[Rtk-users] bright circle on the reconstructed image

[Howard]

Hi Simon,

It appears I have to apologize again and take back the question about the
SART reconstruction.
I re-ran on the big data set (644 projections). This time no memory issue
and SART worked well.
For some reason it did not work on the small set initially. I should have
run more tests before
posting here.

Sorry about the noise and thanks a lot for looking into the "issues".

-howard

On Thu, Sep 18, 2014 at 5:47 PM, Howard <lomahu at gmail.com> wrote:

> Hi Simon,
>
> I apologize for the confusion about the dataset. Actually I did not use
> the same dataset.
> The reason is that the set I used for rtkfdk test had 644 projections and
> with it rtksart
> crashed because of memory (I only allocated 3GB memory to Virtualbox from
> my window7).
> So for the SART test I changed to a small projection set (366 projections)
> where the object is
> a cube with dimension of 5cm x 5cm x 5cm. That's why I chose a smaller
> reconstruction
> size 128*2. With the same dataset when I ran rtkfdk I got very good
> reconstruction.
>
> -howard
>
> On Thu, Sep 18, 2014 at 4:41 PM, Simon Rit <simon.rit at creatis.insa-lyon.fr
> > wrote:
>
>> Hi Howard,
>> Assuming you're using the same dataset as the one you used for the
>> rtkfdk test, you have decreased its size. For example, x was 2*256 mm
>> with rtkfdk and it is now 128*2. On the slice you sent, we could see
>> that a part of the phantom on the right was missing so you should keep
>> the same size except for this side that should be larger, certainly
>> not reduce the size.
>> For the display, I would suggest to use the same window/level as for
>> your FDK reconstruction. The two reconstruction algorithms reconstruct
>> the same average intensities, only the texture is different.
>> Good luck,
>> Simon
>>
>> On Thu, Sep 18, 2014 at 7:03 PM, Howard <lomahu at gmail.com> wrote:
>> > Hi Cyril,
>> >
>> > I am testing SART reconstruction according to what you described,
>> specifying
>> > dimension, spacing and origin:
>> >
>> > rtksart \
>> >    --geometry varianGeometry \
>> >    --regexp .*.hnd \
>> >    --path ../Scan0 \
>> >    --output sart_rec.mha \
>> >    --verbose \
>> >    --spacing 2.0\
>> >    --dimension 128,100,128 \
>> >    --origin -127.5,-99,-127.5
>> >
>> > but the reconstruction gave a plain image with no object/contrast. I
>> checked
>> > the grayscale values on the reconstructed images and did see they ranged
>> > from -1.78 to 3 (this may not be precise but give you some idea). I
>> tried to
>> > adjust window level but couldn't see any contrast. I must miss something
>> > here. Please advise. Thanks!
>> >
>> > -howard
>> >
>> > On Thu, Sep 18, 2014 at 10:59 AM, Cyril Mory
>> > <cyril.mory at creatis.insa-lyon.fr> wrote:
>> >>
>> >> Howard,
>> >>
>> >> For SART reconstructions, you do not need padding, and you can't do
>> Hann
>> >> windowing because no ramp filtering is applied. SART usually outputs
>> >> smoother images, so you probably will not need this denoising trick
>> anyway.
>> >> However, you should be careful about one thing : your object has to be
>> >> fully contained in the reconstructed volume, or you will observe
>> important
>> >> border effects. Use the --dimension, --spacing and --origin options to
>> >> ensure this.
>> >>
>> >> All the best,
>> >> Cyril
>> >>
>> >>
>> >> On 09/18/2014 04:47 PM, Howard wrote:
>> >>
>> >> Thanks very much, Cyril & Simon. Now the fdk reconstruction is working
>> >> beautifully. Jus another quick question: for rtksart executable, are we
>> >> still able to do padding and hanning window trick as well?
>> >>
>> >> On Thu, Sep 18, 2014 at 9:47 AM, Simon Rit
>> >> <simon.rit at creatis.insa-lyon.fr> wrote:
>> >>>
>> >>> Hi,
