A study of IMRT planning parameters on planning efficiency, delivery

efficiency, and plan quality

Kathryn Mittauer
Department of Radiation Oncology, College of Medicine, University of Florida, Gainesville,
Florida 32603 and J. Crayton Pruitt Family Department of Biomedical Engineering,
University of Florida, Gainesville, Florida 32611

Bo Lu, Guanghua Yan, and Darren Kahler

Department of Radiation Oncology, College of Medicine, University of Florida, Gainesville, Florida 32603
Arun Gopal
Department of Radiation Oncology, New York-Presbyterian Hospital, Columbia University, New York,
New York 10032

Robert Amdur and Chihray Liua)

Department of Radiation Oncology, College of Medicine, University of Florida, Gainesville, Florida 32603
(Received 9 January 2013; revised 11 April 2013; accepted for publication 11 April 2013;
published 13 May 2013)
Purpose: To improve planning and delivery efficiency of head and neck IMRT without compromising
planning quality through the evaluation of inverse planning parameters.
Methods: Eleven head and neck patients with pre-existing IMRT treatment plans were selected for
this retrospective study. The Pinnacle treatment planning system (TPS) was used to compute new
treatment plans for each patient by varying the individual or the combined parameters of dose/fluence
grid resolution, minimum MU per segment, and minimum segment area. Forty-five plans per patient
were generated with the following variations: 4 dose/fluence grid resolution plans, 12 minimum seg-
ment area plans, 9 minimum MU plans, and 20 combined minimum segment area/minimum MU
plans. Each plan was evaluated and compared to others based on dose volume histograms (DVHs)
(i.e., plan quality), planning time, and delivery time. To evaluate delivery efficiency, a model was
developed that estimated the delivery time of a treatment plan, and validated through measurements
on an Elekta Synergy linear accelerator.
Results: The uncertainty (i.e., variation) of the dose-volume index due to dose calculation grid vari-
ation was as high as 8.2% (5.5 Gy in absolute dose) for planning target volumes (PTVs) and 13.3%
(2.1 Gy in absolute dose) for planning at risk volumes (PRVs). Comparison results of dose distribu-
tions indicated that smaller volumes were more susceptible to uncertainties. The grid resolution of
a 4 mm dose grid with a 2 mm fluence grid was recommended, since it can reduce the final dose
calculation time by 63% compared to the accepted standard (2 mm dose grid with a 2 mm fluence
grid resolution) while maintaining a similar level of dose-volume index variation. Threshold values
that maintained adequate plan quality (DVH results of the PTVs and PRVs remained satisfied for
their dose objectives) were 5 cm2 for minimum segment area and 5 MU for minimum MU. As the
minimum MU parameter was increased, the number of segments and delivery time were decreased.
Increasing the minimum segment area parameter decreased the plan MU, but had less of an effect on
the number of segments and delivery time. Our delivery time model predicted delivery time to within
1.8%.
Conclusions: Increasing the dose grid while maintaining a small fluence grid allows for improved
planning efficiency without compromising plan quality. Delivery efficiency can be improved by in-
creasing the minimum MU, but not the minimum segment area. However, increasing the respective
minimum MU and/or the minimum segment area to any value greater than 5 MU and 5 cm2 is not rec-
ommended because it degrades plan quality. © 2013 American Association of Physicists in Medicine.
[http://dx.doi.org/10.1118/1.4803460]

Key words: IMRT, plan quality, delivery time, treatment planning

I. INTRODUCTION popularity of IMRT for radiation therapy (RT),1 there is a

Intensity modulated radiation therapy (IMRT) planning
is a complex and time-consuming task involving many
parameters that impact plan quality, treatment planning
time, and delivery time. Consequently, with the growing
greater need for improved planning efficiency. Another need
in the regimen of IMRT is to improve delivery efficiency
while maintaining the plan quality, since a long treatment
time can increase the chance of dose errors due to patient
movement.2, 3

061704-1 Med. Phys. 40 (6), June 2013 0094-2405/2013/40(6)/061704/13/$30.00 © 2013 Am. Assoc. Phys. Med. 061704-1
061704-2 Mittauer et al.: Parametric analysis of IMRT planning
Less “complex”4 IMRT plans allow for improved deliv-
ery accuracy and reduced delivery time.4 Several measures
on how to reduce the plan complexity have been proposed.4–15
These efforts focused on limiting the intensity fluctuations by
setting maximum intensity limits8, 9 or applying a smoothing
filter.4–7, 9, 10 Filtration or maximum intensity limits are ap-
plied either inside the optimization loop (i.e., concurrently
or postiteratively) or outside the optimization loop (i.e., prior
to leaf sequencing). By limiting the intensity spikes, the plan
complexity is reduced through the impact on the number of
MUs delivered per segment, the total number of segments,
and/or the segment width.
However, there are inverse planning parameters that im-
pact the segmentation directly during IMRT optimization:
(1) the minimum segment area parameter (MSAP) limits the
MLC area to greater than a minimum size for each seg-
ment, (2) the minimum MU parameter (MMUP) influences
the plan’s segments by limiting an individual segment’s mini-
mum MU, and (3) the maximum number of segments controls
the total segment number for the plan. Since these parame-
ters are incorporated directly into the optimization loop, plan
quality is not degraded with postprocessing methods, such as
filtration or smoothing.9, 16 Although strategies15 on limiting
the maximum number of segments have been studied using
the Pinnacle treatment planning system (TPS) (Philips Radi-
ation Oncology Systems, Andover, MA), to our knowledge
initial settings for MSAP and MMUP in light of delivery effi-
ciency have not been discussed.
Niemierko and Goitein first reported that a smaller grid
resolution provides superior dosimetric accuracy due to a
smaller interpolation of dose between calculation points.17
Investigators17–19 through both empirical methods18, 19 and
analytical models17, 18 reached a common consensus that a
dose grid resolution of 2 mm is sufficient to minimize dose
errors from a finite grid resolution. Chung et al. reported the
impact on plan quality for various dose grid resolutions us-
ing the Pinnacle TPS.19 However, this study did not report
the cost of planning time nor did it report plan quality dif-
ferences for critical structures. Although dose grid resolution
has been well studied, the impact of fluence grid resolution
has not been investigated at this time. Note the Pinnacle TPS
allows for the user to specify these grid resolutions separately
for dose and fluence.
The purpose of this study is to recommend a set of in-
verse planning parameters for head and neck IMRT planning
that provides a relatively low planning effort without com-
promising plan quality. Towards this goal, this study evalu-
ates a range of fundamental inverse planning parameters and
their effect on plan quality, dose calculation time, and deliv-
ery time. Such a study is useful for setting an initial set of
parameters that allow for fast planning and efficient treatment
delivery, while allowing the user to be confident that planning
calculations are being performed accurately.
In this study parameters of the MSAP and the MMUP are
studied both independently (i.e., varied one at a time) and de-
pendently (i.e., varied simultaneously) in light of plan qual-
ity and delivery time differences. By relaxing the MSAP and
the MMUP, the plan complexity can be reduced; hence, the
Medical Physics, Vol. 40, No. 6, June 2013
061704-2
treatment delivery time can be improved. To assess delivery
time, we have proposed a delivery time model. Similar efforts
include Li and Xing’s delivery time model for a Varian (Var-
ian Oncology System, Palo Alto, CA) linear accelerator.20
However, no such model has been proposed at this time for
an Elekta (Elekta Oncology Systems, Norcross, GA) linear
accelerator. Finally, we evaluate the impact of the calculation
parameters of both dose grid and fluence grid resolution on
plan quality and calculation time. We seek alternative options
for both dose grid and fluence grid resolution based on plan-
ning time considerations. The parameters found in this study
will provide a class solution that can reduce planning time.
II. MATERIALS AND METHODS
II.A. Overview
Eleven head and neck patients were chosen for this
study. Plans were recomputed for each patient with varying
dose grid resolution (4 variations), minimum segment area
(12 variations), minimum MU (9 variations), and the mini-
mum segment area and minimum MU were varied together
(20 variations), while all other variables were held constant.
The patients and plan variations are described in more detail
in Secs. II.B–II.D. An overview of this study can be seen in
Table I. Plan quality, calculation time, and delivery time
served as the study’s endpoints.
II.B. Patients
Eleven head and neck patients previously treated with
IMRT were selected for this retrospective study. Head and
neck cases were chosen because of the planning complex-
ity associated with this site. As displayed in Fig. 1, all ex-
isting treatment plans were prescribed 70 Gy to the high risk
planning target volume (PTV 70) and 56 Gy to the standard
risk PTV (PTV 56) in 35 fractions. Since this fractionation
treated both PTVs concurrently, both PTVs were optimized
simultaneously under the same fractionation. A summary of
the eleven initial treatment plans is given in Table II to show
that a variety of cases were selected.
As shown in Table II, existing patient plans utilized six
or seven coplanar IMRT beams with varying gantry angles.
Based on investigations of the appropriate number of beams
to use for head and neck IMRT,21 6–7 beams was considered a
reasonable choice. Beam angles were chosen to best minimize
doses to planning at risk volumes (PRVs) while maximizing
PTV coverage. Beam energy was 6 MV for all cases.
Patients selected included eight whole neck IMRT cases
and three partial neck IMRT cases. Whole neck IMRT, illus-
trated in Fig. 1, was used when the gross disease extended
into the lower neck region. For these eight patient plans, the
planning isocenter was placed inside of PTV 70. For the three
partial neck IMRT cases, an AP lower anterior neck (LAN)
field was added separately to treat the nodal region to 56 Gy.
A technique described by Li et al. was used in which the
LAN field was feathered by shifting the match line 1 cm per
segment for a total of four segments (Fig. 2).22 On the last
061704-3 Mittauer et al.: Parametric analysis of IMRT planning 061704-3

