[Rtk-users] Fwd: High pixel/voxel values in SART

Chao Wu wuchao04 at gmail.com
Wed Feb 21 06:57:58 EST 2018


L.S.,

I was working on FDK in the past and interative reconstruction methods are
still new to me.
I understand the concept of iteratvie methods but are not aware of
technical details in implementation.

Recently I am trying SART but got streak artefacts in reconstructed slices,
as well as dots with very high value (both negative and positive) at
corners of slices.
When I checked intermediate images in the pipleline I found that those are
introduced in itk::DivideOrZeroOutImageFilter.
You can see from the attached picture: the left half shows the output
of rtk::RayBoxIntersectionImageFilter and the right half the output of
itk::DivideOrZeroOutImageFilter,
both during processing of the first projection in the first iteration.
Apparently, although it contains the whole object, my volume is relatively
small compared to the size of the detector images.
Then the rays intersecting the volume near corners and edges result in
small values in the output of the raybox filter, and subsequently magnify
the pixel values largely after division.
This may not be a problem if the detector images are noiseless, but in
practice this will magnify the noise and they will stay as streaks and dots
in slices.

To correct for this I have something in mind, such as making the volume
bigger and cropping the detector images so that corners and edges of the
volume do not project to the cropped detector; or increasing the threshold
in the divide filter so that low values from edge/corner rays wll be zero
out. Since I am lack of experiences in interative methods, my question is
what the best or common practice will be to handle this? Thanks a lot.

Regards,
Chao
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://public.kitware.com/pipermail/rtk-users/attachments/20180221/9cc6f406/attachment-0001.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: raybox&divide1.png
Type: image/png
Size: 292092 bytes
Desc: not available
URL: <https://public.kitware.com/pipermail/rtk-users/attachments/20180221/9cc6f406/attachment-0001.png>


More information about the Rtk-users mailing list