Soft Tissue Visualization Using a Highly Efficient Megavoltage Cone ...

Soft Tissue Visualization Using a Highly Efficient Megavoltage
Cone Beam CT Imaging System
Farhad A. Ghelmansarai, Ali Bani-Hashemi, Jean Pouliot*, Ed Calderon, Paco
Hernandez, Matthias Mitschke, Michelle Aubin*, Kara Bucci*
Siemens Medical Solutions Inc., Concord, CA, USA
*Univ. of California at San Francisco, San Francisco, CA
ABSTRACT
Recent developments in two-dimensional x-ray detector technology have made volumetric Cone Beam CT
(CBCT) a feasible approach for integration with conventional medical linear accelerators. The requirements
of a robust image guidance system for radiation therapy include the challenging combination of soft tissue
sensitivity with clinically reasonable doses. The low contrast objects may not be perceptible with MV
energies due to the relatively poor signal to noise ratio (SNR) performance.
We have developed an imaging system that is optimized for MV and can acquire Megavoltage CBCT
images containing soft tissue contrast using a 6MV x-ray beam. This system is capable of resolving relative
electron density as low as 1% with clinically reasonable doses ranging from 3 to 16 cGy. There are many
factors such as image noise, x-ray scatter, improper calibration and acquisitions that have a profound effect
on the imaging performance of CBCT and in this study attempts were made to optimize these factors in
order to maximize the SNR. A QC-3V phantom was used to determine the contrast to noise ratio (CNR)
and f50 of a single 2-D projection. The computed f50 was 0.43 lp/mm and the CNR for a radiation dose of
0.02cGy was 43. Clinical Megavoltage CBCT images acquired with this system demonstrate that
anatomical structures such as the prostate in a relatively large size patient are visible using clinically
acceptable radiation doses.
Key Words: Megavoltage cone beam CT, Electronic portal imaging device, Megavoltage soft tissue
imaging.
1. Introduction
The goal of the radiotherapy is to deliver the highest possible dose to the tumor while minimizing the dose
to the surrounding tissue. Conformal Intensity modulated radiotherapy (IMRT) and conformal radiotherapy
have been developed to achieve these goals. However, the benefits of these techniques are limited when
large margins have to be used to accommodate uncertainties in target and normal tissue at the time of
treatment. One possible solution is the computed tomography (CT) integrated with the treatment machine.
There are a few approaches available to CT imaging in the treatment position. One such approach
integrates a KV x-ray tube and an a-Si flat panel detector that is mounted on the C-arm gantry of a linear
accelerator. Second approach is to use of the treatment MV beam to obtain CT images either in fan beam or
cone beam scanning geometry.
The use of MV photons to obtain CT images results in a greater absolute dose deposition per photon
compared to KV photons. Therefore, in order to maintain comparable doses for MVCT and KVCT, the
number of MV photons incident on the patient must be considerably reduced and this reduction decreases
the signal to noise ratio. One mitigating factor for MVCT is that the MV photons are more penetrating than
KV photons. Thus, a greater fraction of incident MV photons will travel through the patient and reach the
detector. This effect will provide some compensation for the reduced number of incident photons,
especially for larger patients. Unfortunately, this effect is countered by the increased penetration of the
photons through the detector.

The ability of an x-ray imaging system to differentiate soft tissues is affected by the difference in
attenuation coefficient at a given energy between the contrasting objects. For KV CBCT imaging the
contrast changes with energy, because both the photoelectric and Compton effects contributes significantly
to attenuation. For MV energies the Compton effect is by far the dominant interaction process, especially
for material with low atomic number (Z), in which the attenuation coefficient is proportional to electron
density and is nearly independent of Z. Object contrast is therefore constant over a wide range of energies
in the MV regime. Since the attenuation coefficients of KV photons are higher than MV photons in soft
tissue, the percent density contrasts will be easier to detect at KV energies. Thus, while the higher
attenuation of KV photons results in greater transmission losses, it also benefits KV contrast by the squared
ratio of the mass attenuation coefficient for KV and MV energies 1 .
In this paper we present an implementation of a highly efficient MV CBCT imaging system that can
resolve soft tissue contrast with clinically acceptable doses.
2. Methods and Materials
2.1 Timing Interface for Cone Beam Acquisition Mode
The timing interface uses the Radiation-on (Rad-on) signal of linear accelerator (linac) to generate trigger
pulses that externally control the frame readout of flat panel imager. In this mode, a-Si sensors integrate the
signal during the radiation interval of each projection and the data readout is performed at the end of each
projection. The detector is set in the external mode. Five trigger pulses (5 frames scan) are sent to the
external trigger input of the flat panel detector prior to the start of first projection to force sensor discharge
by reading 5 frames in absence of radiation. This eliminates the dark current accumulation prior to the start
of the first projection. No readout is allowed during the projection interval. Avoiding readout during
exposure interval improves the image SNR by reducing the effect of the electronic readout noise, and also
removes the linac pulsing artifacts from the image.
Rad-on
Trigger
2.2 Image Receptor
5 Refresh Pulses
First
Projection
Projection period
Second
Projection
Figure-1: Timing Interface for Cone Beam Mode
Third
Projection
Data Projection 1 Data Projection 2 Data Projection 3

