Acknowledgements
Monte Carlo-based Clinical Treatment
Planning: Issues for Consideration
• E. Heath, G. Stroian, K. Al-Yahya, W. Parker, F.
Verhaegen
• F. Deblois and A. Alexander: MMCTP GUI
• Canadian Institutes of Health Research, National
Cancer Institute of Canada and Natural Sciences and
Engineering Research Council for grant funding &
salary support (CIHR, NCIC, NSERC)
Indrin Chetty and Jan Seuntjens
Univerisity of Michigan, Ann Arbor, MI
and
McGill University, Montreal, QC
AAPM Summer School 2006
Outline: “issues for consideration”
•
•
Building blocks of a generic clinical MCTP system
Verification issues in general
•
•
•
•
CT calibration & artifacts
Dose specification
Absolute calibration, MU calculations
Verification issues in detail
•
•
•
•
•
–
Validation of relative output
–
Experimental verification of MCMC-based dose algorithms: beam
modifiers (MLC)
Experimental verification of transport within the phantom/patient
phantom/patient
–
2
Plan 1 in a lung cancer treatment
Dose (Gy)
__ 4-8 __ 8-12 __ 10-20 __ 20-30 __ 30-36 __ 36-38 __ 38-40 __ 41-42 __ 42-
(a)
(b)
EqTAR corrected
Monte Carlo
MCMC-based treatment planning: comparisons of MC versus simple
(correction(correction-based) and modelmodel-based algorithms
Statistical Uncertainties in MCMC-based treatment planning
MCTP and lung (NSCLC)
Retrospective re-planning & outcome association
Summary
Difference map
for PTV area
Outcome?
(c)
AAPM Summer School 2006
treatment plan or
treatment delivery
CT slices
composition of
(e)density matrix
3
Beam information
extractor
Patient modifications:
CT couch removal,
bolus,
CT artifact-cleanup
Validation
beam
configuration
Primary beam
Beam modeling
Beam calculation
Phasespace or source
model for each treatment
field
Absorbed dose to
water calculation
patient Monte Carlo
MC-plan displayed
in popular planning system
Patient dose
calculation
AAPM Summer School 2006
Beam modifiers
Reference and
relative “output” in
defined phantoms
Relative “output” in
patient geometries
with defined materials
6
Validation - Wedge
100
Rel. Dose (%)
80
45 degrees wedge profile, 18 MV Photon Beam
140
3 cm (Measured)
3 cm (XVMC)
3 cm (dosxyz)
120
100
MC
film
CA24
60
40
20
0
-8
-6
-4
-2
0
2
4
6
8
Crossplane (cm)
80
100
60
80
Rel. Dose (%)
40
20
0
-15
-10
-5
0
5
10
Dynamic IMRT Delivery
15
MC
Film
CA24
60
40
20
Off-axis position (cm)
0
-8
Wedge factors
-6
-4
-2
0
Inplane (cm)
2
4
6
8
Output factors
15x15 cm2 field size
Wedge angle
15
30
45
Measurement
0.755
0.623
0.514
Simulation
0.754
0.621
0.514
Uncert. (%)
0.8
0.8
0.8
Difference (%)
-0.1
-0.2
0.0
5x5 cm2 field size
Wedge angle
15
30
45
60
Measurement
0.756
0.616
0.512
0.422
0.421
Simulation
0.765
0.609
0.518
Uncert. (%)
0.7
0.9
0.7
0.7
Difference (%)
0.9
-0.7
0.6
-0.1
Field size
5x5
10x10
15x15
20x20
measured
0.919
1.000
1.036
1.058
simulation
0.918
1.000
1.030
1.055
difference
0.001
0.006
0.003
AAPM Summer School 2006
Reference dose calibration
Reference dose calibration (cont’d)
DwMC (10 × 10,SSD = 100)
= k MC [Gy / particle]
particle
Model-based system: fluence-based calibration
Dwref (10 10,SSD = 100)
= k [Gy /MU]
U
MC calculated reference dose to water per particle
(fluence);
Particle fluence at an accelerator reference point (usually
upstream from monitor chamber)
U: a monitor unit (quantity with dimension MU)
Dw dose to water at the clinical reference point
k = 0.01 Gy / MU.
If accelerator tweaked in terms of dose to tissue:
then k = 0.0101 Gy / MU.
