Hi again,
I all my reasoning I am assuming that whatever registration we are talking about it is unbiased.
> The unknown in my mind is to what extent the derivation of the original
> ROIs depends on the registration algorithm. My thought is that you won't
> necessarily get the same ROIs using a algorithm based on FLIRT as you
> would with FNIRT, and that the degree of similarity or dissimilarity would
> be dependent on the precise manner in which the original ROIs were
> obtained. For example, if the ROIs were based on significant fMRI
> activation in a large group of subjects, it is easy to envision that ROIs
> based on FLIRTING the subjects into MNI space would be larger than those
> obtained by FNIRTING them into MNI space, since FLIRT registration would
> tend to spread the activation out over a larger area in MNI space.
I honestly don't see how this could be done in a particularly objective way. Unless you actually partition the entire brain with some heroic battery of tasks you will always be stuck with a final decision how threshold the individual maps and then how to threshold the aggregate map.
> On the
> other hand, if the ROIs are based on classifying all voxels in each
> individual subject, and then assigning the class based on maximum
> probability after transform into MNI space, then those ensuing ROIs might
> be somewhat independent of whether the transformation used FLIRT or FNIRT,
I guess I would say this is the only sensible way of building ROIs in standard space so my assumption would be that it was done in that way.
> since the ordered rankings of class probabilities within each voxel might
> not change too much, even if the probabilities themselves change (with
> FNIRT expected to have more of the probability contained in a fewer number
> of classes).
I would be surprised if the, given a reasonable sized group, fnirt or flirt would make much of a difference as to where the borders between the regions were put. I agree that you could probably get better borders from a smallish group with fnirt than with flirt.
Regardless of that I still think that when you then have an additional subject that you want to use those ROIs in you would want to do the best possible job of the registration, and that would most likely be non-linear.
What do you think?
Jesper
>
> Does that seem like a reasonable expectation?
>
> cheers,
> -MH
>
>
> --
> Michael Harms, Ph.D.
>
> -----------------------------------------------------------
> Conte Center for the Neuroscience of Mental Disorders
> Washington University School of Medicine
> Department of Psychiatry, Box 8134
> 660 South Euclid Ave. Tel: 314-747-6173
> St. Louis, MO 63110 Email: [log in to unmask]
>
>
>
>
> On 11/28/12 1:56 PM, "Jesper Andersson" <[log in to unmask]> wrote:
>
>> Hi Michael and Donald,
>>
>> I think what professor Ashburner is saying is that it is always a good
>> idea to use the best possible registration. Let's say we have some ROI
>> that has been defined in a bunch of subjects that have been linearly
>> transformed (FLIRTED) into standard space and then averaged and
>> thresholded. If the number of subjects was big enough we would expect
>> that ROI to be a quite good standard space representation of the region
>> in question, much better than we would expect if we just drew one ROI in
>> one subject.
>>
>> Let's now say we have some subject we want to apply that ROI to. I think
>> it would make sense to use the best possible registration (non-linear,
>> though maybe not SPM ;-) ) to register that subject to standard space.
>> After all, using a linear registration is likely to achieve a worse
>> matching of that subject to standard space, and hence to the ROI in
>> question.
>>
>> Does this make sense?
>>
>> Jesper
>>
>> On 28 Nov 2012, at 20:46, Michael Harms wrote:
>>
>>> Hi Donald,
>>> I don't see any way to just warp the data. What you really want to know
>>> in your example is what the Harvard-Oxford labels would have been had
>>> they
>>> been derived using registrations that involved FNIRT. That seems an
>>> empirical question to me. Ideally, one would want to use a set of ROIs
>>> derived using the same templates and algorithms as the study to which
>>> those ROIs are going to be applied. But since that isn't always
>>> possible
>>> you have a make a decision informed by the needs/questions of your
>>> particular study.
>>>
>>> cheers,
>>> -MH
>>>
>>> --
>>> Michael Harms, Ph.D.
>>>
>>> -----------------------------------------------------------
>>> Conte Center for the Neuroscience of Mental Disorders
>>> Washington University School of Medicine
>>> Department of Psychiatry, Box 8134
>>> 660 South Euclid Ave. Tel: 314-747-6173
>>> St. Louis, MO 63110 Email: [log in to unmask]
>>>
>>>
>>>
>>>
>>> On 11/28/12 1:24 PM, "MCLAREN, Donald" <[log in to unmask]> wrote:
>>>
>>>> Michael,
>>>>
>>>> Excellent points. However, the underlying question still remains about
>>>> using existing ROI with newer normalization approaches.
>>>>
>>>> For example, would using the Harvard-Oxford labels be bad if you use
>>>> FSLs FNIRT as they were created with FLIRT? Or is there a way to warp
>>>> the data?
>>>>
>>>> We are not using DARTEL, just SPM's regular non-linear warp.
>>>>
>>>> On Wed, Nov 28, 2012 at 1:50 PM, Michael Harms <[log in to unmask]>
>>>> wrote:
>>>>> I would think that you might raise some eyebrows using ROIs obtained
>>>>> using
>>>>> FSL's FLIRT in data normalized using SPM's non-linear (Dartel?) tool,
>>>>> due
>>>>> to the big potential difference between linear (affine) and non-linear
>>>>> approaches. Of course, it would depend on the anatomical precision
>>>>> that
>>>>> you want/need for your particular study and whether the ROIs involve
>>>>> regions that are particularly sensitive to non-linear registration.
>>>>>
>>>>> cheers,
>>>>> -MH
>>>>>
>>>>> --
>>>>> Michael Harms, Ph.D.
>>>>>
>>>>> -----------------------------------------------------------
>>>>> Conte Center for the Neuroscience of Mental Disorders
>>>>> Washington University School of Medicine
>>>>> Department of Psychiatry, Box 8134
>>>>> 660 South Euclid Ave. Tel: 314-747-6173
>>>>> St. Louis, MO 63110 Email: [log in to unmask]
>>>>>
>>>>>
>>>>>
>>>>>
>>>>> On 11/28/12 11:43 AM, "MCLAREN, Donald" <[log in to unmask]>
>>>>> wrote:
>>>>>
>>>>>> Dear SPM/FSL users,
>>>>>>
>>>>>> It seems that there are several templates (SPM v FSL) as well as
>>>>>> different normalization routines that could result in slight
>>>>>> variations of the localization of the results. Do I need to worry
>>>>>> about these small differences between methods?
>>>>>>
>>>>>> In particular, I want to know if I can take regions defined in a
>>>>>> study
>>>>>> using FLIRT in FSL and use them in my study that has been processed
>>>>>> with SPM's non-linear normalization tool. Can I use them as is or is
>>>>>> there a transform that can I compute and/or apply to the ROIs to get
>>>>>> them into the SPM normalized space.
>>>>>>
>>>>>> Thank you in advance for your input.
>>>>>>
>>>>>> Best Regards, Donald McLaren
>>>>>> =================
>>>>>> D.G. McLaren, Ph.D.
>>>>>> Research Fellow, Department of Neurology, Massachusetts General
>>>>>> Hospital
>>>>>> and
>>>>>> Harvard Medical School
>>>>>> Postdoctoral Research Fellow, GRECC, Bedford VA
>>>>>> Website: http://www.martinos.org/~mclaren
>>>>>> 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.
>>>
>
|