Two image receptors that were used for this study contained two components; a metal build up
plate/phosphor screen and a flat panel imager. The flat panel imager had an active area of 41 x 41 cm 2 , and
contained 1024 x1024 elements of 0.4 mm pitch. The fastest frame rate was 7 frames per second and the
output data was 16-bit. Flat panel detectors generate frames either using “Free Running” or “External
Triggered”. Free running means that the detector sends out continuous frames according to the selected
frame time. External triggered means that there is an external trigger which allows the generation of the
maskable interrupt request.
The phosphor screen of one receptor (receptor 1) was a 133 mg/cm 2 Gd 2O 2S screen (Lanex Fast) and for
the receptor 2 was an 800 µm CsI (TI) scintillator screen. Both receptors were using a 1mm copper buildup
plate. The charge amplifier capacitance of receptors 1 and 2 were 5pF and 1pF, respectively.
2.3 Cone Beam Image Acquisitions and Corrections
Cone beam CT starts by acquiring a large number of projection images acquired as the X-ray source
follows an orbital path. At every angular increment, i.e., every degree, an image is acquired. 201
projections were acquired over 201 degrees (1° rotation between projections) around the subject. The
gantry speed was set to 109° per minute.
The Siemens PRIMUS® linear accelerator parameters were adjusted to allow the delivery of a small
fraction of a Monitor Unit (MU) per image. A very precise dose control was achieved by loading the
accelerator by very small amount of electrons per pulse. In the current design, the injection is phased with
the RF power at startup or during re-phasing of the injector at the programmed angular increment. With
normal beam parameters, the phasing or re-phasing of the injection generates a shift in the resonant
frequency of the accelerator. Because of the lag of the automatic frequency control (AFC), the beam
formation can be significant (hundreds of milliseconds). In order to improve the beam formation, the beam
parameters were adjusted such that the dose per pulse was reduced to approximately 10% of normal, and
the pulse repetition frequency (PRF) was adjusted to 250 pps. The reduced loading resulted in an
insignificant shift in the resonant frequency of the accelerator, resulting in significant improvement of the
beam formation (less than a few ms). The increased PRF improved the granularity of the control system,
the significance of every dose pulse was reduced, the predictability of the system increased, and the beam
formation time decreased to only several milliseconds.
Three corrections were performed; Offset, Gain, and Defected pixel correction. Offset correction is used to
correct the dark current of each pixel for a specified frame or integration time. The integration time for each
projection is equal to the projection period (figure-1), and therefore, the Offset correction images were
acquired with the frame time that is equal to the projection period.
The Gain correction is used to homogenize the flat field image intensity. The Defected pixel correction
allows a software repair of defected pixels to enhance image quality. Improper pixel values are replaced
with the averaged value of the adjacent pixels, while defected pixels are not averaged.
2.4 Preprocessing by Resolution/Dose Adaptive Filter
A unique adaptive noise removal filter was used for image enhancement 2 . The filter utilizes pixel-wise
nonlinear and adaptive techniques to estimate the additive noise power before filtering. The characteristics
of the filter are adapted to the input X-ray flux and image spatial resolution. Filtering is minimal to the
images that receives high X-ray flux, and increases with the decreasing flux. To minimize the loss of
resolution, minimum filtering is applied to the area of an image with large variance to preserve the edges
and other high frequency components of an image 2 .
Before data normalization and reconstruction, the corrected projection images (Offset, Gain, and Defected
pixel) were processed by the resolution/dose adaptive filter using a local matrix size of 8x8.
2.5 Data Normalization and Reconstruction