TABLE I. Breakdown of trials for each case.

Grid resolution trials MSAP trials MMUP trials MMUSAP trials

Dose grid Fluence grid Segment Segment Segment

Trial (mm) (mm) Trial size (cm2 ) MU Trial size (cm2 ) MU Trial size (cm2 ) MU

1 2 2 5 3 3 17 4 2 26 4 3
2 3 3 6 4 3 18 4 3 27 4 5
3 4 4 7 5 3 19 4 4 28 4 7
4 4 2 8 6 3 20 4 5 29 4 10
9 7 3 21 4 6 30 6 3
10 8 3 22 4 7 31 6 5
11 9 3 23 4 10 32 6 7
12 10 3 24 4 20 33 6 10
13 16 3 25 4 35 34 8 3
14 25 3 35 8 5
15 36 3 36 8 7
16 49 3 37 8 10
38 10 3
39 10 5
40 10 7
41 10 10
42 16 3
43 16 5
44 16 7
45 16 10

Note: MSAP, min segment area parameter; MMUP, min monitor unit parameter; MMUSAP, min monitor unit and segment area parameters. (1) There are some trial
redundancies between parameters.

TABLE II. Summary of diagnosis and plan information for the 11 head and neck cases in this study.

Volume (cm3 )

Case no. Disease site No. of IMRT beams MU for 2 mm grid resolution plan PTV 70 PTV 56

1 Posterior pharyngeal wall 7 723 424.1 264.5

2 Oropharynx and nasopharynx 7 1037 420.8 950.1
3 Glottic larynx 7 560 430.7 471.7
4 R. glosso tonsiliar sulcus 7 657 464.4 436.8
5 L. true vocal cord 7 542 78.9 581
6 L. maxillary sinus 7 558 139.9 194.3
7 L. paratracheal and thyroid 6 691 99.3 358.7
8 R. oral tongue 7 + LAN 601 147.5 280.1
9 L. floor of mouth and mandible 6 755 116.9 225.5
10 R. oral tongue and floor of mouth 7 + LAN 505 167.9 251.8
11 L. base of tongue 7 + LAN 516 160.5 325.5
Statistic data Mean 649.5 241.0 394.5
Standard deviation 154.1 156.3 218.2

Note: PTV, planning target volume; LAN, lower anterior neck; PTV 70, high risk PTV; PTV 56, standard risk PTV. (1) PTV 70 volume is excluded from PTV 56 volume.

F IG . 1. Whole neck IMRT with isodose distribution for case no. 2 with display of high risk PTV 70 and standard risk PTV 56.