AAPM Summer School 2006
10
11
AAPM Summer School 2006
12
Reference dose calibration (cont’d)
Reference dose calibration (cont’d)
MC
⎞
D MC (r) ⎛ k
(r)
Dtissue
[Gy/MU] = tissue ⎜
[particle/MU]⎟
U
particle ⎝ k MC
⎠
MC
⎞
D MC (r) ⎛ k
(r)
Dtissue
[Gy/MU] = tissue ⎜
[particle/MU]⎟ B(x, y)
U
particle ⎝ k MC
⎠
The B(x,y) factor accounts for the change in output due to
monitor backscatter.
MC calculated reference dose to water per particle
(fluence);
Particle fluence at an accelerator reference point (usually
upstream from monitor chamber)
1. Relatively small effect (<1%
for 10x10 and smaller)
2. Effect reduces for higher
energies.
3. Effect is smaller for electrons.
4. Varian series only.
This approach ignores backscattering of moving
components into the monitor chamber!
AAPM Summer School 2006
treatment plan or
treatment delivery
CT slices
AAPM Summer School 2006
13
composition of
(e)density matrix
Verhaegen et al 2000, Phys Med Biol
45, 3159 - 70
14
Photon beams: 6 MV, 15 MV, 250 kVp
beam
configuration
Electron beam: 18 MeV
No contaminant particles
Beam information
extractor
Patient modifications:
CT couch removal,
bolus,
CT artifact-cleanup
Beam calculation
Phasespace or source
model for each treatment
field
CT scan
patient Monte Carlo
MC-plan displayed
in popular planning system
Exact geometry
HU interval
ρ interval
Air
[-1000 : -950]
[0.001 : 0.044]
Lung
[-950 : -700]
[0.044 : 0.302]
ICRU tissue
[-700 : 125]
[0.302 : 1.101]
ICRP cortical bone
[125 : 2000]
[1.101 : 2.088]
Material
Dose calculation:
DOSXYZnrc
Courtesy: Frank Verhaegen
AAPM Summer School 2006
16
6 MV photons
2.1
100kV
120kV
140kV
ctcreate ramp
Catphan 100 kV
bonefit 100kV
-3
mass density (g.cm )
1.9
≈ 2% underdose in CTMC
calculation
1.7
1.5
1.3
(5)
1.1
( μen/ρ)w > ( μen/ρ)soft tissue
(8)
(3)
0.9
(10)
(1)
Region 1: large
overestimation of ρ
(9)
(4, 6, 7)
(2)
-300
0.7
-100
100
300
500
700
900
HU
Dose
(Dexactin
–CTMC
DdefaultMap
geometry
)/ Ddefault
Material
(HU, ρ) relation
AAPM Summer School 2006
Courtesy: Frank Verhaegen
17
AAPM Summer School 2006
Courtesy: Frank Verhaegen
18
CT artifacts, editing
Conclusion:
mis-assignment of media and ρ can cause significant dose errors
Question:
Does using only water with ρ derived from (HU,ρ) worse than risking mis-assignment
of media?
6 MV
15 MV
250 kVp
18 MeVe
default
water
only
Catphan
*
default
water
only
Catphan*
default
water
only
Catphan*
default
water
only
Catphan*
μ (%)
0.61
0.30
0.77
0.59
0.25
0.73
0.64
0.75
0.66
0.58
0.17
0.79
σ (%)
0.26
0.36
0.32
0.30
0.33
0.33
0.95
2.0
0.91
0.72
0.92
0.99
Min (%)
-8.5
-7.9
-9.0
-8.2
-6.0
-9.6
-45
-5.2
-45
-12
-10
-12
Max (%)
+10
+5.9
+10
+8.3
+6.6
+10
+49
+78
+47
+9.7
+6.0
+34
• artifacts (dental fillings, etc)
• couch issues (CT vs. treatment)
• bolus or shielding editing onto CT data
Not assigning media (other than water) is sometimes preferable over mis-assigning media!
-> be critical about your CT calibration!
-> be careful with how a manufacturer handles (HU - interaction coefficient)
AAPM Summer School 2006
Courtesy: Frank Verhaegen
19
AAPM Summer School 2006
Dose specification
The debate…
• Assuming proper material specification MC
dose calculations will provide absorbed
dose specified to tissue, Dtissue
In TP, do we really want the dose specified in
terms of Dtissue?
