Hi Rolf,
Thanks a lot for your help. As you said, the ".mat" can catch very gross
failures of registration. That helps.
I always visually inspect the registrations, like what Mark suggested. By so
doing I can have an opinion of which registration works better. Let's say I
have two image volumes a and b. Using a as reference, I register b to a with
one registration method and created a volume c; then I do the same thing
with another registration method and created a volume d. Now I want to
calculate the mutual information between a and c, as well as the mutual
information between a and d. Is there a command to use in FSL to do this?
I understand what Mark said that change of such numbers does not
necessarily indicate a change in the quality of the alignment. I'm just curious
and want to get such numbers to play with them. That's all.
Jim
On Tue, 29 Sep 2009 15:57:48 +0100, Rolf Heckemann
<[log in to unmask]> wrote:
>Hi Jim
>
>When you run FLIRT with default settings to register a pair of images
>(no matter whether they are from the same subject or different ones),
>the output represents the optimal alignment as measured by the
>correlation ratio. You can set an option for FLIRT to use a different
>cost function, e.g. mutual information. This then becomes the measure
>of "goodness", and FLIRT manipulates the alignment parameters until
>this value becomes as "good" as possible.
>
>Perhaps you can see now why it is difficult to answer your question.
>
>If you want to catch very gross failures, such as a pair of images not
>matching at all after registration, and you want to avoid opening and
>looking at each aligned pair, then I suggest you have a look at
>flirt's .mat output. The numbers in there describe the
>transformation. If you have a large number of .mat files, you may
>recognize failed registrations as outliers in one or more of these
>numbers.
>
>Hope that helps
>
>Rolf
>
>
>On Tue, Sep 29, 2009 at 03:35:54PM +0100, Jim Li wrote:
>> Hi Mark,
>>
>> Thanks a lot for the message, :)
>>
>> You know, I'm not doing image registration on different subjects. I have
the
>> same subject whose T2-weighted images and b0 images were done on the
>> same day, and I tried two different ways to register the two image
volumes. I
>> just want to get some values to quantify the registration and see if there's
a
>> pattern. Could you suggest some ready-to-use script? Will fslmaths work?
>>
>> Thanks a lot,
>>
>> Jim
>>
>>
>>
>> On Tue, 29 Sep 2009 08:17:52 +0100, Mark Jenkinson
>> <[log in to unmask]> wrote:
>>
>> >Hi,
>> >
>> >I'm afraid that there is no easy way to compare the quality of
>> >registrations for different subjects. You can use mutual information
>> >or other metrics to compare the quality of two different registrations
>> >of the same images (this is how the registration works), but the values
>> >are not really comparable across different registrations of different
>> >images/subjects, and will tend to be influenced a lot by the FOV,
>> >SNR, artefact level (e.g. amount of motion) and so on. Hence a
>> >larger/smaller number does not necessarily indicate a change in
>> >the quality of the alignment.
>> >
>> >We strongly recommend you visually inspect all your registrations.
>> >This is definitely the best way to assess the quality of the alignment.
>> >
>> >All the best,
>> > Mark
>> >
>> >
>> >On 28 Sep 2009, at 14:46, Jim Li wrote:
>> >
>> >> Hello everybody,
>> >>
>> >> I have another Newbie question about registering two image volumes
>> >> of the
>> >> same subject using the same modality (MRI) after using FLIRT or
>> >> other scripts.
>> >> Can anyone give me a hint?
>> >>
>> >> Basicly, other than visual inspection of the two images after
>> >> registration, are
>> >> there ready-to-use scripts in FSL that can quantify the overall
>> >> goodness of
>> >> registration? I heard people use mutual information, etc. I just
>> >> want to run a
>> >> command to get a number (or more) to quantify the goodness of
>> >> registration...
>> >>
>> >> Thanks a lot, :)
>> >>
>> >> Jim
>> >>
>
>--
>Rolf A Heckemann, MD PhD
>Médecin chercheur
>Fondation Neurodis
>CERMEP - Imagerie du Vivant
>Hôpital Neurologique Pierre Wertheimer
>59 Boulevard Pinel
>69003 Lyon
>
>
>
>1254235444
>
>
|