>>>>>>>>>>>> building
>>>>>>>>>>>> the
>>>>>>>>>>>> PPI
>>>>>>>>>>>> regression model. However, on line 488, it is the uncorrected seed
>>>>>>>>>>>> time
>>>>>>>>>>>> course, Y, that is input to the deconvolution routine and used to
>>>>>>>>>>>> generate
>>>>>>>>>>>> the ppi term:
>>>>>>>>>>>>
>>>>>>>>>>>> C = spm_PEB(Y,P);
>>>>>>>>>>>> xn = xb*C{2}.E(1:N);
>>>>>>>>>>>> xn = spm_detrend(xn);
>>>>>>>>>>>>
>>>>>>>>>>>> % Setup psychological variable from inputs and contast weights
>>>>>>>>>>>>
>>>>>>>>>>>>
%------------------------------------------------------------------>>>>>>>>>>>>
-
>>>>>>>>>>>> ---
>>>>>>>>>>>> PSY = zeros(N*NT,1);
>>>>>>>>>>>> for i = 1:size(U.u,2)
>>>>>>>>>>>> PSY = PSY + full(U.u(:,i)*U.w(:,i));
>>>>>>>>>>>> end
>>>>>>>>>>>> % PSY = spm_detrend(PSY); <- removed centering of psych variable
>>>>>>>>>>>> % prior to multiplication with xn. Based on discussion with Karl
>>>>>>>>>>>> % and Donald McLaren.
>>>>>>>>>>>>
>>>>>>>>>>>> % Multiply psychological variable by neural signal
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>>
%------------------------------------------------------------------>>>>>>>>>>>>
-
>>>>>>>>>>>> ---
>>>>>>>>>>>> PSYxn = PSY.*xn;
>>>>>>>>>>>>
>>>>>>>>>>>> Is there any specific reason why Y, rather than Yc(:,1), is used to
>>>>>>>>>>>> compute
>>>>>>>>>>>> PSYxn? I thought Yc might be more appropriate (?).
>>>>>>>>>>>> Thanks for your help,
>>>>>>>>>>>> Alex
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>> ------------------------------------------
>>>>>>>>>>>>
>>>>>>>>>>>> Alex Fornito
>>>>>>>>>>>> Research Fellow
>>>>>>>>>>>> Melbourne Neuropsychiatry Centre
>>>>>>>>>>>> University of Melbourne
>>>>>>>>>>>> National Neuroscience Facility
>>>>>>>>>>>> Levels 1 & 2, Alan Gilbert Building
>>>>>>>>>>>> 161 Barry St
>>>>>>>>>>>> Carlton South 3053
>>>>>>>>>>>> Victoria, Australia
>>>>>>>>>>>>
>>>>>>>>>>>> Email: [log in to unmask]
>>>>>>>>>>>> Phone: +61 3 8344 1876
>>>>>>>>>>>> Fax: +61 3 9348 0469
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> ------------------------------------------
>>>>>>>>>>>
>>>>>>>>>>> Alex Fornito
>>>>>>>>>>> Research Fellow
>>>>>>>>>>> Melbourne Neuropsychiatry Centre
>>>>>>>>>>> University of Melbourne
>>>>>>>>>>> National Neuroscience Facility
>>>>>>>>>>> Levels 1 & 2, Alan Gilbert Building
>>>>>>>>>>> 161 Barry St
>>>>>>>>>>> Carlton South 3053
>>>>>>>>>>> Victoria, Australia
>>>>>>>>>>>
>>>>>>>>>>> Email: [log in to unmask]
>>>>>>>>>>> Phone: +61 3 8344 1876
>>>>>>>>>>> Fax: +61 3 9348 0469
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> ------------------------------------------
>>>>>>>>>
>>>>>>>>> Alex Fornito
>>>>>>>>> Research Fellow
>>>>>>>>> Melbourne Neuropsychiatry Centre
>>>>>>>>> University of Melbourne
>>>>>>>>> National Neuroscience Facility
>>>>>>>>> Levels 1 & 2, Alan Gilbert Building
>>>>>>>>> 161 Barry St
>>>>>>>>> Carlton South 3053
>>>>>>>>> Victoria, Australia
>>>>>>>>>
>>>>>>>>> Email: [log in to unmask]
>>>>>>>>> Phone: +61 3 8344 1876
>>>>>>>>> Fax: +61 3 9348 0469
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> ------------------------------------------
>>>>>>>
>>>>>>> Alex Fornito
>>>>>>> Research Fellow
>>>>>>> Melbourne Neuropsychiatry Centre
>>>>>>> University of Melbourne
>>>>>>> National Neuroscience Facility
>>>>>>> Levels 1 & 2, Alan Gilbert Building
>>>>>>> 161 Barry St
>>>>>>> Carlton South 3053
>>>>>>> Victoria, Australia
>>>>>>>
>>>>>>> Email: [log in to unmask]
>>>>>>> Phone: +61 3 8344 1876
>>>>>>> Fax: +61 3 9348 0469
>>>>>>
>>>>>>
>>>>>
>>>>>
>>>>> ------------------------------------------
>>>>>
>>>>> Alex Fornito
>>>>> Research Fellow
>>>>> Melbourne Neuropsychiatry Centre
>>>>> University of Melbourne
>>>>> National Neuroscience Facility
>>>>> Levels 1 & 2, Alan Gilbert Building
>>>>> 161 Barry St
>>>>> Carlton South 3053
>>>>> Victoria, Australia
>>>>>
>>>>> Email: [log in to unmask]
>>>>> Phone: +61 3 8344 1876
>>>>> Fax: +61 3 9348 0469
>>>>>
>>>>>
>>>>>
>>>>
>>>
>>
>>
>> ------------------------------------------
>>
>> Alex Fornito
>> Research Fellow
>> Melbourne Neuropsychiatry Centre
>> University of Melbourne
>> National Neuroscience Facility
>> Levels 1 & 2, Alan Gilbert Building
>> 161 Barry St
>> Carlton South 3053
>> Victoria, Australia
>>
>> Email: [log in to unmask]
>> Phone: +61 3 8344 1876
>> Fax: +61 3 9348 0469
>>
>>
>>
>>
>
------------------------------------------
Alex Fornito
Research Fellow
Melbourne Neuropsychiatry Centre
University of Melbourne
National Neuroscience Facility
Levels 1 & 2, Alan Gilbert Building
161 Barry St
Carlton South 3053
Victoria, Australia
Email: [log in to unmask]
Phone: +61 3 8344 1876
Fax: +61 3 9348 0469
|