AAPM Summer School 2006
20
21
• Dtissue is the real
quantity needed!
• Reply: Yes, but life is not
perfect!
• Reply: Former algo’s
are not accurate so
they represent
anything but Dw
• Converting back to Dw
is adding complexity
• Dw is what is being used in
former algo’s.
– material specification is far
from perfect
– CT voxels are too large
– soft tissues within bony
matrix
• Reply: The incremental
complexity is not really
worthwhile talking about.
AAPM Summer School 2006
22
Validation
Primary beam
Beam modeling
Siebers et al 2000
PMB 45, 983
Absorbed dose to
water calculation
Patient dose
calculation
AAPM Summer School 2006
23
AAPM Summer School 2006
Beam modifiers
Reference and
relative “output” in
defined phantoms
Relative “output” in
patient geometries
with defined materials
24
Internal consistency
All tissues to water, plan 2
120
100
Left lung
80
60
CADplan
xVMC
40
20
0
Monte Carlo – all tissues to water
CADplan
All tissues to water, plan 2
120
100
80
cord
The conventional planning
system uses a pencil beam
algorithm once
MLC is used as a
beam shaping device
60
40
20
CADplan
xVMC
0
0
500
1000
Dose (cGy)
1500
0
500
1000
1500
Dose (cGy)
2000
2500
AAPM Summer School 2006, Windsor, Canada
Monte Carlo-based Clinical
Treatment Planning: Issues
for Consideration
Indrin J. Chetty and Jan Seuntjens
University of Michigan, Ann Arbor MI
and McGill University, Montreal QC
AAPM TG 105 report
Many of the findings and recommendations
presented here are reported in the AAPM Task
Group No. 105 report: “Issues
“Issues associated
with clinical implementation of Monte CarloCarlobased treatment planning” submitted to Med
Phys
IJ Chetty,
Chetty, B Curran, J Cygler,
Cygler, J DeMarco,
DeMarco, G
Ezzell,
Ezzell, B Faddegon,
Faddegon, I Kawrakow,
Kawrakow, P Keall,
Keall, H
Liu, CC-M Ma, DWO Rogers, D SheikhSheikh-Bagheri,
Bagheri,
J Seuntjens,
Seuntjens, JV Siebers
How should verification be performed for MC ?
Outline
Experimental verification of MCMC-based dose
algorithms: beam modifiers (MLC) and
transport within the phantom/patient
B.
Clinical treatment planning examples: MC vs.
EPL and CS
C.
Statistical Uncertainties in MCMC-based
treatment planning
D.
Conclusions
A. Experimental verification of MCMC-based
dose algorithms
A Monte Carlo-based treatment planning
system is just another treatment planning
system and as such should be subjected to the
same testing and verification requirements as
any other planning system (such as provided in
AAPM TG-53 – Fraass et al.)
MC is known to be more accurate under nonCPE conditions, and should be verified under
such situations
MC modeling of beam modifying devices
The BEAMnrc code system (NRCC) was
developed for transport through detailed MLC
geometries: component module developed by
Heath and Seuntjens (PMB 48: 4045404564(2003)) for the Varian Millenium 120120-leaf
MLC
Target
Collimator
treatment
head
A.
Vacuum Win
Flattening Filter
Ion Chamber
Jaws
MLC
patient
Courtesy of J Siebers, P Keall, MCV
The BEAMnrc code system is not fast enough
for routine clinical treatment planning
(patient(patient-dependent calculations on the order
of minutes)
Department of Radiation Oncology • University of Michigan Health Systems
1
MC modeling of beam modifying devices
Explicit MLC transport: BEAMnrc module
To improve the efficiency of simulation through
beam modifiers, investigators have developed
other approaches:
(a) Modified transport methods: first Compton
scatter or multiple order scatter, ignoring
Compton electrons
(b) Multiple source models parameterized for
simulating square shapes defined by the jaws;
requires explicit or empiricallyempirically-based model for
transport through the MLC
Explicit “approximate” MLC transport
Heath and Seuntjens (PMB 48: 40454045-64, 2003)
First Compton transport approximation
Tongue-and
groove effect
maximized:
Explicit
approximate
MLC transport
In PEREGRINE
Hartmann Siantar et. al. Med. Phys. 28 (2001)
120.0
1
2
3
4
5
6
7
8
9
10
11
12
13
0
1
Siebers et al. (PMB 47:322547:3225-49, 2002)
Sensitivity of closed leaf profile
to the rounded leaf tip (from Tyagi et al)
2
6
5
5
4
4
3
3
2
2
1
1
0
0
0
Multiple order Compton scatter approx.