Medical Physics, Vol. 40, No. 6, June 2013

061704-4 Mittauer et al.: Parametric analysis of IMRT planning
F IG . 2. AP LAN field of case no. 8 treated to 56 Gy using match line feathering in 1 cm intervals with four IMRT segments. The rightmost digitally reconstructed
radiograph (DRR) is the last segment in which the cord is blocked.
segment of the LAN field, the cord was blocked to reduce its
dose to less than 50 Gy. The planning isocenter was placed at
the junction of the LAN field with the head and neck fields, at
a depth greater than 1 cm.
II.C. Treatment planning
New treatment plans were generated from each of the
eleven cases by varying parameters of dose and fluence grid,
minimum MU, and minimum segment area. Forty-five trials
per patient, 495 trials in total, were reoptimized and/or re-
computed on the Pinnacle TPS (version 9.0) using direct ma-
chine parameter optimization (DMPO) and adaptive super-
position/convolution for the dose calculation. The maximum
number of optimization iterations was set to 25, with intensity
optimization occurring within the first five iterations. Twenty-
five iterations was selected based on UF planning experiences
with DMPO as well as the Pinnacle’s historical recommen-
dations of using a TPS default preset of 25 iterations. After
calculation, all plans were scaled such that the dose to 95%
of PTV 70, D95%, received at least 70 Gy. Differential dose
volume histograms (DVHs) were exported for plan quality re-
sults.
All initial eleven cases were planned according to the pro-
cess and dose objectives found in Table III. The generic dose
objectives are derived from quantitative analysis of normal
tissue effect in the clinic (QUANTEC) data,23 Radiation Ther-
apy Oncology Group (RTOG) protocols, and UF historical
TABLE III. ROIs (regions of interest) with expansions and dose goals for all cases.
ROI Expansion (mm)
CTV (95% Vol > Rx)
CTV (99% Vol > 93% Rx)
CTV (20% Vol < 110% Rx)
Brainstem
Cochlea
Cord
Larynx
Parotid
Note: Contour expansion was performed after physician’s initial contour delineation.
Medical Physics, Vol. 40, No. 6, June 2013
061704-4
data and experiences. Optimization objectives for organs at
risk (OARs) were previously determined by organ classifica-
tion, serial or parallel. Objectives for serial structures were
specified as maximum dose and volumetric maximum DVH.
Parallel structure objectives were defined as volumetric max-
imum DVH. PTV objectives were minimum dose, uniform
dose, and volumetric minimum DVH. Note that the objective
functions during the initial planning were customized for each
patient based upon the tumor size and the anatomy, which can
be slightly different from patient to patient. All optimization
objectives and contours were kept the same as the original
treatment plan when generating the 495 trials.
All contour volumes were previously delineated by clin-
icians and were used on the initial patient treatment plan.
The gross tumor volume (GTV) and positive nodes were ex-
panded, excluding OARs, to delineate the high risk clinical
target volume (CTV), CTV 70. A 3 mm contour expansion
was applied to OARs and CTVs to delineate PRVs and PTVs
(Table III).
II.D. Parameter variations
The dose grid resolution, minimum MU, and minimum
segment area were varied while keeping the all other param-
eters constant. Note that planning parameters are TPS depen-
dent and apply to the Pinnacle TPS. Dose grid resolution was
selected to evaluate the tradeoff between dose uncertainties
caused by grid resolution variation and the cost of planning
PTV and PRV dose goals
3 95% Vol > 70 Gy (PTV70), 56 Gy (PTV56)
3 99% Vol > 65.1 Gy (PTV70), 46.5 Gy (PTV56)
3 20% Vol < 77 Gy (PTV70)
3 0.1 cc ≤ 55 Gy
3 0.1 cc ≤ 45 Gy
3 0.1 cc ≤ 50 Gy
3 Mean ≤ 36 Gy
3 Mean ≤ 26 Gy
061704-5 Mittauer et al.: Parametric analysis of IMRT planning
time. Minimum segment area and minimum MU were se-
lected to study the effect of modulation (i.e., intensity) on plan
quality and treatment delivery time.
Default parameters were the values the parameters take
when not intentionally varied. Table I specifies these default
parameters as boxed values, representing the reference plan
used for dose comparison in Sec. III. Default parameters were
2 mm (dose and fluence grid resolution), 4 cm2 (minimum
segment area), and 3 MU (minimum MU). We infer from pre-
vious work,17–19 a 2 mm dose and fluence grid resolution as
the established standard to measure dose accuracy. Default
parameters of 4 cm2 and 3 MU were chosen empirically to
decrease the likeliness of mechanical inaccuracies and output
fluctuations, which can arise for small MU and/or small field
size deliveries.24–27 A more detailed discussion of the impact
of these parameters on mechanical and dosimetric accuracy is
presented below.
II.D.1. Dose grid and fluence grid parameter
From this point forward, grid resolution variation will refer
to the resolution variation of both dose grid and fluence grid
resolution. If dose grid resolution and fluence grid resolution
are not equal, then the dose grid resolution and the fluence
grid resolution will be noted separately.
Four trials with final calculation grid resolutions of 2,
3, and 4 mm, in addition to a 4 mm dose grid with a 2
mm fluence grid resolution, were computed for each patient
(Table I). These four trials were chosen based on the TPS re-
quirement that the dose grid resolution be a multiple of the flu-
ence grid resolution. While a grid resolution of greater than 2
mm is considered adequate for IMRT planning,17–19 any uni-
form grid resolution of greater than 4 mm is considered inac-
curate in terms of dose calculation quality.18
Optimization grid resolution was kept constant as 4 mm
among the four trials. Calculation times for each trial were
recorded for determining planning efficiency. As previously
noted in Sec. II.C, after the final dose calculation all plans
were scaled by the appropriate number of monitor units to
meet D95% of the PTV 70.
II.D.2. MSAP
The MSAP controls the plan’s modulation by limiting the
MLC area to greater than a minimum size for each segment.
Twelve trials per patient of varying minimum segment area
from 3 to 49 cm2 were computed (Table I). To have a greater
evaluation of the parametric trend, we took the minimum pa-
rameter limit to one unit below our default value of 4 cm2 .
The maximum limit for MSAP was selected in consideration
of the mean high risk treatment volume (Table I). The mean
PTV 70 volume 241 cm3 (planar area 39 cm2 ) would be sat-
isfied as far as coverage by a maximum segmental area of
49 cm2 .
The default value for MSAP was selected based on the
mechanical limitations and dosimetric effects of an Elekta
MLC.24–26 The positional accuracy of the Elekta MLC has
been reported to be within 0.4 mm.25, 26 For example, a 1 mm
Medical Physics, Vol. 40, No. 6, June 2013
061704-5
positional error for a 1 cm2 segment would result in a 10% dif-
ference in the segment area. Furthermore, the dosimetric out-
put (i.e., dose/MU) for an Elekta linear accelerator (LINAC)
fluctuates due to head scatter and loss of electronic equilib-
rium for field sizes below 3 cm2 .27
II.D.3. MMUP
The MMUP influences the plan’s segments by limiting an
individual segment’s MU. Here, a segment cannot contribute
less than a specified MU. The MMUP was chosen to range
from 2 to 35 MU in nine trials (Table I). Similar to Sec. II.D.2,
the minimum limit was taken at one value below the default
value of 3 MU. The maximum limit was selected when the
number of segments per beam approached one.
The default MMUP was selected based on considerations
of an Elekta LINAC. Dosimetric accuracy can become com-
promised when delivering only a few MUs per segment, due
to the LINAC’s servo requiring a few MUs to stabilize the
beam. Sharpe et al. reported dosimetric output fluctuations of
the order of 2% for 1–3 MU deliveries on an Elekta LINAC.27
II.D.4. Minimum MU and segment area
parameters (MMUSAP)
Since both minimum MU and minimum segment area af-
fect the plan’s segments and modulation by different means,
the effect of these two parameters changing together was also
examined. A set of 20 trials for MMUSAP (i.e., MMUP +
MSAP), varying the minimum MU parameter from 3 to 10
MU and the segment area parameter from 4 to 16 cm2 , were
computed for each patient (Table I).
The MMUSAP parameter was included in this study to
validate that the MMUP and MSAP trends would be main-
tained when simultaneously changing both parameters. Thus,
MMUSAP trials were examined for their trend, and were not
taken to the maximum limits of 49 cm2 and 35 MU as in
Secs. II.D.2 and II.D.3.
II.E. Treatment delivery time model
II.E.1. Model derivation
Delivery time for step and shoot IMRT has been previously
described by Li and Xing, and serves as the basis of our de-
livery time model.20 Li and Xing20 previously reported an av-
erage deviation of 2.1% with a maximum difference of 3.9%
(16 s) between calculated and measured times on a Varian
TrueBeam. The Li and Xing20 model provides a valid for-
malism for LINACs with gridded guns and delivery methods
without a gantry position verification control point. However,
a more detailed model is needed to account for machine-
derived metrics. Also expanding the model to include fac-
tors for delivery with nongridded guns, unequally spaced
beam geometry, and MOSAIQ Radiation Oncology Informa-
tion System (Elekta Oncology Systems, Norcross, GA) auto-
field-sequence (AFS) mode would allow for a more universal
model.
061704-6 Mittauer et al.: Parametric analysis of IMRT planning 061704-6

TABLE IV. Inputs, parameters, and outputs for treatment delivery time model.

