[Insight-users] Loading DICOM RTStructs

Fri Oct 16 22:35:46 EDT 2009

Hi Jason,

My C# solution does exactly the same thing, parsing the file into 2d
planes that are then merged into a 3D volume.  There are tricks that
I've found; for instance, structs read into Mimvista's software came
out horizontally flipped after anonymization; I'm not sure if that was
a result of the anonymization process or if that's just something that
Mimvista does.  In any event, my code contains a flag to check whether
or not an image was made using mimvista's software and then flips the
contour accordingly.

Just looking through your code briefly, does it create a mask with a
filled polygon, or an outline?  I'm not familiar enough with itk to be
able to interpret the insertRegion code, or what
GetLargestPossibleRegion does in the way you've set things up.  Since
I didn't have the itk stuff (which is probably doing very nice polygon
handling), my code includes a large chunk on filling in the polygon.
Some RTstruct files have the polygon described such that each point is
about 2 pixels from every other point, regardless of whether all the
points are on a straight line or not, while some have the most
geometrically simple version of the polygon.  I'm not sure how that
would affect your code, so I'll take a look at the files I have.

I've attached my project, which does as you describe but in C# and
using GDCM.  It also has a linear and cubic interpolator that I
borrowed (and the copyright's in that code) to do image resizing; one
of the real reasons to adopt itk was to have better resizing
functions.  If you want to run the results of this code, just put the
gdcm dlls into the directory with the executable, or have them as part
of the path.

When I make production code, I put the dll's into an include
directory, and they are automatically copied from there, so that when
I get a new version of a library, the code is built and the newly
compiled library is put into the includes directory.  I'm looking for
something similar with itk.  One of the big reasons I used C# in this
project is that C# does not require anything other than the compiled
library, so I didn't have to track down the header file locations of
what I was trying to use.  The major drawback is that the C# layer
doesn't appear to work properly on 64 bit platforms, something I
haven't had a chance to debug but I suspect lies with the swig
translation layer.

As you can see from my code, I really dislike monolithic files.  It's
just a stylistic choice, but the CMake paradigm adopted in itk
definitely seems to favor single-file solutions.  Is there a non-CMake
way to make an itk project?  I need the header files to link against,
not just the dlls and libs; is it safe to assume that all the
necessary files are in itk/code/common?  And is there a reason that
only the common library is a dll, and all the others are statically
linked libraries?