Delivered with
even/odd leaves
closed half the
time, resp.
rel. dose
R=1cm
R=2cm
9
9
8
8
7
7
6
100.0
10
R=3cm
11
10
MLC leaf tip
12
11
13
12
13
80.0
60.0
40.0
R=4cm
Film
5cm
Film
DPM
8cm
DPM
6 mm
20.0
0.0
-2
-1
0
1
2
3
X (cm)
Tyagi et al. (Submitted to Medical Physics)
x (cm)
Department of Radiation Oncology • University of Michigan Health Systems
2
The importance of accurate vendor information !
15 MV, 10x10
profile in water
at 10 cm
1
FF w/ Cu
design
0.5
0.25
0
-10.0
-7.5
-5.0
-2.5
Relative Dose
0.75
FF w/ Tungsten
Patient Transport
Experimental verification in phantom
geometries
Solid Line =
Measurement
Distance (cm)
0.0
2.5
5.0
7.5
10.0
Experimental verification in heterogeneous
geometries
water
Verification – high density inhomogeneities
water
water
water
Arnfield et al. (MCV), Med. Phys, 27 (6) 2000
Ma et. al. Med. Phys. 26 (1999)
Verification – Other geometries
MCMC-based beam modifier transport: Summary
• As we move MC closer to the clinic, we need
more clinically realistic patient geometries
ρ=0.31
ρ=1.0
tumor
ρ=0.31
ρ=0.31
S
8 cm
Solid water
8 cm
Solid water
30 cm
30 cm
Rice et al, IJROP, 15, (1988)
Wang et al Med. Phys., 26 (1999)
In summary, regardless of the methods used
in modeling beam modifiers, appropriate
experimental verification is necessary
Experimental testing should include complex
configurations designed to verify the
improved accuracy expected with the use of
the Monte Carlo method
• Accurate measurements are difficult !!!
Department of Radiation Oncology • University of Michigan Health Systems
3
Treatment Planning: The main dosimetric issue
B. Monte CarloCarlo-based treatment planning
Solid = MC , 100%
Dashed = EPL, 100%
Blue = PTV
6 MV oblique fields
Underdosage of the PTV
DVH for the PTV
Percent Volume
Comparison of MC vs. simple (EPL) and
sophisticated heterogeneity correction
methods (CS) in lung, head and neck
cancer planning, and other treatment
sites
MC
EPL
Percent Dose
15 MV conformal lung plan
15 MV conformal lung plan
Note the differences in the spatial dose and
dose gradient due to penumbral broadening
EPL
6 MV AP/PA lung plan: 95% IDL coverage
% Volume
MC
% Dose
MC
CS
EPL
DVH - left lung
MC
% Volume
EPL
EPL
DVH - CTV
MC
EPL
% Dose
19 field (parotid sparing) head/neck forward plan
MC
CS
EPL
Department of Radiation Oncology • University of Michigan Health Systems
4
19 field (parotid sparing) head/neck forward plan
Prostate (4 fields + 3 segments)
MC
EPL
DVH – PTV
MC
EPL
% Volume
Convolution/Superposition
% Dose
Prostate DVH’s
100
DVH’s for the PTV and
normal liver
Difference map (MC-EPL)
sagittal view
100
PTV
80
PTV
80
60
% Volume
% Volume
Liver
MC
EPL
60
rectum
40
< 2%
40
MC
20
20
% Dose
0
20
EPL
0
0
40
60
80
0
Blue/red colorwash < 5%
100
normal liver
% Dose
25
50
75
100
beam penumbra effects
Conventional (1 cm collimation)
MC (1 cm) collimation
Comparison of “model“model-based” algorithms
The agreement between the MC and other modelbased methods (e.g. CS) will strongly depend on
the particular implementation of the algorithm
120
CS (1)
dose
Radiosurgery
CS (2)
MC
100
Ion Chamber
80
60
40
Solberg et. al. Radiother. Oncol. 49 (1998)
Electron disequilibrium always exists whenever the field size
is smaller than the range of the secondary electrons
ρ=0.25
20
ρ=1.0
depth (cm)
0
0
4
8
12
16
20
24
28
32
CS (1): 2 rep.
component
energies
CS (2): 1 rep.
energy averaged
over spectrum
Both CS (1) and
(2) agree in
water-based
tests
Department of Radiation Oncology • University of Michigan Health Systems
5
Summary
It’s not just the technology
The differences found in comparing the MC method
with other algorithms will be highly dependent on the
beam arrangements, field sizes, beam energies, and
tumor size and location
Situations where there is a lack of CPE (small field
sizes, high energies, low density tissues) may pose a
challenge even for CS algorithms because these
algorithms do not explicitly transport electrons
It’s how you drive it !
