[Rtk-users] Fwd: Have you encountered this artifact?

Fri Sep 16 16:37:43 EDT 2016

Thanks for the fast response Simon!

I flipped the angles (360 - angle[deg]) and it worked! Thanks, you were
right all along!
I just didn't get why it makes a difference. I think I do now, as the
resulting image was flipped upside down and not left/right as I expected.
[attached]

The reconstruction is significantly better, I'll look into what should be
included in the reader and what I should keep in my program to keep
conformity with the other readers. Then I'll create a pull request.

Just for the purpose of others hitting the same or a similar bug, I also
attempted:
I did the  SART reconstruction with 10 iterations, lambda=0.3, and Joseph
back/forward projection, *but with no* significant improvement [attached]

And:
If you want you can download the data set from: [Dropbox link to 460MB zip
<https://www.dropbox.com/s/hg2k50vw3f7bt4b/CatPhan.zip?dl=0> (I'll keep it
up as long as Dropbox allows me)] Only the Acquisitions/subfolder is used
along with the Scan.xml (Calibrations folder may be used in the future in
my program, but I'm not sure if you can rely on the existence of the
content).

A MatLab XimReader is available: link
<https://github.com/agravgaard/RTK/blob/master/code/ReadXim.m> (also
available from Varian bitbucket along a with a python version and a
C#->matlab plugin
<https://bitbucket.org/dmoderesearchtools/ximreader/downloads>). Otherwise
my fork with the RTK-style reader is available from the same repository (I
have also added Hnc support, thanks to the Geoff Hugo fork, so bzip2 is a
new dependancy).

Best regards
Andreas

__________________________________

Andreas Gravgaard Andersen

Department of Oncology,

Aarhus University Hospital

Nørrebrogade 44,

8000, Aarhus C

Mail:     andreasg at phys.au.dk

Cell:      +45 3165 8140

2016-09-16 16:13 GMT+02:00 Simon Rit <simon.rit at creatis.insa-lyon.fr>:

