Monte Carlo dose calculation accuracy in 192Ir

1
Monte Carlo dose calculation accuracy in 192Ir
Brachytherapy based on CT and cone-beam CT
images D Goodfellow, A Prichindel, M Bazalova, E
Poon, B Reniers, F Verhaegen McGill University
Health Center, Montreal
2
Introduction With the advantage of Brachytherapy,
that it can deliver highly conformal dose
distributions in a target volume with much less
damage to the healthy surrounding tissue, there
is a need for a more accurate CT image
calibration for individual patient dose
assessment. Conventional clinical calibration
techniques involve assigning uniformly both the
material composition and density of water to the
entire patient geometry. It is based on these
calibrated images that dose is then calculated.
The effect of improperly assigning material types
and densities needs to be accurately quantified
in order to motivate to what extent calibration
techniques need to take place and for what
materials calibration is of greatest importance.
3

• This work
• In this work we investigated the accuracy of
  different calibration and
• material segmentation methods in determining
  accurate dose for an
• 192Ir source in Brachytherapy. To this end,
  we studied
• CT images of Gammex RMI465 Electron density CT
  Phantom obtained
• using a Picker PQ5000 CT scanner and dose
  calculations were
• performed based on Monte Carlo simulations
  using DOSXYZnrc code.
• We studied
• 4 methods (i) uniform water phantom
• (ii) water phantom with densities
  based on 4 calibration
• curves
• (iii) a 4 material segmentation and
  density calibration
• (iv) correct material assignment
  and density calibration.
• the applicability of the Nucletron Simulix Cone
  Beam CT Scanner for

4
Calibration and Experimental Methods This
research focuses on how to properly calibrate CT
scans and design a material segmentation
technique for Brachytherapy treatment planning.
To quantify the errors due to calibration we used
a solid water calibration phantom with inserts of
various tissue-equivalent materials and scanned
this with a Picker PQ5000 CT scanner at 100 kVp,
120 kVp, 130 kVp and 140kVp and 200mA. We then
performed 4 different calibration and material
segmentation techniques uniform water phantom,
water phantom with calibrated density using
calibration curves based on 4 density groups, a 4
material phantom and density calibration, correct
material assignment but density calibrated based
on 4 density groups.
Fig 1. Calibration graph at 100 kVp (left) and
140 kVp (right)
5
Monte Carlo simulations using DOSXYZnrc were then
run on all test phantoms with the source placed
in 4 different locations shown below
. The accuracy of the dose in the ith
voxel for the 4 calibration techniques was then
quantified by using the following formula for
each image
(1) Where
and represent the dose in the exact phantom
and the dose in the test phantom of the ith voxel
respectively. When this was applied to the whole
image a dose accuracy map was obtained. The same
treatment was used for images obtained from a
Nucletron Simulix cone beam CT scanner.
Fig 2. The 4 locations of the source (N, E, S, W)
6
Water Phantom In the water phantom we noticed
very high dose differences of up to 153 in some
cases, particularly in the vicinity of lung and
bone materials.
Fig 3. Dose accuracy map for non-density
calibrated (left) and density calibrated (right)
water phantom with source positioned East (right)

• For most inserts the introduction of calibrated
  densities (Fig. 3 right) reduced the
• dose difference by half (up to a maximum dose
  difference of 101).
• Eliminated the large dose difference trails we
  see in the uncalibrated
• (Fig. 3 left) phantom reduced to 21.

