1. Dosimetric improvements in a cohort of patients with simulated tracheal tumors receiving intensity
modulated radiotherapy (IMRT) accounting for optimization convergence errors (OCEs)
Todd J. Scarbrough, M.D.*
; Samuel R. Faught, M.D.*
; Charles R. Thomas, Jr., M.D.†*
OMHSMitchellMemorialCancerCenter,Dept.ofRadiationOncology,Owensboro,Kentucky,USA
†
OregonHealth&ScienceUniversity,Dept.ofRadiationMedicine,Portland,Oregon,USA
OMHS Mitchell Memorial Cancer Center
1020 Breckenridge Street, Owensboro, Kentucky 42303 USA
email: scarbrtj@gmail.com
phone: 1 270-231-6269
BACKGROUND
In many commercially available treatment planning systems
(TPSs), when planning IMRT, the fluence optimization process
uses a different algorithm than the actual dose calculation work-
space. In the Eclipse™ treatment planning system (v8.9.09.18617,
Varian Medical Systems, Palo Alto, CA, USA), for static field IMRT,
the fluence optimization algorithm used is dose volume opti-
mizer (DVO) v8.9.08. Of the calculation algorithms available in
Eclipse, there is evidence that the analytical anisotropic algo-
rithm (AAA v8.9.08) is significantly more accurate in predicting
delivered dose in most clinical situations, especially air/tissue in-
terface scenarios as encountered in lung or tracheal carcinoma
cases. However, the discrepancy in the DVO and AAA calcula-
tion algorithms induces optimization convergence errors (OCEs)
whichareespeciallynoticeableintheTPSintheseparticularclini-
calsituations:dosesbecomesignificantlyinhomogenouswithin
targets such that plans must be normalized to very low isodose
lines in turn inducing unwanted high target inhomogeneities
and/or hot spots in the plan. Zacarias and Mills (J Appl Clin Med
Phys, 10:3061, 2009) have devised a fluence summing method to
correct for OCEs in the Eclipse™ TPS by forcing the DVO to look
at calculated AAA results. We selected a non-random cohort of
10patientsfromourinstitutionforthistreatmentplanningstudy
and created, within the TPS, simulated tissue masses (using the
assign CT number function to voxels in the Eclipse™ TPS) within
the mid-trachea 3 cm in longitudinal dimension and having AP
and lateral dimensions equal to half the diameter of the trachea.
We then performed IMRT plans on these simulated masses with
or without OCE corrections, and recorded the results.
METHODS
ThemasseswerecreatedasaboveandconstitutedtheCTVwith-
in the simulated plans on 10 patients who had previously been
scannedandtreatedinthedepartmentforthoracicmalignancies
(andweredeceased);thesepatientspreviousscansservedasthe
planning phantoms for this study. A 5 mm isotropic margin was
added to the CTVs to create PTVs within the plans. The median
CTV volume was 4.5 ccs; PTV volume, 17.8 ccs. A nine-field equal-
lyspacedcoplanarstaticfieldbeamarrangementwasemployed
using dMLCs, 600MU/min dose rate. The PTV was optimized to
receive 100% of the Rx dose (60 Gy/30 fx). The maximum plan in-
homogeneity at optimization was set to be 62 Gy. A smoothing
factor of 20/20 X/Y was used for optimization. No other optimi-
zationparameterswereutilized.Eclipse™ DVOv8.9.08wasused
foroptimization.Afterinitialoptimization,planswerecalculated
using Eclipse™ AAA v8.9.08 (Plan 1). All plans were normalized
such that 95% of the PTV received 100% of the Rx dose or more.
PTVminimum,medianandmaximumdoseswererecorded,and
PTV conformity indices (CIs) calculated. Next, each Plan 1 was
re-optimized and re-calculated (again using AAA) to correct for
OCEs using the method and tools of Zacarias and Mills (Plan 2).
A smoothing factor of 0/0 X/Y was used for OCE-corrected opti-
mization.PTVminimum,medianandmaximumdoses,andPTV
CIs, were recorded for each Plan 2. Plan metrics were compared
using the Mann-Whitney test.
RESULTS
Plan 1 non-OCE and Plan 2 OCE-corrected plans differed signifi-
cantly. PTV minimum dose differed for Plan 1 vs 2 (median 56.5
Gy vs 58.3 Gy, p=0.0004). PTV median dose differed for Plan 1 vs
2 (median 62.6 Gy vs 60.7 Gy, p=0.0002). PTV maximum dose dif-
feredforPlan1vs2(67.4Gyvs62.2Gy,p=0.0002).TheCIdiffered
for Plan 1 vs 2 (median 1.77 vs 1.67, p=0.0002). Plan 1 calculations
revealedsignificantlymoretargetinhomogeneity(median112%)
than Plan 2 calculations (median 104%), normalizing 95% of the
PTV volumes to receive 60 Gy Rx dose or more.