>> >>> I fully agree with Cyril. Something else : you might want to mask out
>> >>> what is outside the field-of-view using the rtkfieldofview
>> executable. Your
>> >>> ring corresponds to that border. See an example here:
>> >>> http://wiki.openrtk.org/index.php/RTK/Examples/VarianReconstruction
>> >>> Simon
>> >>>
>> >>> On Thu, Sep 18, 2014 at 10:12 AM, Cyril Mory
>> >>> <cyril.mory at creatis.insa-lyon.fr> wrote:
>> >>>>
>> >>>> Hi Howard,
>> >>>>
>> >>>> The bright ring artifact is typically caused by high-pass filtering
>> (the
>> >>>> ramp filter used in FDK) of truncated projection data (when
>> projections have
>> >>>> non-zero values on the sides, the edge of the projection makes a
>> sharp
>> >>>> transition to zero). A classical answer to this problem is to pad the
>> >>>> projections, for example with values starting from the values on the
>> side
>> >>>> and slowly decreasing to zero.
>> >>>>
>> >>>> rtkfdk can do this padding for you. Just add "--pad 0.1" to your
>> options
>> >>>> (or more than 0.1 if it isn't enough).
>> >>>> Also, I noticed that your reconstructed image seems quite noisy. You
>> >>>> might want to use a Hann windowing of the ramp filter to reduce the
>> noise
>> >>>> amplification effect of high-pass filtering. You can do that by
>> adding
>> >>>> "--hann 0.5" to your options. Note that the argument is a cut-off
>> frequency,
>> >>>> so smaller values mean more smoothing.
>> >>>>
>> >>>> Best regards,
>> >>>> Cyril
>> >>>>
>> >>>>
>> >>>> On 09/18/2014 05:21 AM, Howard wrote:
>> >>>>
>> >>>> Hi,
>> >>>>
>> >>>> I use rtkfdk to reconstruct varian cbct images following the example
>> on
>> >>>> rtk wiki page. The projections were obtained with half fan mode
>> because it
>> >>>> is an abdominal region with the patient lateral x AP dimension of
>> 550x550.
>> >>>> The reconstructed cbct images have a big bright circle around
>> >>>> the image. Please see the image:
>> >>>> http://expirebox.com/download/2f55c9d8f0bc83138b954d2fec08e78c.html
>> >>>>
>> >>>> I am using Ubuntu Linux box to reconstruct. Two sequential commands
>> were
>> >>>> used:
>> >>>>
>> >>>> ./rtkvarianobigeometry --xml_file ProjectionInfo.xml --path Scan0/
>> >>>> --regexp Proj_.*.hnd -o varianGeometry
>> >>>> and
>> >>>>
>> >>>> rtkfdk \
>> >>>>    --geometry varianGeometry \
>> >>>>    --regexp .*.hnd \
>> >>>>    --path ../Scan0 \
>> >>>>    --output rec.mha \
>> >>>>    --verbose \
>> >>>>    --spacing 2.0,2.0,2.0 \
>> >>>>    --dimension 256,100,256 \
>> >>>> The projections are stored in the standard varian format: Scan0 and
>> >>>> ProjectionInfo.xml with 644 projection images.  I assume that rtkfdk
>> handled
>> >>>> the half fan geometry properly. What did I do wrong then?  Many
>> thanks!
>> >>>>
>> >>>> -howard
>> >>>>
>> >>>>
>> >>>> _______________________________________________
>> >>>> Rtk-users mailing list
>> >>>> Rtk-users at public.kitware.com
>> >>>> http://public.kitware.com/mailman/listinfo/rtk-users
>> >>>>
>> >>>>
>> >>>> --
>> >>>> --
>> >>>> Cyril Mory, Post-doc
>> >>>> CREATIS
>> >>>> Leon Berard cancer treatment center
>> >>>> 28 rue Laënnec
>> >>>> 69373 Lyon cedex 08 FRANCE
>> >>>>
>> >>>> Mobile: +33 6 69 46 73 79
>> >>>>
>> >>>>
>> >>>> _______________________________________________
>> >>>> Rtk-users mailing list
>> >>>> Rtk-users at public.kitware.com
>> >>>> http://public.kitware.com/mailman/listinfo/rtk-users
>> >>>>
>> >>>
>> >>
>> >>
>> >> --
>> >> --
>> >> Cyril Mory, Post-doc
>> >> CREATIS
>> >> Leon Berard cancer treatment center
>> >> 28 rue Laënnec
>> >> 69373 Lyon cedex 08 FRANCE
>> >>
>> >> Mobile: +33 6 69 46 73 79
>> >
>> >
>>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://public.kitware.com/pipermail/rtk-users/attachments/20140918/d667978b/attachment-0009.html>