Hm, I see what you mean. I have been looking at these images as raw files through ImageJ since I'm reading them as raw files when processing them. The large differences in intensity between the input and output FLAIRs are evident in this case. When I look at them in fslview, I see that the difference in intensities is not as significant.
-Anil
----- Original Message ----
From: Mark Jenkinson <[log in to unmask]>
To: [log in to unmask]
Sent: Mon, June 21, 2010 3:10:20 PM
Subject: Re: [FSL] Issues with intensity scaling when using FLIRT
Dear Anil,
I see no problem with any of these.
Everything seems consistent - the values in the FSLView intensity
window, the histogram generated by FSLView and the stats output
by fslstats. For example:
ws183$ fslstats FLAIR_masked.nii -r -R
0.000000 6163.926270 0.000000 18566.042969 1009.216190 3510.841900
ws183$ fslstats cFLAIR_masked.nii -r -R
0.000000 5669.895996 0.000000 17078.000000 857.194649 2773.019085
These differences are certainly consistent with interpolation effects.
What version of FSL and FSLView are you using?
I am using FSL version 4.1.5 and FSLview version 3.1.5.
I notice that the scl_slope in your nifti images is
set to non-unity values (e.g. 15.069840 in FLAIR_masked
and 3.338950 for PD_masked). These values represent
scalings of the intensities which FSL will apply to the
values stored in the non-header part of the image, as
dictated by the NIFTI format. There may have been
older versions of FSLView that did not display the
scaled intensities, in which might explain why you
quote a mean value in the original FLAIR around 130
whereas I see values that are much bigger than this.
If this is the problem then you can easily save
a scaled version (where the scl_slope is reset to
1.0 and the stored values are scaled instead) by doing:
fslmaths FLAIR_masked ScaledFLAIR_masked
I hope this helps solve the problem.
All the best,
Mark
On 21 Jun 2010, at 20:27, Anil Krishna Vasireddi wrote:
> The reference number is 979615.
> I have uploaded a set of three images from the same brain. The input FLAIR (FLAIR_masked.nii), the input PD (PD_masked.nii), and the output FLAIR (cFLAIR_masked.nii).
>
> Thanks again-Anil
>
>
>
> ----- Original Message ----
> From: Mark Jenkinson <[log in to unmask]>
> To: [log in to unmask]
> Sent: Fri, June 18, 2010 10:42:43 AM
> Subject: Re: [FSL] Issues with intensity scaling when using FLIRT
>
> Him
>
> This sounds very concerning and very unusual.
> Can you please upload the relevant images to:
> http://www.fmrib.ox.ac.uk/cgi-bin/upload.cgi
> and send us the reference number.
>
> Hopefully we will be able to identify the problem quickly.
>
> All the best,
> Mark
>
>
>
> On 18 Jun 2010, at 16:35, Anil Krishna Vasireddi wrote:
>
>> The resolution of the PD is higher than the resolution of the FLAIR.
>>
>> PD: 288x288x42
>> FLAIR: 160x160x40
>>
>> To test the increase in intensity, I look at intensities in the same regions of the brain within the two volumes. To quantify the degree of scaling, I compare the mean intensity of the two volumes. The mean intensity within the original FLAIR typically rests around 130. The mean intensity in the registered FLAIR varied for different volumes, ranging from 900 to 2500.
>>
>>
>> -Anil
>>
>>
>>
>> ----- Original Message ----
>> From: Mark Jenkinson <[log in to unmask]>
>> To: [log in to unmask]
>> Sent: Thu, June 17, 2010 4:59:57 PM
>> Subject: Re: [FSL] Issues with intensity scaling when using FLIRT
>>
>> Hi,
>>
>> There is no scaling or modification of the intensities except for
>> interpolation. Is it possible that what you are seeing are just
>> interpolation effects? If the resolution of the PD is lower than
>> the FLAIR then that will make the effects worse.
>>
>> If you still do not think it is just interpolation effects then can
>> you describe how you are testing this and give us some
>> numerical examples?
>>
>> All the best,
>> Mark
>>
>>
>>
>>
>> On 17 Jun 2010, at 22:40, Anil Vasireddi wrote:
>>
>>> Hi,
>>>
>>> I have been using FLIRT to register FLAIR volumes of a given brain to the corresponding proton density (PD) volumes. I am using the following simple command for this purpose:
>>>
>>> flirt -in <FLAIR volume> -ref <PD volume> -out <registered FLAIR volume>
>>>
>>> The input and reference volumes were similar enough for the above command to be sufficient for a successful registration.
>>>
>>> However, the I have been noticing that the image intensities of voxels in the output FLAIR volumes are considerably larger than the intensities from the input FLAIR volumes. It doesn't seem as though the intensities were simply multiplied by a scalar factor either.
>>>
>>> Is it possible to use the flirt command without altering the image intensities from the input volumes during registration?
>>>
>>> Any help would be greatly appreciated.
>>>
>>> -Anil
>>>
>>
>>
>>
>>
>
>
>
>
>
|