Input (plan specific) Parameter (machine specific) Output

Name Symbol Name Symbol Name Symbol

dt G
Gantry degree spacing xG Gantry speed dx Time Ttotal
No. of beams NB Gantry lag toG
dt MLC
MLC travel distance (cm) xMLC MLC speed dx
No. of segments NSeg MLC lag + intersegmental beam-on delay toMLC
MU per fraction xMU AFS-gantry delay TCPG
dt MU
Dose rate dx
Interbeam beam-on delay toMU

Notes: AFS, auto-field-sequence mode. (1) Inputs are treatment plan dependent and correspond to Eq. (4). (2) Parameters
are machine dependent and correspond to Eqs. (1)–(4). (3) Output is the total treatment time for any step and shoot IMRT
plan.

For the success of such a model, parameters unique to the results of Eqs. (1)–(3), with the exception of the following
treatment machine must first be defined. Second, inputs from terms, because they are plan dependent: xG (the gantry degree
the patient-specific treatment plan would need to be deter- spacing between first and last beam), xMLC (the sum of the
mined. Such model (Table IV) would then use corresponding maximum leaf travel distance of each segment), and xMU (the
plan-based inputs and machine-based parameters to output the MU per fraction).
total treatment delivery time. This model relies on the following assumptions: TCPG ap-
Three independent equations can be formulated to model plies only to MOSAIQ users with AFS mode on an Elekta
the machine-based parameters from the linear relationship of LINAC, and was determined to be 4.9 s. AFS mode uses an in-
parametric incremental change and time: terbeam control point to verify the gantry position once a new
dt G G gantry angle has been reached. The MLC is able to update
TG = x + toG , (1) as the gantry rotates to the next beam, thus the first segment
dx
for each beam is excluded from the MLC speed and interseg-
mental beam-on delay; instead, the first segment of each beam
dt MLC MLC uses interbeam beam-on delay.
T MLC = x + toMLC , (2)
dx

dt MU MU II.E.2. Parameter acquisition

T MU = x + toMU , (3)
dx To assess the treatment delivery for this study, an Elekta
where x represents the displacement (e.g., gantry degree spac- Synergy LINAC was used to determine the machine-specific
ing, MLC travel distance, or the number of delivered monitor parameters of Eqs. (1)–(3). The machine parameters along
units), t0 is the lag time associated with the ramp-up (e.g., ac- with the individual plan inputs were used in the delivery time
dt model [Eq. (4)] to calculate the delivery time for all 495 treat-
celeration and/or deceleration), and dx is the inverse of the
velocity. Specifically, toMLC
is the intersegmental beam-on de- ment plans.
lay and MLC lag, and toMU is the interbeam beam-on delay. To verify the accuracy of the delivery time model, a se-
With the inclusion of the above machine-derived param- lection of 33 plans were delivered with step and shoot deliv-
eters with the plan-specific input values, the total treatment ery using MOSAIQ AFS mode. Measured times were then
delivery time for any unique IMRT plan can be determined. compared to calculated times from our model and a modified
The machine-characterized, plan-specific delivery time model Li and Xing20 model. We modified the Li and Xing20 model
for step and shoot delivery becomes to represent Elekta delivery by including the following fac-
tors from our model: intersegmental beam-on delay, inter-
dt G G
T total ∼
= (x ) + toG (N B − 1) + T CPG (N B − 1) beam beam-on delay, and AFS gantry delay.
dx As previously noted in Sec. II.E.1, the user must first char-
dt MLC MLC   acterize their machine. Details of this parameter acquisition
+ (x ) + toMLC N Seg − N B
dx will be described here. Three simple tests were developed to
characterize the LINAC’s metrics in terms of gantry, MLC,
dt MU MU
+ (x ) + toMU (N B ), (4) and dose rate. Such novel method allows for the user to eas-
dx ily parameterize their machine for delivery time. Each test
where NB is the number of beams, NSeg is the number of seg- measured a parameter’s rate of change through intervals rang-
ments, and TCPG is the gantry verification control point time. ing from the smallest measurable amount of change to the
The remaining terms of Eq. (4) are taken directly from the largest practical amount of change. The data were plotted as

Medical Physics, Vol. 40, No. 6, June 2013

061704-7 Mittauer et al.: Parametric analysis of IMRT planning 061704-7

the amount of change, x, versus time, T, and a linear fit was trials, like the other redundant trials listed in Table I, have the
performed. The three tests are as follows: same plan quality results. Evaluated dose was taken from the
DVH, using the dose criteria in Table III: mean dose (larynx,
(1) Gantry speed was measured in service mode by man-
parotid), 0.1 cc dose (spinal cord, cochleae, and brainstem),
ually rotating the gantry through various incremen-
D20% and D99% (PTV 56, PTV 70), and D95% (PTV 56).
tal angles, e.g., 10o , 20o , 70o , 90o , 180o , 200o , 340o ,
PTV 70 D95% was excluded since all plans were scaled to be
and 360o while measuring the time. No difference was
equal at D95%. Percentage differences for PRVs and PTVs
found between counterclockwise and clockwise rota-
were calculated between the evaluated DVH dose of each trial
tion, but this could be an issue with an unbalanced
and the DVH dose from its respective reference plan. These
gantry.
percentage differences will be referred to as dose differences,
(2) MLC speed was determined by delivering an IMRT
which are the dose calculation differences in DVH results that
plan designed with symmetric leaf sequences and sin-
result from changing the specified parameter.
gle direction leaf sequences. The sequences were rep-
Note for MSAP, MMUP, and MMUSAP plans, statistical
etitions of transitions from small fields to large fields,
results were eliminated for PRVs where dose was negligi-
and vice versa. Thus, both MLC extension and retrac-
ble; specifically, if the dose was less than 50% of either the
tion was accounted for. Time was calculated from the
0.1 cc or mean dose objective from Table III. These PRVs
LINAC “beam-off” indicator at the completion of one
are not statistically relevant because of the ease of achieving
segment to the delivery of the first MU on the next seg-
optimization objectives.
ment. Primary jaws and backup jaws were also tested,
and for an Elekta LINAC their speed was negligible
(i.e., faster) compared to the MLC speed.
III.A. Dose and fluence grid parameter
Note for nongridded guns there is a beam-on delay due
to the dose rate and gun ramp-up period. This beam- Dose differences among the grid resolutions of 2, 4, and
on lag time is inherently included in the above testing the 4 mm dose grid with 2 mm fluence grid resolution are
condition. However, this beam-on lag is representative displayed on the cumulative DVH (Fig. 3) for case no. 9. The
of the IMRT intersegmental beam-on delay and not the results for case no. 9 were representative of the results for all
interbeam beam-on delay. The subtle difference is that cases. The 3 mm grid resolution was eliminated from Fig. 3 to
in the former intersegmental beam-on delay the gun re- better display the maximum and minimum DVH differences
mains in active mode and in the latter interbeam beam- for grid resolution.
on delay the gun current must ramp-up from standby For PTVs and PRVs, the 4 mm dose grid with 2 mm flu-
mode to active mode. The latter contributes to a greater ence grid resolution was found to be most comparable in
time delay and is measured in the following test. Note terms of DVH to the 2 mm grid resolution in Fig. 3. The
that this test does not decouple MLC lag from acceler- 4 mm grid resolution varied the most from the 2 mm grid
ation/deceleration and intersegmental beam-on delay. resolution (Fig. 3). Specifically, Fig. 3 shows PTV 70 for the
From our observations, MLC lag is much less of a con- 4 mm grid resolution lacking a steep gradient; this represents
tribution to time compared with the beam-on delay. a breakdown of dose coverage and loss of dose homogeneity.
(3) Dose rate and interbeam beam-on delay were deter- Note that the DVH results for the 3 mm grid resolution fell
mined by delivering static fields of varying MU from
1 to 20 MU. Static fields were used to measure the
delay at the first segment of each new beam as the
gun warmed up out of standby mode, thus character-
istic of interbeam beam-on delay. Time was measured
between the “beam-on” and the “beam-off” indicator.
For gridded gun machines instantaneous dose rate is
possible because the gun remains in active mode, thus
the interbeam beam-on delay would be negligible.