Thanks,
Mark

>
> On Fri, Oct 16, 2009 at 4:30 PM,  <Jason.Dowling at csiro.au> wrote:
>> Hi Mark,
>>
>> No I don't use an itk mesh.  I needed a binary volume for each structure  (you can then mesh your binary volume if you want).  In the DICOM RT file each each structure is represented as a set of points for a single slice (nb there might be > 1 sets of points for a single contour in a single slice, eg for bone).   I work on  a  temporary 2D slice which is inserted back into the 3D volume after it's updated.   I use some of Mathieu's code from the excellent GDCM 2.0 library  to extract the RT points.   For each structure region , I  convert the physical RT points into image space and generate pointlist,  which are used to generate  polygon spatial objects for each contour .  Then I use the SpatialObjectToImageFilterType to copy that structure component onto the 2D slice.  And iterate through the RT file:  at the end of I have one binary volume for each structure.
>>
>> If anyone's interested, last year I also wrote an itk RTOG file parser which does a similar thing.
>>
>> I've sent the code and test data to you - please let me know if you have any feedback?
>>
>> Cheers,
>> Jason
>>
>> ________________________________________
>> From: Mark Roden [mmroden at gmail.com]
>> Sent: Saturday, 17 October 2009 1:54 AM
>> To: Dowling, Jason (ICT Centre, Herston - RBWH)
>> Cc: mathieu.malaterre at gmail.com; insight-users at itk.org; gdcm-developers at lists.sourceforge.net
>> Subject: Re: [Insight-users] Loading DICOM RTStructs
>>
>> Hi Jason,
>>
>> That would be very useful, thanks.
>>
>> My code, as I said, is written using the C# wrappers to gdcm.  It then
>> converts the masks into a char image, where a 1 indicates the presence
>> of a mask and a 0 does not, and I was going to implement a marching
>> squares algorithm to go back to an rtstruct.  I'm guessing that your
>> code goes to an itk mesh; is there a way to convert a mesh into an
>> image and back again?  I'm messing around with segmentations, and the
>> papers I'm finding are all pixel-based rather than mesh-based, so
>> getting the mesh is just part of the problem.
>>
>> Thanks!
>> Mark
>>
>> On Fri, Oct 16, 2009 at 12:27 AM,  <Jason.Dowling at csiro.au> wrote:
>>> Hi Mathieu and Mark,
>>>
>>> Yes, I've written a C++ DICOM-RT Struct importer which uses ITK and GDCM 2. I've been testing it on MR and CT volumes of the pelvis (I have also testing it on some head and neck and chest volumes) over the past few weeks.
>>> I have been planning to write an Insight Journal paper over the past week and will try to do this over the weekend.  I'm happy to send you the code today (as is) if you would like to test/compare it?
>>>
>>> Cheers,
>>> Jason
>>>
>>>
>>> -----Original Message-----
>>> From: Mathieu Malaterre [mailto:mathieu.malaterre at gmail.com]
>>> Sent: Friday, October 16, 2009 6:13 PM
>>> To: Mark Roden
>>> Cc: insight-users at itk.org; Dowling, Jason (ICT Centre, Herston - RBWH); gdcm-developers
>>> Subject: Re: [Insight-users] Loading DICOM RTStructs
>>>
>>> Hi Mark,
>>>
>>>  I made a mistake and had a private conversation with Jason Dowling (CC), without CCing gdcm-dev sorry :(
>>>  He also has been working on a C++ RT Importer using gdcm 2.x. This has been tested on multilpe platform and gives proper results.
>>>  If you would like we could compare both C++ code and merge them. I know Jason was really close to making a release to the IJ. Jason, any comments ?
>>>
>>> cheers
>>>
>>> On Fri, Oct 16, 2009 at 1:43 AM, Mark Roden <mmroden at gmail.com> wrote:
>>>> Hi all,
>>>>
>>>> I note that there is now the capability of exporting DICOM RT structs:
>>>>
>>>> http://www.insight-journal.org/browse/publication/316
>>>>
>>>> However, I can't find any information on how to import these structs.
>>>> It may just be that my google-fu is weak, so if there is such an
>>>> example or the like, please let me know.
>>>>
>>>> I've written C# code to wrap around gdcm to go into my own image
>>>> format, so I can use that if I have to, but the conversion would
>>>> probably be irritating since I'm extremely inexperienced with the itk.
>>>>  I'd rather follow some sample code.
>>>>
>>>> Barring that, should I just post my importer to the development list
>>>> to get some help on converting it into an itk C++ class?
>>>>
>>>> Thanks!
>>>> Mark
>>>> _____________________________________
>>>> Powered by www.kitware.com
>>>>
>>>> Visit other Kitware open-source projects at
>>>> http://www.kitware.com/opensource/opensource.html
>>>>
>>>> Please keep messages on-topic and check the ITK FAQ at:
>>>> http://www.itk.org/Wiki/ITK_FAQ
>>>>
>>>> Follow this link to subscribe/unsubscribe:
>>>> http://www.itk.org/mailman/listinfo/insight-users
>>>>
>>>
>>>
>>>
>>> --
>>> Mathieu
>>>
>>
>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: dcmtoblock160ct2009.zip
Type: application/zip
Size: 30272 bytes
Desc: not available
URL: <http://www.itk.org/pipermail/insight-users/attachments/20091016/d571bc7d/attachment-0001.zip>