Imagine, the following is real data: 0 0 0 1 1 1 1 0 0 0
A -- Imagine that conv(0) gives: 0 0 0 1 0 0 0 0 0 0
B -- Imagine that conv(4) gives: 0 0 0 10 10 10 10 0 0 0
C -- Imagine that 10*conv(4) gives: 0 0 0 100 100 100 100 0 0 0
The results would be as follow:
A misses the activity, B and C produce the same statistical image
because the error is the same in both models. The beta coefficient
will be a 1/10 of the value of B in C.
The point is that the critical component is the shape of the HRF, not
the peak or amplitude in the model. The beta coefficient will do the
scaling.
Thus, you should use duration=5s.
On Thu, Apr 9, 2009 at 11:57 AM, Dorian P. <[log in to unmask]> wrote:
> Dear Donald,
>
> I plotted all BOLD responses for durations 0s, 1s, 2s, 5s, 10s, 80s.
> It is exactly as you say, the plateau happens for stimuli > 10s. This
> is different from what is explained in Design Efficiency tutorial (>
> 2s).
>
> Importantly choosing 0s vs 1s still makes still a big difference to
> the model. Don't know if this is normal, but it makes me think of a
> strong bias on SPM regressors. Let's say the real bold response is
> like 5s convolve for a 5s stimulus. If this is true and I use dur = 0s
> to convolve HRF then the error will be very high exactly for those
> areas that have a high correlation with the task. In other words I
> will miss the real activation just because I used dur = 0s. On the
> other hand what I may be getting is the areas that have weak
> activation related to the task, still related, but weak.
>
> So what is the best thing to assume?
> 1. The SPM convolve for 5s stimulus is not exact (not reflecting the
> real BOLD) or
> 2. A 5s (or even 1s) stimulus will produce erroneous results if I use dur = 0s.
>
> Thanks to Donald for his nice explanation.
>
> Attached is the image of predicted bold responses. Here is the script
> I used to obtain that:
> %---------------------------------------
> samplespersecond=16
> hrf=spm_hrf(1/samplespersecond)
> dur=zeros(size(hrf))
>
> dur0=dur; dur0(1)=1;
> bold0=conv(dur0,hrf);
>
> dur1=dur; dur1(1:samplespersecond)=1;
> bold1=conv(dur1,hrf);
>
> dur2=dur; dur2(1:2*samplespersecond)=1;
> bold2=conv(dur2,hrf);
>
> dur5=dur; dur5(1:5*samplespersecond)=1;
> bold5=conv(dur5,hrf);
>
> dur10=dur; dur10(1:10*samplespersecond)=1;
> bold10=conv(dur10,hrf);
>
> dur80=dur; dur80(1:80*samplespersecond)=1;
> bold80=conv(dur80,hrf);
>
> plot(bold0,'g'); hold;
> plot(bold1,'r');
> plot(bold2,'r');
> plot(bold5,'b');
> plot(bold10,'c');
> plot(bold80,'m');
> %-----------------------------------------------------
>
> 2009/4/9 MCLAREN, Donald <[log in to unmask]>:
>> I believe this is the case (the graphs are consistent with Rik's
>> website). However, remember that there is not true instantaneous
>> events in real-life. The goal is to pick an HRF that matches in shape.
>> As long as the shape matches, then you can vary the scaling to match
>> the amplitude.
>>
>> I'm not sure what the basis for choosing 0s duration was in the past.
>> Choosing 0s versus 1s probably does not make very much difference
>> (based on previous posts).
>>
>> Going from ER to BR is a continum. With the inception of convolution,
>> they are treated the same way. In earlier days (see Ogawa for block;
>> see Buckner for trial averaging events), the methods for analyzing
>> them were quite different. I'd say any event lasting less than 10
>> seconds in an event-related design, some will call this an epoch. As
>> you extend an event beyond 10 seconds it becomes a block. The critical
>> difference is that the response truly reaches a plateau after about 10
>> seconds of stimulation.
>>
>> Note, consistent with the work by Fox et al on transients at the
>> beginning of a block, you see that in reflected in this model as well.
>>
>>
>> On Thu, Apr 9, 2009 at 10:57 AM, Dorian P. <[log in to unmask]> wrote:
>>> Hi Donald,
>>>
>>> From your example I can see the difference in amplitude between 0s and
>>> 5s can be 100-1000 times higher. Is this the real case of the BOLD
>>> response for brief and longer stimuli? Is this what SPM does for
>>> different durations? It's impressive.
>>>
>>>
>>> I've read around much more reasons why to use durations > 0, but only
>>> a few for using dur = 0s. What is the real benefit of using a 0s
>>> duration? In what cases would this be better? From what we say here it
>>> appears that dur=0 isn't making an experiment event-related, because
>>> we can treat also durations of 1s as blocks and the design is still ok
>>> (and event-related from my perspective).