III. RESULTS AND DISCUSSION

The differences, correlations, and tradeoffs associated with
the IMRT planning parameters were evaluated in terms of de-
livery and planning efficiency and plan quality. We also derive
recommendations to the user.
Trial plans were compared to a specified reference plan,
as noted in Table I and Sec. II.D: using 2 mm grid resolution
F IG . 3. DVH curves of PRVs and PTVs for case no. 9 for the grid resolu-
and 3 MU with 4 cm2 for MSAP, MMUP, and MMUSAP. tion parameter. Notes: (1) medium solid curve represents 2 mm dose/fluence
Note that MSAP, MMUP, and MMUSAP reference plans had grid, thin solid curve represents 4 mm dose/fluence grid, and dashed curve
the same default IMRT parameters. Thus, these reference represents 4 mm dose grid with 2 mm fluence grid, (2) flu., fluence.

Medical Physics, Vol. 40, No. 6, June 2013

061704-8 Mittauer et al.: Parametric analysis of IMRT planning 061704-8

TABLE V. PRV and PTV dose differences for grid resolution parameter.

PRV Dmax Dmax D

Grid resolution (mm) Cochlea Serial Parallel All PRVs

3 13.3% 5.8% 0.9% 1.0%

4 7.9% 5.3% 1.5% 1.2%
4, 2 flu. 5.5% 3.5% 0.7% 0.7%

PTV Dmax Dmax D

Grid resolution (mm) PTV 70 D99% All PTVs All PTVs
3 2.8% 1.4% 0.5%
4 8.2% 3.3% 1.0%
4, 2 flu. 6.2% 1.9% 0.5%

Notes: flu., fluence; Dmax , max dose difference; Dmax , mean max dose difference; D, mean dose difference. (1) 3
and 4 mm denote matching dose/fluence grid. (2) 4, 2 flu. denotes 4 mm dose grid with 2 mm fluence grid. (3) Reference
taken at 2 mm dose/fluence grid. (4) Dmax was the cochlea and PTV 70 D99% among all cases.

between the DVH for the 4 mm dose grid with 2 mm fluence
grid resolution and the DVH for the 4 mm grid resolution.
Table V lists the maximum dose difference (Dmax ), the
mean dose difference (D), and the mean maximum dose
difference (Dmax ), for PRVs and PTVs. All dose differences
in Table V were measured as absolute percentage differences
from the 2 mm grid resolution reference plan. To best de-
termine the dose variability of the grid resolution, sign con-
vention was not used because we sought to measure the ab-
solute change of dose due to grid resolution. Note that each
dose objective for PTV 56 and PTV 70 was considered sepa-
rately when averaging in the last two columns as five criteria:
PTV 56 (D99%, D20%, D95%) and PTV 70 (D99%, D20%).
The PRV maximum dose difference, being the greatest
PRV percentage difference for all cases of that grid resolu-
tion, was the cochlea at 13.3% or 2.1 Gy (3 mm grid reso-
lution); 7.9% or 1.8 Gy (4 mm grid resolution); and 5.5% or
1.2 Gy (4 mm dose grid with 2 mm fluence grid resolution)
(Table V). This result was due to the cochlea having a 0.1 cc
volumetric objective where dose is determined by the tail-end
of the DVH. As expected, small volumes with small volumet-
ric objectives were more susceptible to dose inaccuracies due
to grid resolution.
For PRVs, the mean maximum dose difference of Table V
is the mean across each maximum PRV percentage difference.
PRV mean maximum dose difference was calculated by deter-
mining the greatest deviation for each PRV, and then averag-
ing the results among serial and parallel PRVs. Table V shows
higher dose differences for serial than for parallel PRVs. Since
the serial structures were defined by 0.1 cc dose objectives and
parallel structures were typically defined by mean dose ob-
jectives, the same conclusion as above applies here that small
volumetric objectives have greater dose variability due to grid
resolution.
The greatest dose deviation for the PTV maximum dose
difference in Table V was 8.2% or 5.5 Gy, for PTV 70 D99%,
4 mm grid resolution. Thus, overall, the dose variability due
to grid resolution was found to be as high as 8.2% for PTVs
and 13.3% for PRVs.
One anomaly in Table V was that the 3 mm grid resolu-
tion PTV maximum dose difference and PTV mean maximum
dose difference were more comparable to the 2 mm grid res-
olution than the 4 mm dose grid with 2 mm fluence grid res-
olution. This result was only seen for PTV 70. The anomaly
is attributed to the plans being scaled to PTV 70 D95% using
the nearest whole MU. Thus, whole number approximation
led to some plans being scaled higher than exactly D95%, re-
sulting in D99% and D20% being somewhat inconsistent for
PTV 70.
The PRV and PTV mean dose differences remained under
1.2% for all trials (Table V), meaning the effect of grid res-
olution averaged over all structures was less than 1.2%. Note
there was no difference between dose objectives passing or
failing among the four grid resolutions for all cases.
The effect of grid resolution on calculation time and plan
MU is presented in Table VI. The mean plan MU (Table VI)

TABLE VI. Planning calculation time and total plan MU for grid resolution parameter.

Calc. time Total MU for plan

Grid resolution (mm) 3 4 4, 2 flu. 3 4 4, 2 flu.

Mean − 69.8% − 84.4% − 62.7% 0.8% 1.8% 0.4%

Std. dev. 3.0% 2.0% 5.6% 0.5% 0.5% 0.4%
2 mm reference Mean Calc. Time 3.0 min Mean MU 645 MU

Note: flu., fluence grid; 2 mm, 2 mm dose/fluence grid; 3 mm, 3 mm dose/fluence grid; 4 mm, 4 mm dose/fluence grid; 4, 2 flu., 4 mm dose grid with 2 mm fluence grid.