C. Statistical uncertainties in
MC-based treatment planning
“The only certainty is that nothing is
certain”
Pliny the Elder
Roman scholar & scientist (23 AD - 79 AD)
Implementation is critical !
Courtesy: Solberg (UNMC)
Statistical uncertainties
Probability
“Jittery” isodose lines due to the stochastic nature
of the MC method are quite different from dose
distributions computed with conventional
(deterministic) algorithms
σ ~ 1/√N,
8.0
N= total no. of
0.05 Gy
particles simulated
6.0
σ
4.0
2.0
μ
0.0
0.8
0.9
Dose (Gy)
1.0
1.1
1.2
In tx planning,
Relative uncertainty
= σ/ μ ~ 1/√dose
Statistical uncertainties
• Two sources of uncertainty: treatment head
simulation (latent uncertainty – term coined by
Sempau) and the patient simulation
• Latent uncertainty in the phase space is systematic
in nature, while the uncertainty in the phantom dose
is random
• The statistical uncertainty in calculated dose will
approach (as a function of 1/√N, where N is the
number of simulated particles), the finite, latent
uncertainty associated with the phase space,
regardless of the number of times the phase space
is sampled
Department of Radiation Oncology • University of Michigan Health Systems
6
Statistical uncertainties: previous work
• Several investigators have published on
statistical uncertainties in Monte Carlo dose
calculation (see the chapter for references)
3F lung plan (RT_DPM): relative uncert.
uncert.
rel. uncert. = (1σ/μ)x100 %
10 million particles
AP
55%
9%
25%
• Clinical planning studies show that a statistical
uncertainty of 2% or less does not significantly
affect isodose lines, DVHs, or biological indices
L LAT
7%
20%
GTV
PA
Clinical plan: one sigma % uncertainty
Clinical plan: one sigma % uncertainty
1.5 billion particles
150 million particles
20%
15%
5%
5.5%
9%
3.5%
0.5%
4%
1.8%
1.5%
0.5%
rel. uncert. = (1σ/μ)x100 %
rel. uncert. = (1σ/μ)x100 %
Effect of uncertainties on the 95% IDL
Effect of uncertainties on DVHs
150
million
10
50million
million
1.5
billion
100
10 milion
150 million
1.5 billion
Volume (%)
80
60
40
20
Dose (%)
0
90
95
100
105
DVH for a
plan w/given
uncertainty is
derived by
convolving the
DVH for the
“0%”uncert.
plan with the
given random
uncertainty
distribution
110
Department of Radiation Oncology • University of Michigan Health Systems
7
Uncertainty volume histograms (UVHs
(UVHs))
Volume (%)
100.0
Direct UVHs for the CTV
cumulative UVH
% vol.
50.0
150E6
80
500E6
60
1500E9
cumulative DVH
75
50
25
0
0
20
0
0.0
1.0
2.0
3.0
4.0
5.0
6.0
7.0
8.0
Relative uncertainty (%)
Effect of uncert.:
uncert.: parallel organs (lung)
No. of % σ/μ
histories (min.)
% σ/μ
(max.)
% σ/μ
(mean
dose)
MLD
(Gy)
%
NTCP
66 Gy
10E6
6.5
100.0
46.0
6.5
0.51
150E6
1.7
41.0
18.9
6.8
0.55
500E6
0.9
23.8
10.5
6.8
0.55
1500E9
0.5
17.9
6.1
6.8
0.55
Statistical uncertainties: 4D planning
Exhale
Intermediate
states
Inhale
0
20
40
DE
80
100
Relative uncertainty (%)
60
IJ Chetty et al. “Reporting and
analyzing statistical uncertainties in
Monte Carlo–based treatment
planning” in press Red Journal
Effect of uncert.:
uncert.: serial organs (cord)
No. of
histories
% σ/μ
(min.)