>>>
>>> Thanks again :)
>>> Dorian.
>>>
>>>
>>> 2009/4/9 MCLAREN, Donald <[log in to unmask]>:
>>>> See comments below
>>>>
>>>> On Thu, Apr 9, 2009 at 10:00 AM, Dorian P. <[log in to unmask]> wrote:
>>>>> Dear all,
>>>>>
>>>>> After revisiting some threads here and the Design Efficiency monograph
>>>>> I am confused about the effects of duration on the HRF shape. I
>>>>> thought the difference between a delta function and a boxcar should be
>>>>> the length of the peak because (citing Design Efficiency, Part VII,
>>>>> http://imaging.mrc-cbu.cam.ac.uk/imaging/DesignEfficiency) "for longer
>>>>> than 2s duration trials, the response begins to plateau".
>>>>>
>>>>> But in the same section is noted:
>>>>> "After convolution with the IR, a difference in the duration of a
>>>>> trial causes a difference in the scaling (size) of the predicted
>>>>> response". This is shown in their Fig 19 with trials of 8s having huge
>>>>> amplitudes compared to the others. Same effect is shown in a Matlab
>>>>> example from Poldrack in this thread:
>>>>> https://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind01&L=SPM&P=R30702&I=-3&K=1&X=10D5DD76B0C377C9B3
>>>>>
>>>>> Here come my questions:
>>>>> 1. What is the real difference between the HRF convolved for a
>>>>> duration of 0s and a duration of 5s ?
>>>>
>>>> Amplitude and shape
>>>>
>>>>> 2. Can somebody suggest some commands (similar to Poldrack's) to
>>>>> simulate different durations and the resulting HRF for an event?
>>>>
>>>> samplespersecond=16
>>>> hrf=spm_hrf(1/samplespersecond)
>>>> dur=zeros(size(hrf))
>>>> dur0=dur; dur0(1)=1;
>>>> dur1=dur; dur1(1:samplespersecond)=1;
>>>> ..
>>>> ..
>>>> dur5=dur; dur5(1:5*samplespersecond)=1;
>>>> bold0=conv(dur0,hrf)
>>>> bold1=conv(dur1,hrf)
>>>> ...
>>>> ...
>>>> bold5=conv(dur5,hrf)
>>>> plot(bold0,'g-'); hold plot(bold5,'b+')
>>>>
>>>>
>>>>
>>>>
>>>>>
>>>>> I tried to create a simulation myself and found out that SPM.xBF.bf
>>>>> has the delta function of my experiment (attached as JPG). Apparently
>>>>> it is created in spm_get_bf.m and help shows this:
>>>>> "spm_get_bf prompts for basis functions to model event or epoch-related
>>>>> responses. The basis functions returned are unitary and orthonormal
>>>>> when defined as a function of peri-stimulus time in time-bins.
>>>>> It is at this point that the distinction between event and epoch-related
>>>>> responses enters."
>>>>>
>>>>> 3. But if SPM.xBF.bf is the same for the whole experiment, how are
>>>>> durations 0s and durations 5s convolved in a mixed model design
>>>>> (event-related + blocked)? Isn't it enough that I put different
>>>>> durations in the SPM interface?
>>>>
>>>> SPM.xBF.bf is the hrf, it is convolved with a vector of when the
>>>> stimulus is on. SPM.xBF.bf is in microtime.
>>>>
>>>> ON periods convolved with SPM.xBF.bf will produce your regressors. The
>>>> basis function is the same for 0 and 5 second durations. The duration
>>>> causes the regressors to be different, not the basis function.
>>>>
>>>> Hope this helps.
>>>>
>>>>>
>>>>> Would really appreciate your answers.
>>>>>
>>>>> Dorian.
>>>>>
>>>>
>>>>
>>>>
>>>> --
>>>> Best Regards, Donald McLaren
>>>> =====================
>>>> D.G. McLaren
>>>> University of Wisconsin - Madison
>>>> Neuroscience Training Program
>>>> Office: (608) 265-9672
>>>> Lab: (608) 256-1901 ext 12914
>>>> =====================
>>>> 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 (608) 265-9672 or
>>>> email.
>>>>
>>>
>>
>>
>>
>> --
>> Best Regards, Donald McLaren
>> =====================
>> D.G. McLaren
>> University of Wisconsin - Madison
>> Neuroscience Training Program
>> Office: (608) 265-9672
>> Lab: (608) 256-1901 ext 12914
>> =====================
>> 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 (608) 265-9672 or
>> email.
>>
>
--
Best Regards, Donald McLaren
=====================
D.G. McLaren
University of Wisconsin - Madison
Neuroscience Training Program
Office: (608) 265-9672
Lab: (608) 256-1901 ext 12914
=====================
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 (608) 265-9672 or
email.
|