Medical Physics, Vol. 40, No. 6, June 2013

061704-9 Mittauer et al.: Parametric analysis of IMRT planning
increased by 0.8% (3 mm grid resolution), 1.8% (4 mm grid
resolution), and 0.4% (4 mm dose grid with 2 mm fluence
grid resolution). The planning time (Table VI) was reduced by
69.8% (3 mm grid resolution), 84.4% (4 mm grid resolution),
and 62.7% (4 mm dose grid with 2 mm fluence grid resolu-
tion), compared to the mean calculation time for the reference
plan of 3 min.
The 4 mm dose grid with 2 mm fluence grid resolution
saved considerable calculation time, about 2/3 the calculation
time of the 2 mm grid resolution (Table VI). As previously
noted, the 4 mm dose grid with 2 mm fluence grid resolu-
tion was most comparable to the 2 mm grid resolution for
plan quality results (Fig. 3 and Table V). This indicates that a
4 mm dose grid with 2 mm fluence grid resolution is a viable
alternative to a 2 mm grid resolution setting with considera-
tion of dose accuracy and calculation time.
Technically, all segments are the same among plans, since
plans were optimized as the one before the grid resolution
was changed; thus, there is no effect from optimization. Dose
differences are, therefore, a direct measure of the uncertain-
ties of the dose calculation after the plan has been scaled by
the appropriate MUs to achieve the prescription (i.e., PTV 70
D95%). Since it is known that interpolation of a larger grid
resolution leads to these dose differences,17–19 interpolation
was the cause of the dose volume index uncertainty.
F IG . 4. Mean (a) and (b) and maximum (c) and (d) percentage dose differences for all PRVs and PTVs with varying MSAP (a) and (c) and MMUP (b) and (d)
for all cases. Notes: (1) The max is the max magnitude of plan quality degradation being a negative sign for D20% and D99% and a positive sign for all other
structures. (2) Reference taken at 3 MU, 4 cm2 .
Medical Physics, Vol. 40, No. 6, June 2013
061704-9
III.B. Minimum segment area parameter
Figures 4(a)–4(d) illustrate the effects on dose for PTVs
and PRVs with varying MSAP and MMUP. The PTV and
PRV dose differences are noted separately as mean [Figs. 4(a)
and 4(b)] and maximum dose differences [Figs. 4(c) and 4(d)]
with respect to the reference plan (Table I). Note that Fig. 4
adheres to sign convention. A positive dose difference denotes
plan quality degradation for all PRVs and for PTV D20% (i.e.,
PTV hotspot). On the other hand, a negative dose difference
denotes a loss of plan quality for PTV D95% and D99% (i.e.,
PTV coverage).
For MSAP there was a marked increase in the maximum
dose difference in Fig. 4(c) from 4% (MSAP of 5 cm2 ) to
11% (MSAP of 6 cm2 ). Then, dose differences remained be-
low 12% for plans with MSAP < 25 cm2 . Note the anomaly
at a MSAP of 3 cm2 [Fig. 4(c)]. Here, the cochlea had a large
maximum dose difference at 6%, but remained under its dose
objective.
Another measure of how MSAP changes plan quality is
the threshold of MSAP that dose objectives fail. The dose
to PRVs remained within the objectives for all plans with a
MSAP of <6 cm2 . For the PTV objectives, PTV 56 D95%
and D99% failed at greater than 8 and 5 cm2 , respectively.
PTV 70 D20% and D99% objectives also failed at 25 and
061704-10 Mittauer et al.: Parametric analysis of IMRT planning 061704-10

III.D. Minimum MU and segment area parameters

F IG . 5. Percentage dose differences averaged over all PRVs and PTVs with
varying MMUSAP for all cases. Notes: (1) Only D20% is shown for PTV
results due to sign convention. (2) Reference taken at 3 MU, 4 cm2 .
36 cm2 , respectively. Limiting the MSAP to ≤5 cm2 kept all
PRVs and PTVs (except for PTV 56 D99%) within the dose
objectives as well as all dose deviations to under 4% (Fig. 4).
Figure 5 displays the two parameters MSAP and MMUP
changing simultaneously as designated by MMUSAP. Dose
differences were averaged over all PRVs and PTVs (D20%
only), with respect to the reference plan, as noted in Table I.
Note that Fig. 5 uses only D20% for PTVs, which denotes
the hotspot, or the maximum dose encompassing 20% of the
PTV. Figure 5 adheres to sign convention, and thus only an
increase in dose is relevant.
When simultaneously changing both MSAP and MMUP,
the trends of MSAP and MMUP (Fig. 4) were maintained for
the MMUSAP (Fig. 5). Mean dose differences for MMUSAP
remained under 2.5%, but were slightly larger than the re-
spective MSAP or MMUP dose differences of Fig. 4. This
means that as both parameters were increased individually or
together the dose distribution varied more (i.e., the plan be-
came hotter) from reference 3 MU, 4 cm2 plan.

III.C. Minimum MU parameter

Figure 4(b) shows a marked increase in PRV mean dose
differences for plans of MMUP > 10 MU. As the MMUP in-
creases, the PTVs lose dose homogeneity. Specifically, this
result was seen with PTV D20% for MMUP > 10 MU
[Figs. 4(b) and 4(d)].
As described in Sec. III.B, thresholds at which dose objec-
tives are maintained serve as another evaluation of the plan
quality. For all plans with MMUP of <7 MU, dose to the
PRVs remained within the dose objectives. For the PTVs,
dose objectives were met for plans of MMUP <20 MU for
PTV 56 (D95%) and PTV 70 (D20% and D99%). For PTV
56 D99%, plans failed with MMUP ≥4 MU. A MMUP ≤ 5
MU allowed all dose objectives to be met (except for PTV 56
D99%) and dose deviations to be under 4% (Fig. 4).
Overall, plan quality degrades with increasing MSAP or
MMUP (Fig. 4). The effect of MSAP was more marked on
PRVs than on PTVs [Fig. 4(a)]. However, the effect of MMUP
was evident on both PRVs and PTV D20% [Fig. 4(b)]. This
suggests that MMUP may be relaxed, if PRV dose tolerances
and the PTV hotspot are not of concern. But if the PTV
hotspot is an issue, then only MSAP may be relaxed, but cau-
tion should still be used for the dose to PRVs.
III.E. Delivery time model
The machine-derived parameters of our model were com-
pared against vendor specifications. The results are tabulated
in Table VII along with results converted into vendor equiva-
lent units for ease of comparison.
The delivery time model [Eq. (4)] was compared and ver-
ified against 33 delivered plans with times that ranged from
3 to 12 min. Figure 6 shows the results of measured times
for the 33 plans versus calculated times using our model and
the modified Li and Xing20 model (i.e., altered for Elekta de-
livery). The greatest deviations between measured and cal-
culated times were 1.8% (10 s for a 9 min plan) and 5.7%
(10 s for a 3 min plan) and the average deviations were 0.83%
and 1.7% for respective calculated times using our model and
the modified Li and Xing20 model.
The Li and Xing20 model provides a valid formalism for
specific LINACs and delivery types, as previously noted in
Sec. II.E. Our method provides new machine-derived met-
rics and additional terms for a more detailed account of time.
Our model also allows for factors not previously included in
the Li and Xing20 model such as the geometry of unequally
spaced beams, nongridded gun delay due to intersegmental

TABLE VII. Delivery time model parameters of an Elekta LINAC compared to vendor specifications.