> Hi,
> You can try any iterative reconstruction, they can also handle short
> scans. Start with a few iterations of rtksart or rtkconjugategradient.
> However, the nature of the artifacts indicate more a problem in the
> geometry in my opinion. I have seen such errors when, for example, rotating
> in the wrong direction. I can have a look if you share the dataset.
> Cheers,
> Simon
>
> On Fri, Sep 16, 2016 at 2:56 PM, Andreas Gravgaard Andersen <
> andreasg at phys.au.dk> wrote:
>
>> Thanks for the suggestions, Simon and Cyril!
>>
>> I have been carefully looking though the geometry and from what I
>> understand of the transformations matrices, the geometry looks correct/(as
>> expected).
>>
>> HOWEVER: I found out that the reason for the Hnd to behave differently
>> were because had used half-fan scans (full-arc).
>> When I used a full-fan (half-arc) scan of Hnd projections the same
>> artifacts occurs!
>>
>> Are there other (built-in) means of improving half-arc scans, than the
>> parker short scan filter?
>>
>> Parker short scan does a decent job, but the result is still far from the
>> quality of the Varian software reconstruction at least for the CatPhan.
>>
>> Best regards
>> Andreas
>>
>>
>> __________________________________
>>
>> Andreas Gravgaard Andersen
>>
>> Department of Oncology,
>>
>> Aarhus University Hospital
>>
>> Nørrebrogade 44,
>>
>> 8000, Aarhus C
>>
>> Mail:     andreasg at phys.au.dk
>>
>> Cell:      +45 3165 8140
>>
>>
>>
>> 2016-09-14 9:10 GMT+02:00 Cyril Mory <cyril.mory at creatis.insa-lyon.fr>:
>>
>>> One suggestion since it works with the Hnd projections:
>>> You can run rtkprojections twice (with the Hnd projections, then with
>>> Xim projections) and output two projection stack files and two geometry
>>> files, then compare the projection stack files by subtracting one to the
>>> other (with SimpleRTK or clitk) and the geometry files with diff. If they
>>> are identical, then I do not see any reason why the reconstructions should
>>> be different, so my guess is that you will find differences.
>>>
>>>
>>> On 09/13/2016 10:18 PM, Simon Rit wrote:
>>>
>>>> Hi,
>>>> I have almost never worked with Varian data but it looks like a
>>>> geometry problem. Maybe the problem comes from a bad ordering of the
>>>> projections which results in assigning a bad geometry to each
>>>> projection. How did you name your projections? Maybe check that the
>>>> order matches that of the RTK geometry file. Otherwise, there might be
>>>> an issue in the creation of the geometry file itself.
>>>> All this sounds good, happy bug hunt and don't hesitate to share your
>>>> code when you feel it's ready.
>>>> Simon
>>>>
>>>> On Tue, Sep 13, 2016 at 7:06 PM, Andreas Gravgaard Andersen
>>>> <andreasg at phys.au.dk> wrote:
>>>>
>>>>> Dear RTK experts,
>>>>>
>>>>> I am reconstructing Varian ProBeam projections of the Xim image
>>>>> format. I
>>>>> have written the reader myself - very similar to the Hnd one already
>>>>> available with RTK.
>>>>> Links to my fork: [XimReader, XMLReader, GeometryReader]
>>>>>
>>>>> The reader apparently works (Images and angles displays as expected in
>>>>> UI),
>>>>> however when reconstructing with a regular FDK I get a reconstructed
>>>>> image
>>>>> that is smeared out around the high and low density areas [see attached
>>>>> image]
>>>>>
>>>>> I'm using half arc, full fan images with no bow-tie filter from Scripps
>>>>> (~520 projections). Fixed detector and source (offset=0) with SID=2m,
>>>>> SDD=3m.
>>>>>
>>>>> For the Hnd projections the reconstruction works perfectly (Same
>>>>> algorithm).
>>>>> The reconstruction of the Xim projections performed on Varian software
>>>>> works
>>>>> perfectly.
>>>>>
>>>>> Without the Parker Short Scan Filter the first and last projections
>>>>> creates
>>>>> streaks across the reconstruction as if they were way too bright.
>>>>> If the first few projections are excluded, the following projection
>>>>> will act
>>>>> the same way.
>>>>>
>>>>> The projections are corrected for beam hardening and all the
>>>>> projections
>>>>> have the expected attenuation.
>>>>> No "smearing" filters (like median) is used, and iterative
>>>>> reconstruction
>>>>> makes the same artifacts.
>>>>>
>>>>> Setting the value of the first and last projection to zero has the same
>>>>> effect as excluding. Changing the ramp filter only changes noise, not
>>>>> the
>>>>> artifacts.
>>>>>
>>>>> Have any of you had a similar problem? Am I missing something?
>>>>> Any suggestions are welcome I'm running out of ideas.
>>>>>
>>>>> Best regards
>>>>> Andreas
>>>>>
>>>>> __________________________________
>>>>>
>>>>> Andreas Gravgaard Andersen
>>>>>
>>>>> Department of Oncology,
>>>>>
>>>>> Aarhus University Hospital
>>>>>
>>>>> Nørrebrogade 44,
>>>>>
>>>>> 8000, Aarhus C
>>>>>
>>>>> Mail:     andreasg at phys.au.dk
>>>>>
>>>>> Cell:      +45 3165 8140
>>>>>
>>>>>
>>>>> _______________________________________________
>>>>> Rtk-users mailing list
>>>>> Rtk-users at public.kitware.com
>>>>> http://public.kitware.com/mailman/listinfo/rtk-users
>>>>>
>>>>> _______________________________________________
>>>> Rtk-users mailing list
>>>> Rtk-users at public.kitware.com
>>>> http://public.kitware.com/mailman/listinfo/rtk-users
>>>>
>>>>
>>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://public.kitware.com/pipermail/rtk-users/attachments/20160916/661c803e/attachment-0010.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: CatPhan-SART_and_Flipped.png
Type: image/png
Size: 305991 bytes
Desc: not available
URL: <http://public.kitware.com/pipermail/rtk-users/attachments/20160916/661c803e/attachment-0010.png>