CONCLUSIONS
Correcting for OCEs in this simulated group of patients with in-
traluminal tracheal tumors resulted in significantly more confor-
mal and homogenous plans.
contour as usual:
normal structures,
CTVs, PTVs, etc.
initial optimization:
PLAN_base
Re-optimize using
PLAN_base as a base dose
plan:
PLAN_dosecorr
sum plans:
1)convertfluencesto.txtfiles
dcm2Ascii.exe(onC:drive)
2)sumfluencesinMicrosoftExcel
http://tinyurl.com/imrt-oce
3)importnewsummedfluencesinto
eachnewfield
PLAN_final
Normalize as usual; if final
plan is unacceptable,
repeat re-optimization
contour as usual:
normal structures,
CTVs, PTVs, etc.
initial optimization:
PLAN_base
Re-optimize using
PLAN_base as a base dose
plan:
PLAN_dosecorr
sum plans:
1)convertfluencesto.txtfiles
dcm2Ascii.exe(onC:drive)
2)sumfluencesinMicrosoftExcel
http://tinyurl.com/imrt-oce
3)importnewsummedfluencesinto
eachnewfield
PLAN_final
Normalize as usual; if final
plan is unacceptable,
repeat re-optimization
FIGURE 3. Correcting for OCEs. An initial plan is optimized & calculated in Eclipse. Next, a second plan using the initial plan as a
base dose plan is optimized and calculated. The two plans’ fluences are summed beam-by-beam using the simple software pro-
grams outlined above. The summed fluences are imported back into Eclipse, replacing the initial fluences from the initial plans.
This plan is calculated, and then normalized per the planner’s specifications. The resultant plan usually results in significantly bet-
ter homogeneity and overall PTV coverage.
FIGURE 2. Nine-field, coplanar, equispaced beam arrangement as used for all plans.
FIGURE 1. Simulated tracheal tumor (the clinical target volume, CTV) created to occupy ½ the tracheal
lumen, and an inferior-superior extent of 3 cm. The CTV was assigned zero Hounsfield units. A margin
of 5 mm was added to create the planning target volume (PTV), which was the optimization target.
FIGURE 4. Optimization parameters for “PLAN_base” as outlined in the METHODS section.
FIGURE 5. The fluence patterns for each plan, as created by the optimization algorithm,
are converted to numerical text format (and individual “.txt” files) by the dcm2ascii.exe
program. The .txt files are then loaded into a custom-designed Microsoft Excel macro
(http://tinyurl.com/imrt-oce); this yields new summed fluences which are then loaded
back into the Eclipse treatment planning software. For most situations, only a single
dose-correction plan is necessary to yield desired calculated plan results.
Plan inhomogeneity max: 62 Gy
PTV prescription: 60 Gy
ForPLAN_dosecorr,PLAN_baseis
selected as a base dose plan, and
0/0 smoothing is set, as well.
PLAN_base PLAN_final
The majority of the PTV is cov-
ered by ≥105% of the pre-
scribed dose due to opti-
mization convergence
errors generated by
the DVO algorithm.
The PTV has a maxi-
mum dose inhomo-
geneity of 11.7%.
This could be clini-
cally significant
depending on
PTV overlap into
critical structures.
For PLAN_final, only a small pro-
portion of the PTV volume is re-
ceiving ≥103.5% of the dose and
the maximum inhomogeneity in
the PTV is only 4.7%. Furthermore,
less monitor units will be used in
PLAN_final vs. PLAN_base, which re-
sults in slightly faster treatment times and
slightly less scatter dose to the patient.
FIGURE 6. Correcting for OCEs generates plans which result in significantly less target inhomogeneity and significantly better overall dose conformity.
Air/tissue interfaces
(or high density/low
density interfaces)
are regions in the
body where radia-
tion dose absorption
changes rapidly. The
anisotropic analytic al-
gorithm (AAA) models
this reasonably well, and
is a dose calculation al-
gorithm available in the
Eclipse treatment planning
system. However, the optimi-
zation algorithm for static field
IMRT (dose volume optimizer,
DVO) does not model this dose
phenomenon as well as AAA. Yet
AAA uses the fluences created by
the DVO.
We can “correct”
the relative DVO
error by running
second plans
using AAA cal-
culation of the
first plan as a
basedoseplan.
Then, we sum
the fluences of
each beam from
each plan togeth-
er, and re-calculate
thenewplanusingAAA.
“problem”
areas