Delivery time model parameters Comparison

Parameter Name Measured results (10−3 ) Numerical equivalent Elekta specifications

dt
MLC speed dx 7.70 min/cm 2.2 cm/s 2 cm/s
to 51.17 min 3.1 s/intersegment N/A
dt
Gantry speed dx 2.65 min/degree 57.2 s/rev 60 s/rev
to 36.06 min 2.2 s/interbeam N/A
AFS-gantry T CPG 81.67 min 4.9 s/interbeam N/A
dt
Dose rate dx 1.87 min/MU 526.3 MU/min 535 MU/mina
to 79.18 min 4.8 s/interbeam N/A

Notes: (1) Measured results from characterization testing [Eqs. (1)–(3)] of an Elekta Synergy. (2) Numerical equivalent are measured results converted into the same units
as the vendor specifications.
a
Dose rate is unique to each Elekta LINAC.

Medical Physics, Vol. 40, No. 6, June 2013

061704-11 Mittauer et al.: Parametric analysis of IMRT planning
F IG . 6. Treatment delivery time model tested against 33 delivered IMRT
plans with calculated times from our model and a modified Li and Xing
(Ref. 20) model. Notes: (1) The modified Li and Xing (Ref. 20) model in-
cludes two additional terms for beam-on delay and auto-field-sequence mode
delay. (2) Measured times delivered on an Elekta LINAC.
beam-on delay as well as interbeam beam-on delay, and a
gantry control point delay due to the MOSAIQ AFS mode
to verify gantry position. Overall, our proposed delivery time
formalism provides a valid method to assess delivery time,
and allows for the characterization of the user’s specific ma-
chine, a novel technique.
III.F. Number of segments and plan MU
To assess how varying the parameters of MSAP, MMUP,
and MMUSAP affects the plan quality and delivery time, the
number of segments and plan MU were evaluated. Figure 7
displays the results as percentage differences in the delivery
time, total number of segments, and plan MU, with the refer-
ence plans of Table I.
III.F.1. Plan quality
The factors affecting the dose distribution for MSAP can
be seen in light of the segment shape and the plan MU. As
the MSAP is increased, the total number of segments did not
change considerably [Fig. 7(a)] due to the fact that the MMUP
remained small. Therefore, the optimizer still attempted to
F IG . 7. Percentage difference in treatment delivery time, number of segments, and plan MU for the 3 MU, 4 cm2 reference plan versus plans of varying sizes
of MSAP (a) and MMUP (b) for all cases.
Medical Physics, Vol. 40, No. 6, June 2013
061704-11
utilize all available segments to provide the necessary plan
conformality and PTV coverage. However, when the MSAP
was increased, similar segment shapes were created, the ef-
ficiency of generating plan conformality decreased, the MU
required to obtain PTV coverage decreased, and the overall
result resembles a 3D conformal plan, (i.e., the total MU de-
creased with an increase in the MSAP) [Fig. 7(a)]. Note that
the total delivery time was largely dependent on the number
of segments (Fig. 7), because the MLC movement is the most
time-costly parameter of the whole IMRT delivery process as
described in Sec. III.F.2.
On the other hand, by limiting the MMUP, the number
of segments was affected considerably. As the MMUP is in-
creased, the plan was forced to consolidate segments. This
was because the plan attempts to maintain the same total dose
(i.e., similar MU) to the PTV for a given set of dose objec-
tives. The plan MU remained approximately constant only up
to 10 MMUP; beyond 10 MMUP the plan lost conformality
and efficiency increased (i.e., the plan became similar to a 3D
conformal plan) as seen in Fig. 7(b) and, as expected, the to-
tal delivery time decreased significantly with the number of
segments.
From our previous results of Secs. III.B and III.C, we
found that the effect of MMUP was more costly in terms of
plan quality than the effect of MSAP [Figs. 4(c) and 4(d)].
From the above discussion we can also conclude that the ef-
fect of MMUP on the number of segments was more costly to
the plan quality than the effect of MSAP on the plan MU.
III.F.2. Delivery efficiency
The calculated delivery times for all cases are shown in
Fig. 7 as percentage differences with the mean delivery time
of 9.1 min for reference plans. Delivery efficiency (i.e., treat-
ment delivery time) improved at and above 6 MU for MMUP,
but no marked improvement was found with MSAP (Fig. 7).
For MMUSAP, delivery time was the most efficient in com-
parison to plans with the same MSAP or same MMUP, as
expected.
Thus, it is the intersegmental time (i.e., MLC speed and
gun ramp-up delay) that is more costly than intrasegmental
061704-12 Mittauer et al.: Parametric analysis of IMRT planning
time (i.e., number of MU delivered). Although it was not
seen for the cases in this study, the plan MU could affect
the delivery times for dose rate dependent plans. Li and Xing
showed that a dose rate of 1400 MU/min compared to 400
MU/min can improve delivery efficiency by up to 15% for the
same head and neck IMRT plan.20 Thus, for dose rate depen-
dent cases or for future delivery capabilities with an improved
MLC speed design,20 treatment efficiency could be improved
with MSAP and/or MMUP. But for most current clinical
cases only MMUP, and not MSAP, can improve delivery
efficiency.
III.G. Treatment quality considerations
A longer treatment time will not only increase the chance
of patient movement but also decrease the biological response
of the treatment. Paganetti found that the biological response
is reduced by 12% for a 30 min delivery time as compared
to a 1 min treatment.28 Increasing the beam-on time also in-
creases radiation leakage, which could lead to normal tissues
receiving up to 20% of the prescribed dose.29
Another consideration with regards to the treatment quality
is the deliverability of the plan. The machine’s capabilities,
as described in Sec. II.D, must be taken into consideration
when limiting inverse planning parameters. Dosimetric and
mechanical accuracy testing can be addressed using AAPM
TG 142 guidelines.30
To create a more comprehensive understanding of mod-
ulation parameters, research is needed in the area of opti-
mization objectives to allow for a more efficient IMRT work-
flow. Studying the effect on plan quality and planning time
when changing the optimization grid resolution would be a
useful evaluation for the current clinical practice. Expand-
ing the delivery time model to include sliding window de-
livery and even modulated arc therapy is yet another area of
interest.
IV. CONCLUSION
Our study has shown for a fixed number of beams and
gantry angles, and the greater the number of segments, the
better the plan in terms of dose distribution. However, this oc-