% σ/μ
(max.)
% σ/μ
(mean
dose)
10E6
12.6
97.3
42.2
150E6
3.3
28.0
12.9
3.7
500E6
1.8
15.2
7.1
2.0
1500E6
1.0
8.8
4.2
1.2
% σ/μ
(D>0.5
Dmax)
14.3
Statistical uncertainties: 4D planning
Inhale (mapped onto exhale CT)
Exhale
5.5%
4%
1.8%
Dint
60
40
dose (cGy)
20
0.0
See: Rosu et al.
Med Phys
32:248732:2487-95
(2005)
100
40
25.0
Deformable
organ
dosimetry
% vol.
10E6
150E6
500E6
1500E9
100
10E6
75.0
UVH’s/DVH’s for normal lung tissue
0.5%
6%
0.6%
0.5%
2%
0.5%
DI
Cumulative (exh+inh)
3.5%
Dosevoxel = wE DE + wintDint + wI DI
W = weighting factors derived from a breathing function: Lujan et
al. Med Phys (Med. Phys. 26, 715-20 (1999))
1.4%
0.45%
0.4%
Ref: IJ Chetty et al.
“Reporting and analyzing
statistical uncertainties in
Monte Carlo–based treatment
planning” in press Red Journal
Department of Radiation Oncology • University of Michigan Health Systems
8
σ D >0.5 D
max
Statistical uncertainties – Summary
• AAPM TG-105 report recommendations:
Report uncertainties to volumes rather than
individual points, i.e. σD>0.5Dmax, σPTV, or σPRV
σ D >0.5 D
• Point doses (ex. Dmax points) are subject to
large statistical fluctuation (See Kawrakow and
Fippel, PMB: 45, 2163-83(2000))
max
• Point (individual voxel) dose uncertainties may
be a concern for serial organs and requires
further investigation
• Prescribing doses to a line may obviate the
statistical issues with point doses
D. Conclusion
• A properly commissioned MC-based dose algorithm
will improve dose calculation accuracy in 3D-CRT
and IMRT treatment planning and is likely to
improve dose-effect correlations
• Clinical implementation of MC-based systems must
be performed thoughtfully and physicists must
understand the differences between MC-based and
conventional dose algorithms
• Successful implementation of clinical MC algorithms
will require strong clinician support and an
understanding of the paradigm shift with MC
algorithms
Acknowledgements
Neelam Tyagi
Mihaela Rosu
Eduardo Acosta
Martha Coselmon
Jean Moran
Dale Litzenberg
Daniel McShan
Randall Ten Haken
Bruce Curran
Alex Bielajew
Feng Ming (Spring) Kong
Benedick Fraass
Grant Support: NIH P01-CA-59827 and R01 CA106770
AAPM-TG 105 co-authors
Department of Radiation Oncology • University of Michigan Health Systems
9
Outline: “issues for consideration”
•
•
Building blocks of a generic clinical MCTP system
Verification issues in general
•
•
•
•
CT calibration & artifacts
Dose specification
Absolute calibration, MU calculations
Verification issues in detail
–
Validation of relative output
–
Experimental verification of MCMC-based dose algorithms: beam
modifiers (MLC)
Experimental verification of transport within the phantom/patient
phantom/patient
–
•
Retrospective lung treatment planning
• Patients from a Phase I/II multicentre
clinical trial (stage IIIA + B NSCLC,
concomitant chemo)
• Treatment planning:
– plan I (PTV1 dose up to 40 Gy)
– plan II (PTV2 boost to 60 Gy)
– Typically 3-fields per plan, MLC shaped, wedged
(CADplan, no heterogeneity corrections)
MCMC-based treatment planning: comparisons of MC versus simple
