Hi,
Since you extract the median from all the voxels within
the mask, I suspect it would be fine to go with this approach.
All the best,
Mark
On 31 Jan 2011, at 21:29, Michael Harms wrote:
> Hi Mark,
> Yeah, after seeing that the necessary variance already exists in the
> 'sigmasquareds' file, I realized that the mask would be the hardest
> part.
>
> What do you think about defining a single GM mask in standard space, and
> then propagating that back to the native space for the calculation? Not
> as precise as your suggested approach, but do you think it would really
> matter that much? Friedman and Glover used a single GM mask defined in
> standard space when they implemented this in their paper.
>
> thanks for your thoughts,
> -MH
>
> On Mon, 2011-01-31 at 21:12 +0000, Mark Jenkinson wrote:
>> Hi Michael,
>>
>> This is pretty easy to do.
>> The hardest part is getting the mask right.
>> Once you do have the mask, all you need to do
>> is go to the FEAT directory and run:
>> fslmaths mean_func -mul mean_func -div stats/sigmasquareds -sqrt SFNR_image
>> fslstats SFNR_image -k GM_mask -P 50
>> which will give you the median value within the mask.
>>
>> So now to the tougher bit - getting the mask.
>> I would start by generating a segmentation of the
>> T1 structural image. After this, apply the registration
>> warps (ideally using a fieldmap-derived warp, but if
>> not a 6/7 DOF registration will do) to get the GM
>> segmentation back into the space of the functional
>> images. Then I would run BET on the functional image,
>> multiply the BET mask by the transformed GM mask
>> and then threshold the resulting image around 0.9
>> and binarise this. The resulting binary image should be
>> a reasonable description of a GM mask, and given that
>> we will take the median of the voxelwise results, the final
>> result will be fairly insensitive to small errors in this mask.
>>
>> I hope this helps.
>>
>> All the best,
>> Mark
>>
>>
>>
>>
>> On 27 Jan 2011, at 20:52, Michael Harms wrote:
>>
>>> Hello,
>>> I'm wondering if anyone using FSL has implemented the calculation of an
>>> SFNR (signal-to-fluctuation-noise ratio) for each 1st level run using a
>>> gray matter mask for task-based fMRI data -- i.e., "SFNR-Type-3-GM-SM-
>>> Resid" in the terminology of Friedman and Glover (NI, 2006, 33:471-481).
>>>
>>> I believe that all the pieces are available in FSL to compute something
>>> substantially similar. I'm just wondering if someone has already worked
>>> out the details and has a script available that they would be willing to
>>> share...?
>>>
>>> thanks,
>>> -MH
>>>
>>> --
>>> Michael Harms, Ph.D.
>>> --------------------------------------------------------------------
>>> Conte Center for the Neuroscience of Mental Disorders
>>> Washington University School of Medicine
>>> Department of Psychiatry, Box 8134
>>> Renard Hospital, Room 6604 Tel: 314-747-6173
>>> 660 South Euclid Ave. Fax: 314-747-2182
>>> St. Louis, MO 63110 Email: [log in to unmask]
>>> --------------------------------------------------------------------
>>>
>
|