7
4-Material Phantom For most inserts, the
segmentation into 4 bins (air, lung, tissue,
bone) results in a dose difference below 2 .
Fig 4. Dose accuracy map for 4 material phantom
at 140 kVp energy with source positioned North
(top), East (right), West (left) and South
(bottom)
In tissues such as lung and brain, where dose
accuracy dropped to 1.00.6 and 0.60.4
respectively, material segmentation had a
positive effect. Conversely higher density
inserts such as CB2 30 , cortical bone, inner
bone and bone mineral resulted in dose accuracies
ranging from 40.6 to 205, due to a
mis-assignment of media.
8
7-Material Phantom Within distances of 7 cm from
the source location the dose accuracy was below
2 with the only exception being the lung insert
where the inhomogeneity of the insert led to dose
differences just above 2.
Fig 5. Dose accuracy map for 7 material phantom
at 140 kVp energy with source positioned North
(top), East (right), West (left) and South
(bottom)
9
100kVp Calibration Curves Applied on 140KVp image
We were interested in investigating the impact of
neglecting to create a calibration curve for each
scanning energy. We therefore calibrated the
densities of a CT phantom, whose image was
acquired at 140 kVp, with the calibration curve
designed for 100 kVp images.
Fig 6. Dose difference of CT image acquired at
140 kVp with calibration curve
of 100 kVp (left) Density
calibration curves at 100 kVp and 140 kVp (right)
Overall we noticed dose differences below 2
everywhere. The highest discrepancy can be seen
in a dose difference trail of approximately -11
caused by the cortical bone cylinder (SB3). This
is because the use of a calibration curve
designed for higher CT numbers results in an
underestimation of cortical bone density as seen
below in Fig.6 (right).
10
Summary Histograms The dose accuracy values given
in Figure 7 represent the maximum dose difference
in each cylindrical insert
Fig 7. Dose accuracy (including statistical
error) for all density calibration and material
segmentation techniques
Liver (LV1), resin (CB3) and brain (SR2) do not
need any form of density calibration or material
segmentation since by assigning to them the
density of water (and material type water) they
result in dose accuracy below 2. The 4 material
segmentation technique is sufficient to keep dose
accuracy below 2 for dense inserts. High density
tissues such as bone inserts can only result in
dose accuracy below 2 if correct material
composition is assigned to them. inserts can only
result in dose accuracy below 2 if correct
material composition
11
Fig 8. Dose fluctuations within inserts for all
density calibration and material segmentation
techniques (for all material inserts)
Dose differences within the cylindrical inserts
in 7 material calibration do not fluctuate by
more than 1 and thus we conclude that CT image
quality contributes to no more than 1 to the
dose differences. The water phantom method
however gives rise to higher fluctuations arising
from the gradient behaviour (dose difference
increasing as we go trough the cylinder) .
12
Cone Beam CT Scanner in Treatment Planning To
obtain the Cone Beam CT image we used the wider
5cm beam that produces images with a consistently
lower standard deviation in materials of uniform
density and more accurate HU values. The 4
material calibration was performed and a Monte
Carlo simulation was then carried out. The
accuracy map is displayed below
Figure 8. Dose accuracy map for 4 material
phantom scanned with a Cone Beam CT scanner with
source positioned East (right)
The average accuracy in the SB3, BR12, SR2 and
LN300 cylinders were 0.460.14, 0.90.5,
1.30.6 and 1.20.6 respectively, all of which
are under 2.
13

• Conclusions
• In this work MC simulations were used to
  determine the importance of various calibration
  techniques in brachytherapy. We reach the
  following conclusions
• The failure to calibrate for densities and
  material types in a water phantom leads to errors
  up to 153
• Liver (LV1), resin (CB3) and brain (SR2) can be
  assigned the composition and density of water.
• 4 material segmentation technique is sufficient
  to keep dose accuracy below 2 for low density
  materials (i.e. lung).
• High density tissues (i.e. bone inserts) require
  additional material segmentation to obtain dose
  accuracies below 2.
• The CT image quality contributes to no more than
  1 to the dose differences
• Early simulations with Cone Beam CT images
  suggest that there is potential for this to be
  used as a treatment planning tool in the future.

References 1. Brauner, M. and Neidhardt, J. Y. 
Surgical and Radiologic anatomyVolume 11, number
1/ March,1989 Springer Paris 2. Nath, Ravinder
Overview of Brachytherapy Physics Yale
University School of Medicine New Haven,
Conneticut. 3. Verhaegen, F. and Devic, S
Sensitivity study for CT image use in Monte Carlo
treatment planning, Phys. Med. Biol. 50, 937-946
(2005).