(correction(correction-based) and modelmodel-based algorithms
Statistical Uncertainties in MCMC-based treatment planning
MCTP and lung (NSCLC)
Retrospective re-planning & outcome association
Summary
•
•
•
•
AAPM Summer School 2006
• Delivered dose simulated from delivered MU’s
AAPM Summer School 2006
1
Patient 3
2
Patient 3
Patient 2
Patient 2
Gy
Gy
Gy
Gy
Gy
PTV2
PTV2
Gy
Gy
EqTAR
Monte Carlo
Gy
EqTAR
Monte Carlo
EqTAR underestimates the volume
that receives a dose between 95%
and 100% of the dose
120
100
30
80
PTV
20
Patient 3 - PTV2
60
10
40
Eqtar
20
xVMC
0
0
20
40
60
80
-10
0
-20
0
2000
4000
Dose (cGy)
6000
8000
-30
-40
% of planned PTV dose
100
120
Mean PTV Dose
Maximum PTV Dose
Patient
EqTAR
MC
Diff
EqTAR
MC
Diff
1
62.0
60.8
-1.9%
67.9
64.8
-4.6%
2
60.8
59.9
-1.5%
63.6
62.8
-1.3%
3
61.9
60.3
-2.5%
67.0
63.7
-5.0%
80
4
60.1
58.6
-2.5%
62.9
62.8
-0.2%
60
5
60.7
60.2
-0.8%
65.5
64.7
-1.2%
6
61.3
60.8
-0.8%
66.1
64.9
-1.8%
7
61.5
59.6
-2.2%
63.5
60.9
-3.1%
8
63.1
59.3
-6.0%
67.9
63.7
-6.2%
9
60.1
57.1
-5.1%
63.3
62.1
-1.9%
10
58.9
57.2
-2.9%
63.0
61.2
-2.8%
11
60.4
61.6
2.0%
65.0
65.3
0.4%
12
60.9
59.1
-2.6%
64.2
63.1
Average
-2.2%
120
Patient 3 - co
100
EqTAR
xVMC
40
20
0
0
1000
2000
3000 4000
Dose (cG
5000
6000
-1.7%
-2.5%
EqTAR underestimates volume
receiving 10-20% of planned dose
120
Patient 3 contralateral lung
100
15
EqTAR
xVMC
80
contralateral lung
10
5
60
0
0
40
-5
20
-10
20
40
60
80
100
120
-15
0
0
2000
4000
6000
Dose (cGy)
120
% of planned dose
8000
EqTAR overestimates volume receiving
0-10% of planned dose
Ipsilateral lung
Patient 3, ipsilateral lung
100
15
80
EqTAR
xVMC
60
10
5
40
0
20
-5
0
20
40
60
80
-10
0
0
2000
4000
Dose (cGy)
6000
8000
-15
% of planned dose
100
120
100
80
EqTAR
xVMC
60
40
20
0
0
2000
4000
6000
8000
Heart
4
2
0
-2
0
20
40
60
80
100
120
100* ( V_M C- V_EqTAR) / V_t
(V_EqTAR - V_MC)/V_total *100
Patient 3, he
120
-4
-6
-8
-10
Dose (cG
% of planned dose
100*(MC – EqTAR)/EqTAR
Cord
Tumour
bearing lung
Contralateral
lung
heart
Patient 1
Patient 2
Patient 3
6.2%
6.3%
19.1%
-4.2%
-2.2%
-7.5%
20.3%
15.9%
6.1%
-4.5%
-6.1%
-7.7%
Patient 4
-1.6%
2.2%
8.7%
-2.2%
Patient 5
Patient 6
Patient 7
Patient 8
Patient 9
Patient 10
Patient 11
Patient 12
5.3%
6.6%
2.6%
0.0%
1.7%
2.2%
7.4%
7.3%
1.9%
9.9%
0.3%
3.6%
4.8%
6.1%
4.6%
-1.0%
15.7%
14.4%
2.8%
6.7%
11.1%
18.4%
8.2%
13.5%
35.8%
41.4%
17.3%
32.6%
4.6%
24.2%
5.8%
11.1%
1.5%
(4.8%)
11.8%
(5.4%)
e-scatter
12.7%
(17.3%)
Average
St.Dev.
5.3%
(5.3%)
Beam model
How can an improved dose calculation algorithm
be useful in relation to outcome?
¾ Revise known dose-response or dosecomplication relations
¾ Study associations of outcome with difference
maps
Beam model
Mean lung dose (normalized using 2 Gy/ fraction scheme)
Radiation Pneumonitis
25.000
A main complication for lung cancer
radiation therapy
Bio-model indicators for radiation
pneumonitis:
• Mean lung dose (MLD)
• Vdose (V20 or V30)
• NTCP
20.000
15.000
10.000
y = 1.3637x 0.9264
5.000
0.000
0.000
5.000
10.000
15.000
20.000
25.000
MLD (Gy, CadPlan)
25
20
need for large-scale
replanning calculations!
