Hi,
Thanks for the quick response. Unfortunately the design of my study limits
me quite a bit. I do need to downsample, and I need to do it before
thresholding. This is because I'm combining it with a dataset that's
natively 4x4x4mm, and I need to threshold the two images after they're
combined.
Looking at the intensity of the 1mm and 4mm voxels, they do seem to be
more or less equivalent (though it's not easy to make a 100% accurate
comparison due to the resampling). Even at a Z threshold of 0 and P of 1,
there's nothing. This is a really strange issue, and I think I'm out of
ideas for now..
Sean
On 2013-03-22 1:58 PM, "wolf zinke" <[log in to unmask]> wrote:
>Hi,
>
>Maybe the best response will come from the experts with a deeper insight
>into the tools. But here is my shot anyway. When you downsample data,
>you have a problem of how to get the information of higer resolution
>voxel into new voxel at lower resolution. As far as I understand it,
>flirt applys a blurring prior to the downsampling (that can be disabled
>by the option -noresampblur). The blurring and the interpolation of the
>downsample reduces the voxel intensities, and hence it also blurres the
>significance of active voxel. Also does the downsample affect the
>smoothness of the data, which also is important for the determination of
>significant cluster. Further, the cluster tool takes the number of voxel
>within a patch into account, which also is reduced after downsampling.
>So I guess you introduce a lot of trouble with this step.
>
>Why do you need to downsample? If necessary, would it be an option to
>first apply the thresholding, and downsample the thresholded maps?
>
>cheers,
>wolf
>
>On 03/22/2013 05:46 PM, McWhinney, Sean wrote:
>> Hi Wolf,
>>
>> I've determined that this problem has something to do with downsampling.
>> Running easythresh using FSL's native 1x1x1mm activation map on a
>>1x1x1mm
>> brain mask produces exactly what I would expect to see. However, running
>> the command below on the mask and map to downsample the two, then
>>running
>> easythresh again, results in not a single cluster making it through.
>>
>> flirt -interp nearestneighbour -in <file> -ref <file> -applyisoxfm 4
>>-out
>> <file_4mm>
>>
>>
>> I've confirmed that the two files have the same voxel dimensions and
>>size,
>> so I'm not sure what could be causing this problem. Are you aware of why
>> downsampling my data may cause this?
>>
>> Regards,
>> Sean
>>
>> On 2013-03-21 8:40 PM, "wolf zinke" <[log in to unmask]> wrote:
>>
>>> Hi,
>>>
>>> Hard to say what's going on without more details. You mentioned that
>>>the
>>> data was'nt created by feat, so what format is it. Are this z-values?
>>> If, for example, you try to input 1-p data, this for sure would not
>>>work.
>>>
>>> What do you mean by robust activation? Did you already apply some
>>> suitable thresholding (including multiple testing corrections)?
>>>
>>> If the data has the correct format, and voxels are clearly (i.e.
>>> significantly) activated, then you should not get empty maps with
>>> easythresh. If they are empty, it means either no voxel survives the
>>> thresholding, or you do not imput the correct data.
>>>
>>> I hope this helps,
>>> wolf
>>>
>>> On 03/21/2013 09:23 PM, McWhinney, Sean wrote:
>>>> Hi Wolf,
>>>>
>>>> Thanks for the response. It seems that regardless of what input
>>>> parameters
>>>> I use, not a single voxel comes out active in any case, even when I
>>>>set
>>>> the thresholds very very lightly. This is odd because the map has some
>>>> very robust activation. Have you experienced this before? Some strange
>>>> patterns near the edges suggest it may be a transformation issue.
>>>>
>>>> I could try using cluster instead, but I wasn't able to figure out how
>>>> to
>>>> output a thresholded image using that.
>>>>
>>>> Thanks,
>>>> Sean
>>>>
>>>> On 2013-03-21 5:02 PM, "wolf zinke" <[log in to unmask]> wrote:
>>>>
>>>>> Hi,
>>>>>
>>>>> Have a look at the easythresh tool, that might simplify things for
>>>>>you
>>>>> a
>>>>> bit.
>>>>>
>>>>> cluster will also do. However, when you look into the easythresh
>>>>> script,
>>>>> you will notice, besides the relevant options for a cluster based
>>>>> thresholding, that it also produces an estimate for the smoothness of
>>>>> the data (smoothest).
>>>>>
>>>>> I hope this helps,
>>>>> wolf
>>>>>
>>>>>
>>>>> On 03/21/2013 05:17 PM, Sean wrote:
>>>>>> I'm looking for a way to perform FSL's cluster thresholding from the
>>>>>> command line. I'm hoping to do it the same way FEAT does it in
>>>>>> post-stats, but I'll be running it on a single NII file that wasn't
>>>>>> generated by FEAT. I've looked into the cluster command, but as far
>>>>>> as I
>>>>>> can tell this only reports statistics about clusters. Is this
>>>>>>possible
>>>>>> to do, and if so, how is it done? Are there any extra files (i.e.,
>>>>>> registration transformation matrix, etc) that I'll be needing?
>>>>>>Thanks
>>>>>> for your help.
|