An algorithm for metal streaking artifact reduction in cone beam CT

1
An algorithm for metal streaking artifact
reduction in cone beam CT M. Bazalova1,2, G.
Landry1, L. Beaulieu3, F. Verhaegen1,4 1-McGill
University, Montreal, Canada, 2-Stanford
University, Stanford, CA, Centre Hospitalier Univ
de Quebec, Quebec, QC, Canada, 4-Maastro,
Maastricht, The Netherlands
In the next step, the interpolated projections
were reconstructed using an x-ray imaging
simulation program (ImaSim) developed at our
institute. The distances source-to-isocenter,
detector-to-isocenter, and the data overlap for
the half fan geometry were taken into account.
CBCT images were filtered and backprojected with
the Feldkamp algorithm. To reduce calculation
speed and noise, projection data were downsampled
by a factor of two before the backprojection. The
reconstruction of a single 886886 pixel slice
takes approximately 35 minutes on a 3 GHz
processor. The deleted metals were superimposed
on the reconstructed images based on the
interpolated projections which produced the final
artifact corrected image with the metals.
The patient study was the most challenging test
for our correction method. The identification of
metal traces based on identification of metal
traces in the artifact corrupted images worked
very well, as demonstrated in fig 4a and 4b. Fig
4c represents the interpolated projection and the
vaginal cylinder rod is made invisible.
Introduction
A number of patients undergoing HDR brachytherapy
in our institution are imaged with cone beam
computed tomography (CBCT). CBCT helps identify
patient anatomy and the position of the catheter
or vaginal cylinder in which the radiation source
is inserted and the tumor treated. Image quality
in CBCT can suffer from severe artifacts if
metals, such as vaginal cylinders or hip
prostheses, are present in the patient body.
Metal artifacts in CT have been studied
extensively, however, research in CBCT imaging is
sparse. This abstract presents a correction
algorithm for CBCT metal artifacts based on
sinogram interpolation methods used in CT.
b)
a)
c)
Results
Figure 4 Original patient projection (a), masked
projection (b) and interpolated projection (c).
Materials and Methods
The small water phantom CBCT images suffered from
severe artifacts, as demonstrated with 3D volume
rendering in fig 2a and with an axial slice in
fig 2b. Our reconstruction algorithm produced
images with less, however still significant,
streaking artifacts that closely resemble CT
metal artifacts (fig 2c). The correction
algorithm introduced in this paper significantly
diminished the metal streaking artifacts (fig
2d). The teflon cylinders can be identified more
easily than in the original CBCT images.
Due to the vaginal cylinder geometry, the
interpolation is hardly noticeable. Fig. 5a and
5b show an original and artifact corrected slice,
respectively. The reduction of streaking
artifacts in the vicinity of the cylinder is
evident. The coronal slices at the position of
the vaginal cylinder (fig 5c and 5d ) also
demonstrate the effectiveness of the artifact
correction algorithm. Note that the streaks
caused by the metal rings are not corrected for
since this was not the aim of our method. Only
the projections corresponding to the central rod
were interpolated. To correct for the rings, a
different threshold for identification of metal
voxels in the image domain has to be chosen.
CBCT images of a 15 cm diameter water phantom
with three 2 cm diameter steel cylinders and two
teflon inserts and a 30 cm diameter solid water
phantom with tissue equivalent inserts and with
or without two large steel cylinders mimicking
hip prostheses were scanned on a CBCT scanner
(Simulix, Nucletron). CBCT images of a GYN
patient with metallic vaginal cylinder were also
used in this study. The correction technique was
designed to reduce metal artifacts caused by the
steel cylinders and by the central rod of the
vaginal cylinder. It can be easily modified for
broader applications. Figure 1 Small
water phantom original CBCT projection (a),
projection with deleted metal traces (b),
interpolated projection (c). The Nucletron CBCT
scanner operates with a 120 kV x-ray tube
rotating at a distance of 100 cm from the
isocenter and the x-rays are detected by an
amorphous silicon flat panel with 10241024
pixels of 0.40.4 mm2. The detector-to-isocenter
distance is 52 cm. CBCT images are reconstructed
from approximately 500 views. The scanner can
operate in half fan or full fan modes requiring
two different reconstruction techniques. The
artifact correction algorithm is described here.
First, the metal traces in the projection data
were identified. In the case of the small water
phantom, the metal traces could be segmented
directly in the projection space due to the
simple geometry (fig 1a-b). However, a more
sophisticated approach had to be taken in the
large phantom and the patient studies. Metals
were first segmented in the original
reconstructed images using a fixed threshold,
which worked well in the studied cases. Metal
traces of these voxels in the projection space
were found by projection of the voxels from the
source onto the flat panel and the corresponding
detector signal was deleted. This was done for
each x-ray tube position. Consequently, the
deleted data were filled in using interpolation
of the neighboring data in the direction
perpendicular to the scanner rotation axis (fig
1c). This direction of interpolation is the
optimal direction for correction of artifacts
caused by long objects parallel to the SI
direction, such as vaginal cylinders, tungsten
shielding or hip prostheses.
f)
g)
a)
b)
a)
b)
a)
b)
c)
a)
b)
d)
c)
d)
c)
d)
c)
Figure 2 3D volume rendering of the small
phantom based on CBCT scanner reconstructed
images (a). An axial slice reconstructed by the
CBCT scanner (b), using our reconstruction
algorithm (b) and the original projections and
CBCT image reconstructed using the modified
interpolated projections (d).
Figure 5 Axial original (a) and corrected (b)
slice. Coronal original (c) and corrected (c)
slice. The arrows indicate artifact suppression.
The CBCT images of the large phantom are
presented in fig 3. Fig 3a shows streaking
artifacts around the steel cylinders, however,
the artifacts between the steel cylinders are
less pronounced than in the small phantom.
Nevertheless, the corrected image in fig. 3b
reduces the artifacts in the vicinity of the
steel cylinders and the image quality is similar
to the CBCT image taken without the steel inserts.
Conclusions
The artifact correction algorithm introduced in
this study significantly improves image quality
and enables to define phantom geometry and
patient anatomy in the regions obscured by the
artifacts. The ultimate test of the method will
be correction of artifacts for a patient with
bilateral hip prostheses. This study has the
potential to be translated into the clinic. The
interpolated projections can be uploaded to the
CBCT reconstruction PC and corrected CBCT images
can be reconstructed directly on the scanner
computer. Further investigation into the
resulting image quality is warranted.
a)
b)
c)
Figure 3 Original CBCT image (a), artifact
corrected CBCT image (b) of the large phantom
with steel cylinders and CBCT image of the large
phantom without steel. All images reconstructed
by our algorithm with no scatter correction.