The corrected (Offset, Gain, Dead pixel) and pre-processed (resolution/dose adaptive filtering) projection
images were normalized using the pixel data in the open area of the image.
///////
Volumetric reconstruction is accomplished with using an extension of Feldkamp filtered back projection in
cone beam geometry 5 . The reconstruction time for generating 256 3 CT data was ??
2.6 Phantoms for Image Quality Analysis
The followings are the list of phantoms that were used and all were positioned at isocenter. The source to
imager distance (SID) was 140cm for all phantom images.
• A QC-3V phantom 3 was used to determine the spatial resolution and contrast to noise ratio (CNR)
of 2D projection images.
• A ball phantom with a diameter of 16cm consisting of water-equivalent plastic was used for
contrast/spatial resolution experiments. The 1cm thick slice in the middle of the ball contains
several tissue equivalent inserts that are 3 cm in diameter (Figure-2). There are also some inserts
with smaller diameters (1, 0.5, and 0.25 cm). There are air filled holes with diameters of 1, 1.5, 2,
and 3mm within the middle slice of phantom for spatial resolution analysis.
• A Quality Assurance System for Advanced Radiotherapy (QUASAR) body phantom 4 that
contains Electron Density (ED) rods of lung inhale, polyethylene, water equivalent, trabecular
bone, and dense bone with EDs of 0.190, 0.945, 1.002, 1.117, and 1.512; respectively. The ED of
acrylic is 1.137.
• An anthropomorphic head phantom with implanted gold seeds (1mm in diameter and 3mm long)

3.1 2D QC-3V phantom Projections
Figure 2: Contrast resolution phantom
3. Results
Figures 3a and 3b illustrate the QC-3V phantom images acquired with receptors 1 and 2. The QC
images were analyzed by the PIPS-pro 3 software to compute the CNR, f50, f40, f30, and are shown in
Table-1.
Receptor Dose (cGy) CNR f50 (lp/mm) f40 (lp/mm) f30 (lp/mm)
1 0.025 43 0.39 0.50 0.64
2 0.035 69 0.41 0.53 0.66
Table-1: Computed spatial resolution and CNR

(a): 0.025cGy receptor 1 (b): 0.035cGy receptor 2
3.2 Reconstructed Images
3.2.1 Phantom Images
Figure 3: QC-3V phantom images
Figure 4 shows the QUASAR and contrast/spatial resolution (ball) phantom images that were acquired
with receptor 1. The total delivered dose was 18cGy and no preprocessing filtering used. From Figure
4(a) 2% electron density variation between acrylic and trabecular bone is resolvable without applying
the adaptive filtering. Applying the resolution/dose adaptive filter will reduce the standard deviation in
pixel intensities by a factor of ~0.5. Using the ratio of pixel intensity difference (due to density
difference) to root mean square uncertainty in pixel intensities of receptor 1, it can be estimated that
the receptor 2 due to its higher CNR compared to receptor 1 and in combination with preprocessing
adaptive filtering is capable of resolving 1% contrast with radiation doses as low as 10cGy.
The left image on figure 5 illustrates that muscle insert with contrast of 0.5% is resolvable with
receptor 1 using 163cGy dose and without applying dose/resolution adaptive filtering. By using only
the odd projections used for the previous reconstruction (163cGy) that corresponds to the dose value of
81.5cGy (center image of figure 5), the 0.5% contrast is still detectable. The right hand image of figure
5 shows that the minimum required dose in which the 0.5 % contrast becomes discernible with
receptor 1 and by applying the adaptive filter is ~41MU (threshold of detection of 0.5% contrast). Note
that all figure 5 images have the cupping artefacts that are due to the scattered, and no scatter/beam
profile corrections have been applied to these images.
Figures 4(b) and 6(a) illustrate that the 2mm air holes and gold seeds implanted (1mm x 3mm) within
a head phantom are resolvable.
Figure 6 (a) shows axial, coronal and sagittal images of a head phantom that is acquired with receptor
2 by delivering only 1cGy. Figure 6 (b) displays the raw image (no adaptive filtering) of the 2D
projection of this phantom acquired with 0.005cGy. The mean/standard deviation (SD) ratios of this
2D projection (on open field regions) before and after applying dose/resolution adaptive filter are 26
and 68, respectively. This indicates the possibility of the use of receptor 2 for 4D-CT.