15
10
y = 1.1421x 0.9606
5
0
0
5
10
15
MLD (Gy, EqTAR)
20
25
Gy
How can an improved dose calculation
algorithm be useful in relation to outcome?
¾ Revise known dose-response or dosecomplication relations
¾ Study associations of outcome with difference
maps
Monte Carlo
Conventional
Correlating late complications with dose
Post-treatment complications (patient 1)
Automatic fibrosis segmentation using:
¾
Automatic lung volume segmentation on planning and
diagnostic CT images
¾
Calculated Pre/Post RT tissue density changes
corresponding to Pre RT lung volume
¾
Mean tissue density changes corresponding to
physician-identified radiographic fibrosis grades*:
Grade 1 fibrosis: from 0.123 to 0.279 g/cc
Grade 2 fibrosis: from 0.279 to 0.546 g/cc
Grade 3 fibrosis: from 0.546 to 0.799 g/cc
9
9
9
+5 to 6 Gy
I. Rosen, T. Fischer, J.A. Anatolak et al., Radiology, 221:614-622, 2001
Post-treatment CT (+ 1y)
AAPM Summer School 2006
Dose-fibrosis correlation
Fibrosis segmentation through image registration
Post1 (RT+20d)
Pre (RT-1d)
DoseDose-response curves for the RTRT-induced fibrosis in the
ipsilateral lung
Post1 - Pre
RT+82d
RT+20d
All grades fibrosis (ips)
100
CADPLAN
80
60
40
20
0
80
60
40
20
0
0
20
40
Dose [Gy]
60
80
Post3 - Pre
20
40
Dose [Gy]
60
80
60
40
20
0
0
RT+188d
40
Dose [Gy]
60
80
MC
80
100
CADPLAN
60
40
20
0
20
40
Dose [Gy]
20
Ipsilateral (RT) lung
CADPLAN
80
0
20
40
100
MC
Fibrosis probability [%]
Fibrosis probability [%]
Post3 (RT+243d)
MC
60
All grades fibrosis (ips)
100
CADPLAN
CADPLAN
80
0
0
All grades fibrosis (ips)
Grade 3 fibrosis
Grade 2 fibrosis
Grade 1 fibrosis
MC
Fibrosis probability [%]
Post2 - Pre
All grades fibrosis (ips)
100
MC
Fibrosis probability [%]
Fibrosis probability [%]
CADPLAN
RT+133d
All grades fibrosis (ips)
100
Post2 (RT+144d)
22
G. Stroian et al 2006
60
80
0
20
40
Dose [Gy]
RT+237d
60
80
Percent volume [%]
Dose difference map - planning CT
MC
80
60
40
20
0
0
AAPM Summer School 2006
G. Stroian et al 2006
23
AAPM Summer School 2006
5 10 15 20 25 30 35 40 45 50 55 60 65
Dose [Gy]
24
Dose-fibrosis correlation (cont’d)
Conclusions
DoseDose-response curves for the RTRT-induced fibrosis in the
contralateral lung
80
60
40
• beam model accuracy
• heterogeneities
60
40
• MC planning presents specific clinical issues in addition to
the issues one is already familiar with.
• The new dose information can be used retrospectively in
two ways:
20
20
0
0
20
40
Dose [Gy]
60
0
80
5 10 15 20 25 30 35 40 45 50 55 60 65
Dose [Gy]
Contralateral (LT) lung
All grades fibrosis (contr)
20
CADPLAN
CADPLAN
– revise dose - response relationships
– to correlate to complications
MC
100
MC
15
P e r c e nt v olum e [% ]
Fibrosis probability [%]
MC
80
Percent volume [%]
Fibrosis probability [%]
CADPLAN
MC
0
Patient #2;
RT+362d
– The devil is in the details
– Dosimetric differences are due to two components:
100
CADPLAN
Patient #4;
RT+82d
• MC planning = raising the bar
Contralateral (LT) lung
All grades fibrosis (contr)
100
10
5
80
60
40
20
0
0
0
10
20
30
Dose [Gy]
40
AAPM Summer School 2006
50
60
0
5
10 15 20 25 30 35 40 45 50 55 60 65
Dose [Gy]
G. Stroian et al 2006
25
AAPM Summer School 2006
26