curred at the expense of increased delivery time. Thus, modu-
lation factors such as MMUP, which largely impact the num-
ber of segments, greatly affected plan quality and delivery
time. Based on our results, we give the following planning
parameter setting recommendations: (1) 5 cm2 for MSAP, (2)
5 MU for MMUP, and (3) 4 mm dose grid with 2 mm fluence
grid resolution for the final dose calculation settings. When
required, delivery efficiency can be improved by increasing
either the MMUP or MMUSAP, but not the MSAP. However,
we do not recommend increasing these modulation parame-
ters beyond our recommended settings, for the sake of deliv-
ery efficiency, due to the overall cost to plan quality. The pro-
posed delivery time model provides a valid method to assess
delivery time.
Medical Physics, Vol. 40, No. 6, June 2013
061704-12
ACKNOWLEDGMENT
The authors report no conflicts of interest in conducting the
research.
a) Author to whom correspondence should be addressed. Electronic mail:
liucr@shands.ufl.edu
1 T. Bortfeld, “IMRT: A review and preview,” Phys. Med. Biol. 51, R363–
R379 (2006).
2 B. Lu, J. Li, D. Kahler, G. Yan, K. Mittauer, W. Shi, P. Okunieff, and C. Liu,
“An approach for online evaluations of dose consequences caused by small
rotational setup errors in intracranial stereotactic radiation therapy,” Med.
Phys. 38, 6203–6215 (2011).
3 J. Peng, D. Kahler, J. Li, S. Samant, G. Yan, R. Amdur, and C. Liu, “Char-
acterization of a real-time surface image-guided stereotactic positioning
system,” Med. Phys. 37, 5421–5433 (2010).
4 R. Mohan, M. Arnfield, S. Tong, Q. Wu, and J. Siebers, “The impact of
fluctuations in intensity patterns on the number of monitor units and the
quality and accuracy of intensity modulated radiotherapy,” Med. Phys. 27,
1226–1237 (2000).
5 X. Sun and P. Xia, “A new smoothing procedure to reduce delivery seg-
ments for static MLC-based IMRT planning,” Med. Phys. 31, 1158–1165
(2004).
6 L. Ma, “Smoothing intensity-modulated treatment delivery under hardware
constraints,” Med. Phys. 29, 2937–2945 (2002).
7 S. Spirou, N. Fourier-Bidoz, J. Yang, C. Chui, and C. Ling, “Smoothing
intensity-modulated beam profiles to improve the efficiency of delivery,”
Med. Phys. 28, 2105–2112 (2001).
8 M. Coselmon, J. Moran, J. Radawski, and B. Fraass, “Improving IMRT
delivery efficiency using intensity limits during inverse planning,” Med.
Phys. 32, 1234–1245 (2005).
9 M. Matuszak, E. Larsen, and B. Fraass, “Reduction of IMRT complexity
through the use of beam modulation penalties in the objective function,”
Med. Phys. 34, 507–520 (2007).
10 M. Matuszak, E. Larsen, K. Jee, D. McShan, and B. Fraass, “Adaptive dif-
fusion smoothing: A diffusion-based method to reduce IMRT field com-
plexity,” Med. Phys. 35, 1532–1546 (2008).
11 D. Craft, P. Suus, and T. Bortfield, “The tradeoff between treatment plan
quality and required number of monitor units in intensity-modulated radio-
therapy,” Int. J. Radiat. Oncol., Biol., Phys. 67, 1596–1605 (2007).
12 S. Webb, “A simple method to control aspects of fluence modulation in
IMRT planning,” Phys. Med. Biol. 46, N187–N195 (2001).
13 C. McGarry, C. Chinneck, M. O’Toole, J. O’Sullivan, K. Prise, and
A. Hounsell, “Assessing software upgrades, plan properties and patient
geometry using intensity modulated radiation therapy (IMRT) complexity
metrics,” Med. Phys. 38, 2027–2034 (2011).
14 A. McNiven, M. Sharpe, and T. Purdie, “A new metric for assessing IMRT
modulation complexity and plan deliverability,” Med. Phys. 37, 505–515
(2010).
15 E. Ludlum and P. Xia, “Comparison of IMRT planning with two-step and
one-step optimization: A way to simplify IMRT,” Phys. Med. Biol. 53,
807–821 (2008).
16 B. Hårdemark, A. Liander, H. Rehbinder, and J. Loef, “P3IMRT direct ma-
chine parameter optimization,” Pinnacle White Paper, No. 4535 983 02483,
Philips (2004).
17 A. Niemierko and M. Goitein, “The influence of the size of the grid used
for dose calculation on the accuracy of dose estimation,” Med. Phys. 16,
239–247 (1989).
18 J. Dempsey, H. Romeijin, J. Li, D. Low, and J. Palta, “A Fourier analysis
of the dose grid resolution for accurate IMRT fluence map optimization,”
Med. Phys. 32, 380–388 (2005).
19 H. Chung, H. Jin, J. Palta, T. Suh, and S. Kim, “Dose variations with vary-
ing calculation grid size in head and neck IMRT,” Phys. Med. Biol. 51,
4841–4856 (2006).
20 R. Li and L. Xing, “Bridging the gap between IMRT and VMAT: Dense an-
gularity sampled and sparse intensity modulated radiation therapy,” Med.
Phys. 38, 4912–4919 (2011).
21 R. Popple, J. Fiveash, and I. Brezovich, “Effect of beam number on organ-
at-risk sparing in dynamic multileaf collimator delivery of intensity modu-
lated radiation therapy,” Med. Phys. 34, 3752–3759 (2007).
22 J. Li, C. Liu, S. Kim, R. Amdur, and J. Palta, “Matching IMRT fields with
static photon field in the treatment of head-and-neck cancer,” Med. Dosim.
30, 135–138 (2005).
061704-13 Mittauer et al.: Parametric analysis of IMRT planning
23 L. Marks, E. Yorke, A. Jackson, R. Ten Haken, L. Constine, A. Eisbruch,
S. Bentzen, J. Nam, and J. Deasy, “Use of normal tissue complication prob-
ability models in the clinic,” Int. J. Radiat. Oncol., Biol., Phys. 76, S10–S19
(2010).
24 T. J. Jordan and P. C. Williams, ‘‘The design and performance char-
acteristics of a multileaf collimator,’’ Phys. Med. Biol. 39, 231–251
(1994).
25 C. Liu, T. Simon, C. Fox, J. Li, and J. Palta, “Multileaf collimator character-
istics and reliability requirements for IMRT Elekta system,” Int. J. Radiat.
Oncol., Biol., Phys. 71, S89–S92 (2008).
26 T. Simon, D. Kahler, W. Simon, C. Fox, J. Li, J. Palta, and C. Liu, “An
MLC calibration method using a detector array,” Med. Phys. 36, 4495–
4503 (2009).
Medical Physics, Vol. 40, No. 6, June 2013
061704-13
27 M. Sharpe, B. Miller, D. Yan, and J. Wong, “Monitor unit settings for in-
tensity modulated beams delivered using a step-and-shoot approach,” Med.
Phys. 27, 2719–2725 (2000).
28 H. Paganetti, “Changes in tumor cell response due to prolonged dose de-
livery times in fractionated radiation therapy,” Int. J. Radiat. Oncol., Biol.,
Phys. 63, 892–900 (2005).
29 E. E. Klein and D. A. Low, “Interleaf leakage for 5 and 10 mm dynamic
multileaf collimation systems incorporating patient motion,” Med. Phys.
28, 1703–1710 (2001).
30 E. E. Klein, J. Hanley, J. E. Bayouth, F. Yin, W. Simon, S. Dresser, C.
Serago, F. Aguirre, L. Ma, B. Arjomandy, and C. Liu, “Task Group 142
report: Quality assurance of medical accelerators,” Med. Phys. 36, 4197–
4212 (2009).