(a) 18cGy QUASAR (b) 18cGy contrast/spatial resolution ball phantom
Figure 4: QUASAR and contrast/spatial resolution phantom images with receptor 1 (without preprocessing
adaptive filtering)
Figure 5: Contrast/spatial resolution phantom images with receptor 1. Left: 163cGy, Center: 81.5cGy,
Right: 41cGy with adaptive filtering
(a) 1cGy reconstructed images (b) Projection image with 0.005cGy

3.2.2 Patient Images
Figure 6: Receptor 1Head phantom images with gold seeds
All patient images were acquired at SID of 145cm. Figure 7 shows a pelvis image acquired with
receptor 1. Total dose was 16cGy and preprocessing adaptive filter was applied to the projection
images. The contribution of beam profile and scatter effects that cause shading artefacts are minimized
by obtaining empirical data (from different anatomies, SID, and patient sizes) and linking to the
correction images. Figure 8 compares the axial view of 8cGy MV CBCT with the planning CT image
and shows that the most of the soft tissue structures such as prostate are detectable with MV CBCT.
Figures 9 and 10 depict axial, coronal and sagittal head/neck images that were acquired with receptors
1 and 2 by delivering 16 and 6cGy, respectively. Optic nerves are detectable in both figures. Figure 11
compares the lateral views of MV CBCT images acquired for three different patients over the past year
and shows the improvement in image quality.
One distinct advantage of MV CBCT imaging is that image quality is maintained even when high-Z
materials such as tooth fillings, dental implants, fiducial markers or hip replacements are present. For
kV imaging, these objects create strong artifacts. This is illustrated in the figure 12.
Figure 7: Pelvis MV CBCT with receptor 1 using 16cGy

Figure 8: Axial view of
the pelvic area.
Left: regular CT.
Center; MV CBCT
using 8 cGy. Right: an
insert displays the
MVCBCT image superimposed
on CT
Figure 9: Head/neck MVCB CT with receptor 1 using 16cGy

Figure 10: Head/neck MVCB CT with receptor 2 using 6cGy
Figure 11: The lateral views of MV CBCT images acquired
for three different patients over the past year, which show the
improvement in image quality and required dose.
Left: first MV CBCT clinical image of a Head&Neck patient
Panel RID 1640, dose=8 cGy).
Center: receptor 1, Dose=8 cGy.
Right: receptor 2, Dose=3 cGy.
Figure 12: Images of a Head & Neck patient with dental implants. Left: an
axial view of the planning kV CT shows large artifacts due to the presence
of the metallic objects. Right: the same anatomical area is shown on an
axial view of a MV CBCT acquired with receptor 2. The dental implants
are visible as well as other surrounding normal tissues which are not
visible on the kV CT.

Figure 13 shows neck/chest MVCT axial, coronal and sagittal images that were acquired with
receptors 1 and by delivering 13 cGy.
Figure 13: Neck/chest MV CBCT with receptor 1 using 13cGy
4. Summary
We have developed an imaging system that is optimized for MV and can acquire Megavoltage CBCT
images containing soft tissue contrast using a 6MV x-ray beam with irradiation in range of 3 to 16cGy.
The main components of this system are a high quantum efficient flat panel detector, readout interface
with linac, calibration/corrections process, image preprocessing with dose/resolution adaptive filter,
and scatter corrections. It can resolve a contrast resolution of 1%, gold seeds of 1mm x 3mm (1mm in
diameter and 3mm in length), and air structures of 2mm in diameter.
Clinical MV CBCT images revealed that structures such as prostate and optical nerve are visible with
irradiation of 3 to 8cGy. This system (using receptor 1) can also be used for routine portal imaging
applications without risk of saturation for high dose/high energy treatment/verification imaging, or
dosimetry applications. This system can also deliver good quality MV CBCT images with 1 cGy that
opens the possibility of the use of this system for 4D-CT.
REFERENCES
1. H. Barrett, S. Gordon, and R. Hershel, Statistical limitations in transaxial tomography, Comput. Biol.
Med. 6, 307-323, 1976.
2. F. Ghelmansarai, P. Virador, M. Mitschke, A new dose and resolution adaptive filter, AAPM
(American Association of Physicist in Medicine) 46 th Annual Meeting, 2004.
3. R. Rajapakshe, K. Luchka, and S. Shalev, A quality control test for electronic portal imaging devices,
Med. Phys. 23 (7) 1237-1244, 1996.

4. Modus Medical Devices Inc., QUASAR Body Phantom User’s Guide, May 2001.