Hi,
Is it possible that you were using an older version of FSL?
It used to be the case that if FLIRT was used with only integer images then you would
not see the non-binary values, but internally it would still make them. In that case you
ended up with an implicit threshold of around 1.0, which is OK in some applications
but not in others. We have always recommended that people think about their
application and explicitly choose a suitable threshold for their particular case. This
has been our recommendation for as long as we've had FSL, and I strongly
encourage you to do the explicit thresholding in all cases.
All the best,
Mark
On 9 Jan 2011, at 09:52, Susan Lee wrote:
> In both of the analyses, I used exactly the same command, which was:
>
> flirt -in <mask> -ref example_func -applyxfm -init reg/standard2example_func.mat -out S102_mask
>
> I did not define -interp nearestneighbor for either analyses. The only difference in the 2 cases was that the first analysis used functionally defined ROIs from a higher-level analysis, whereas the second analysis used structurally defined ROIs created from the Harvard Oxford atlas. In both cases, the standard-space mask was binarized and in integer format prior to applying the transform.
>
> Thanks to anyone who can help clarify... I just like to make sure that I understand what I'm doing and how these 2 cases differ.
>
> Susan.
>
>
> On Jan 9, 2011, at 4:20 AM, Andreas Bartsch wrote:
>
>> Did you use nearest neighbour Interpolation?
>> Andreas
>> ________________________________________
>> Von: FSL - FMRIB's Software Library [[log in to unmask]] im Auftrag von Susan Lee [[log in to unmask]]
>> Gesendet: Sonntag, 9. Januar 2011 10:14
>> An: [log in to unmask]
>> Betreff: Re: [FSL] binary mask is no longer binary
>>
>> Mark, thanks for your prompt reply. Quick follow-up question...
>>
>> In a previous PPI analysis, functional ROIs were defined using a
>> cluster from a group analysis. These clusters were binarized, and
>> converted into individual functional space using identical flirt
>> commands. The individual ROIs that resulted from this process were
>> still in integer format and binary (i.e., we did not need to threshold
>> and re-binarize these masks).
>>
>> Could you please help me understand the difference between these 2
>> scenarios?
>>
>> Thank you so much!
>>
>> Susan.
>>
>> On Jan 9, 2011, at 4:02 AM, Mark Jenkinson wrote:
>>
>>> Hi,
>>>
>>> This is covered in the FSL course lecture on registration.
>>> The relevant slide text is:
>>>
>>> Mask values are normally 0 and 1 (integer format)
>>> - Interpolation gives values in between
>>> - if truncated to integer ==> mask "shrinks”
>>> - Need to re-threshold (binarize) the transformed mask
>>> - "Correct" thresholding depends on the particular case
>>> - Threshold near 0.0 to include partial-volume edges
>>> - Threshold near 1.0 to exclude partial-volume edges
>>> - Threshold at 0.5 to keep the same size (approx)
>>>
>>> So the answer depends on your application, and whether
>>> you want to be conservative, inclusive or keep the
>>> mask about the same size (in mm^3).
>>>
>>> All the best,
>>> Mark
>>>
>>>
>>> On 9 Jan 2011, at 08:23, Susan wrote:
>>>
>>>> Hello,
>>>>
>>>> I have converted a binary structural ROI (created from the Harvard
>>>> Oxford atlas) into individual functional space for use in a PPI
>>>> analysis. The resulting mask (when viewed in FSLview) is no longer
>>>> binary.
>>>>
>>>> To convert the mask, this is what I did:
>>>>
>>>> flirt -in STGmask -ref example_func -applyxfm -init reg/
>>>> standard2example_func.mat -out S102_STGmask
>>>>
>>>> Based on searching the forum, my best guess is that I need to apply
>>>> a threshold to this new mask using some combination of -thr and -
>>>> bin... but I'm not quite sure how to go about choosing the
>>>> appropriate -thr value. Any help would be greatly appreciate!
>>>>
>>>> Susan.
>>>>
>
|