[caret-users] too many crossovers

[David Van Essen]

Bela,

Donna's comments are on the mark.  Two additional points:

1) As you evaluate landmarks, use the 'Show First Link Red' option to  
make sure that landmarks are not only in a reasonable location  
relative to other landmarks but also are oriented consistently in the  
source and target surfaces.

2) A general question:  Is there a particular reason you are using  
'Colin' as the registration target?  In general, we encourage  
registration to the PALS-B12 atlas unless there is a specific reason  
for using a different target?

David VE

On Nov 19, 2008, at 3:46 PM, Donna Dierker wrote:

> Hi again Bela,
>
> There are a couple of instances where this problem exists:
>
>> there's no way to topologically match borders without twisting the  
>> surface, because landmarkA is right of landmark B in source, but  
>> vice versa in target.
> Compare the attached captures:
>
> source_lateral.jpg with target_lateral.jpg : green & brown borders  
> on posterior end of sylvian border in source; on anterior end in  
> target
> source_ventral.jpg with target_ventral.jpg : magenta to upper right  
> of blue in source; on upper left in target
>
> Imagine the twisting that will be needed to yank them into  
> alignment.  It won't work.
>
> Donna
>
> On 11/19/2008 01:45 PM, Donna Dierker wrote:
>> Hi Bela,
>>
>> I haven't looked at your target dataset, but right away I can see  
>> some impending nightmares in your source dataset borders.
>>
>> The attached captures show junctions where separate landmarks meet  
>> or come into close proximity with one another.  This spells  
>> trouble.  You want some spacing at the landmark junctions.  I used  
>> to get crossovers in the medial wall until I following the core 6  
>> landmark guidelines on trimming the medial wall near the frontal  
>> and calcarine junctions (http://brainvis.wustl.edu/help/landmarks_core6/landmarks_core6.html/#mw 
>> ).  Obviously, you're doing something more customized (not using  
>> core 6 landmarks, using colin as target), but the same principle  
>> applies.
>>
>> See what trimming some points (on both source and target) near  
>> these junctions does for reducing your crossovers.
>>
>> Other causes of crossovers include:
>>
>> * errors/defects in the spheres themselves, especially along the  
>> landmarks
>> * big spacing in the landmark border points, which can happen if  
>> you project borders from a flat surface (instead of drawing on  
>> flat; projecting; and appending to in-tact medial wall/calcarine  
>> borders on a closed topo surface)
>> * oddities in the borders besides hooks/kinks/reversed polarity:  
>> closted/nested borders; borders crossing over each other; there's  
>> no way to topologically match borders without twisting the surface,  
>> because landmarkA is right of landmark B in source, but vice versa  
>> in target.
>> * misidentifying the hemisphere in the spec files
>>
>> Donna
>>
>> On 11/19/2008 12:29 PM, Bela Bhatia wrote:
>>> Donna,
>>>
>>> I checked by spinning the borders and they look ok - no kinks or  
>>> sharp turns.
>>> I have uploaded 2 zip files
>>> NPRArchive.tar.gz
>>> Right_Hem_Atlas_11_final.tar.gz
>>>
>>> What are the variables that affect crossovers during spherical  
>>> Registration?
>>>
>>> Thanks for your help!
>>>
>>> Bela
>>>
>>> Donna Dierker wrote:
>>>
>>>
>>>
>>>> Hi Bela,
>>>>
>>>> One thing you can do is view your source (and target, if non- 
>>>> standard) borders on the source sphere.  Then, select D/C:  
>>>> Surface Miscellaneous: Drawing mode: Hide surface.  Spin the  
>>>> sphere around (or use the toolbar buttons to switch from medial,  
>>>> lateral, etc.), and make sure there are no hard right turns or  
>>>> kinks in any of your borders.  The medial wall ventral is a  
>>>> common trouble spot for hooks/kinks.  Make sure you turn Drawing  
>>>> Mode back to Tiles and lighting when you're done inspecting the  
>>>> borders.
>>>>
>>>> If you see no problems, then send me a zipped archive of your  
>>>> source dataset (including all the files selected on the  
>>>> registration menu).  Also, either enumerate the source settings  
>>>> or give me a screen capture of the source tab on the registration  
>>>> dialog, so I know which files you selected.  Upload the zipped  
>>>> archive here:
>>>>
>>>> http://pulvinar.wustl.edu/cgi-bin/upload.cgi
>>>>
>>>> If you're using a standard atlas target, like the PALS_B12.LR  
>>>> borders, then just say so and no upload is necessary for the  
>>>> atlas target.  Otherwise, I'll need that, too.
>>>>
>>>> I'll try it on my end and let you know what I find.
>>>>
>>>> Donna
>>>>
>>>> On 11/19/2008 11:27 AM, Bela Bhatia wrote:
>>>>
>>>>> Hello,
>>>>>
>>>>> I am getting too many crossovers while processing spherical  
>>>>> registration.
>>>>>
>>>>> The border starting points are all corect.
>>>>>
>>>>> What else can I do?
>>>>>
>>>>> Thanks
>>>>> Bela
>
> < 
> source_lateral 
> .jpg 
> > 
> < 
> source_ventral 
> .jpg 
> > 
> < 
> target_lateral 
> .jpg 
> ><target_ventral.jpg>_______________________________________________
> caret-users mailing list
> caret-users at brainvis.wustl.edu
> http://brainvis.wustl.edu/mailman/listinfo/caret-users