Elisa,
I almost overlooked your tiny reference to my name ;) but in any case,
you may really want to be careful in this age range: processing infant
data has a number of problems, including issues such as brain size which
is not an issue from ~6 years onward but is in the younger ones. Also,
there are tissue classes in the very young (such as unmyelinated white
matter) that the algorithm cannot deal with correctly as it has no
information that it exists in the first place.
Hence, trying to squeeze together in one analyses subjects that are too
different is likely not a good idea. Your alternative (subdividing into
several analyses) may work better but then you cannot directly compare
results anymore.
In my experience, there are datasets of children < 2 where segmentation
in cat12 worked surprisingly well (as judged visually), but in others it
fails spectacularly, and not surprisingly so. I would therefore tend to
agree with Christian (although I would probably still try it to see if
it works ;)
Cheers
Marko
Elisa Marchetta wrote:
> Dear Christian,
>
> I understand your doubt about studying so young subjects.
>
> Actually, because of all the problems that you listed, it's not possible to analyze all the data togheter and I was thinking of acting like you suggested by sub-dividing them into smaller groups and by using customized templates in the age range of those sub-groups.
>
> Because the most of subjects were acquired at an age between 5 and 18 I'll try to use TOM atlases and I'll keep those too young in stand by, unless Marko Wilke has useful advice on the processing of children less 2 years.
>
> Thank you for support
> Best regards
> Elisa
>
>
>>> - I'm trying both on healthy subjects and on a population
>>> characterized by a disease, in this case when necessary we alreday
>>> masked alterations to avoid segmentation problems.
>>> In both cases we have a wide age range, from 7 months to 14-17 years
>>> and we're using TPM different by TOMs to cover also the youngest ages
>>> not included in TOM.
>> The TOM toolbox only supports ages from 5-18 years and these data should
>> also work with CAT12 without any larger issues. However, children data
>> younger than 2 years will be difficult or impossible to process with
>> CAT12 even if you have customized TPMs or templates. The changes due to
>> development and the differences to adult brains are simply too large and
>> registration and segmentation will usually fail. If you have even
>> younger subjects the intensities in the T1 image are also affected
>> (inverse) and only customized software can deal with these data.
>>
>> Moreover, it will be difficult at all to process and more important to
>> analyze data in that large age range because the changes due to
>> development will be huge. This will be a general issue if you include
>> subjects younger than 2-5 years (difficult to define a cutoff) because
>> it will be difficult or not meaningful to create a template that fits to
>> all data. If you use different templates you will have a bias due the
>> template selection and cannot analyze the data together.
>>
>> However, if you don’t want to analyze all data together this will be
>> easier to handle and you can use the same customized template for a
>> certain age range. Again, you have to think about excluding the subjects
>> which are too young because of processing issues.
>>
>> Maybe Marko Wilke can also comment because he has more experience with
>> children data and has maybe already tried to process data < 2 years of
>> age.
>>
>> Best,
>>
>> Christian
>>
>>
>>>
>>> - I've already reset the origin in AC both in TPMs and our images to
>>> minimize realignment errors.
>>> I was wondering if using different options of affine regularization as
>>> the "average size template" instead of the default "ICBM space
>>> template", may influence the results about surface.
>>>
>>> Otherwise, since my aim is to obtain values in some specific areas, is
>>> it possible to bypass the default internal atlas, generate surfaces
>>> for the whole brain without separating the two hemispheres and then
>>> use ad hoc templates, in the same space of TPMs, to separate the two
>>> hemispheres and to extract values only from the regions of interest?
>>>
>>> Thank you
>>> Elisa
>>>
>>> On Wed, 9 Nov 2016 09:45:56 +0000, Christian Gaser
>>> <[log in to unmask]> wrote:
>>>
>>>> Dear Elisa,
>>>>
>>>> On Tue, 8 Nov 2016 10:27:33 +0000, Elisa Marchetta
>>>> <[log in to unmask]> wrote:
>>>>
>>>>> Dear CAT experts,
>>>>>
>>>>> I started using CAT12 to calculate cortical thickness and sulcal
>>>>> depth.
>>>>>
>>>>> After some tests on adults' T1 images, I've started using it on
>>>>> pediatric population. To have a more accurate segmentation I changed
>>>>> the default TPM with specific pediatric TPM of corresponding age.
>>>>
>>>> Here I need more information: What type of pediatric population (age,
>>>> any diseases?) and what kind of specific TPM have you used (using TOM
>>>> toolbox, or any other TPM)?
>>>>
>>>>> The segmentation works perfectly and in many cases the surface
>>>>> calculation too.
>>>>>
>>>>> Unfortunately in many other cases, during the surface calculation,
>>>>> something goes wrong and in the results the two hemispheres are
>>>>> badly recognized, with the left hemisphere taking into account also
>>>>> a great portion of the right one and the right hemisphere cut in a
>>>>> bad way.
>>>>
>>>> CAT12 uses an internal atlas to divide the hemispheres and to fill
>>>> subcortical structures and ventricles. If your pediatric data deviate
>>>> too much this internal atlas might not fit very well. However, the
>>>> issues could be also caused by realignment errors. Please can you try
>>>> to set the AC first and run the preprocessing again for a few
>>>> selected subjects, where these issue occured.
>>>>
>>>> Best,
>>>>
>>>> Christian
>>>>
>>>>>
>>>>> I guess the problem is related with age, but sometimes subjects with
>>>>> the same age and similar images have very different results.
>>>>>
>>>>> Do you have an idea on how can I solve this problem?
>>>>>
>>>>> I thank you in advance for any advice about it
>>>>> Best regards
>>>>> Elisa
>>>
--
____________________________________________________
Prof. Dr. med. Marko Wilke
Facharzt für Kinder- und Jugendmedizin
Leiter, Experimentelle Pädiatrische Neurobildgebung
Universitäts-Kinderklinik
Abt. III (Neuropädiatrie)
Marko Wilke, MD, PhD
Pediatrician
Head, Experimental Pediatric Neuroimaging
University Children's Hospital
Dept. III (Pediatric Neurology)
Hoppe-Seyler-Str. 1
D - 72076 Tübingen, Germany
Tel. +49 7071 29-83416
Fax +49 7071 29-5473
[log in to unmask]
http://www.medizin.uni-tuebingen.de/kinder/epn/
____________________________________________________
|