Home Search Collections Journals About Contact us My IOPscience

Dosimetric verification of a commercial inverse treatment planning system

This content has been downloaded from IOPscience. Please scroll down to see the full text.

1999 Phys. Med. Biol. 44 463

(http://iopscience.iop.org/0031-9155/44/2/013)

View the table of contents for this issue, or go to the journal homepage for more

Download details:

IP Address: 134.99.128.41
This content was downloaded on 05/12/2013 at 20:51

Please note that terms and conditions apply.

Phys. Med. Biol. 44 (1999) 463–478. Printed in the UK PII: S0031-9155(99)96991-9

Dosimetric verification of a commercial inverse treatment

planning system

Lei Xing†§, Bruce Curran‡, Robert Hill‡, Tim Holmes‡, Lijun Ma†,
Kenneth M Forster† and Arthur L Boyer†
† Department of Radiation Oncology, Stanford University, 300 Pasteur Drive, Stanford,
CA 94305-5304, USA
‡ Nomos Corporation, 2591 Wexford-Bayne Road, Sewickley, PA 15143, USA

Received 25 August 1998, in final form 27 October 1998

Abstract. A commercial three-dimensional (3D) inverse treatment planning system, CorvusTM

(Nomos Corporation, Sewickley, PA), was recently made available. This paper reports our
preliminary results and experience with commissioning this system for clinical implementation.
This system uses a simulated annealing inverse planning algorithm to calculate intensity-modulated
fields. The intensity-modulated fields are divided into beam profiles that can be delivered by means
of a sequence of leaf settings by a multileaf collimator (MLC). The treatments are delivered using
a computer-controlled MLC. To test the dose calculation algorithm used by the Corvus software,
the dose distributions for single rectangularly shaped fields were compared with water phantom
scan data. The dose distributions predicted to be delivered by multiple fields were measured using
an ion chamber that could be positioned in a rotatable cylindrical water phantom. Integrated
charge collected by the ion chamber was used to check the absolute dose of single- and multifield
intensity modulated treatments at various spatial points. The measured and predicted doses were
found to agree to within 4% at all measurement points. Another set of measurements used a
cubic polystyrene phantom with radiographic film to record the radiation dose distribution. The
films were calibrated and scanned to yield two-dimensional isodose distributions. Finally, a beam
imaging system (BIS) was used to measure the intensity-modulated x-ray beam patterns in the
beam’s-eye view. The BIS-measured images were then compared with a theoretical calculation
based on the MLC leaf sequence files to verify that the treatment would be executed accurately
and without machine faults. Excellent correlation (correlation coefficients > 0.96) was found for
all cases. Treatment plans generated using intensity-modulated beams appear to be suitable for
treatment of irregularly shaped tumours adjacent to critical structures. The results indicated that
the system has potential for clinical radiation treatment planning and delivery and may in the future
reduce treatment complexity.

1. Introduction

Recent developments in external beam radiation therapy exploit the use of fields in which
intensity is varied across the beams. It has been demonstrated that dose distributions can
be produced to conform to complex target shapes using intensity modulated photon beams.
The beam fluence maps needed to achieve the dose distributions are usually generated using
techniques now described as inverse planning. Inverse planning begins with a prescribed dose
distribution or prescribed dose volume histograms (DVHs) to the target and sensitive structures,
and derives a set of intensity modulated beams by optimizing an objective function (Brahme
1988, Bortfeld et al 1990, Mackie et al 1993, Webb 1994, Mohan et al 1994, Holmes et al 1995,
§ E-mail address: lei@reyes.stanford.edu

0031-9155/99/020463+16$19.50 © 1999 IOP Publishing Ltd 463

464 L Xing et al

Wang et al 1996, Yu et al 1995, Xing and Chen 1996, Xing et al 1998). This study investigated
the use of the CorvusTM inverse treatment planning system (Nomos Corporation, Sewickley,
PA) for 3D intensity modulated radiation therapy (IMRT). The Corvus is an extension of
the PeacockTM system developed by the same company (Carol et al 1992, Woo et al 1996).
The Peacock system is limited solely to the use of MIMiC (multivane intensity-modulation
collimator) for dose delivery in a slice by slice manner. The Corvus system is capable of
computing inverse plans that can be delivered using multiple fixed-gantry fields whose entire
areas up to 40 × 40 cm2 are modulated by full-field MLC. (The maximum field size without
MLC carriage movement is 14 × 4 cm2 for Varian accelerators (Varian Oncology Systems,
Palo Alto, CA). Larger field sizes in the direction of leaf motion are treated as abutting fields.)
A typical Corvus system consists of a planning workstation, a compute engine, and
ancillary input/output devices. The planning workstation is a PC-based computer with
NEXTStep operating system (Apple Computer, Inc., Cupertino, CA). The compute engine
consists of 2-B Pentium or Pentium II processors (Intel Corporation, Santa Clara, CA)
loosely coupled in groups of 2, 4 or 6, depending on the particular configuration. The
Corvus system allows multiple plans to be queued and calculated while the user is free to
review calculation results or plan additional patients. Corvus supports input of CT and MR
patient volume sets via various proprietary image protocols as well as DICOM (Digital Image
Communication in Medicine) (NEMA 1998). The inverse planning process is carried out
in eight steps in order to obtain an optimal plan. These include: (1) acquisition of patient
information; (2) image registration; (3) image fusion (optional); (4) identifying target volumes
and anatomical structures; (5) setting dose prescriptions; (6) computing the optimal intensity
maps and computing dose distributions; (7) displaying the results; (8) generating MLC leaf
sequencing files. The first four steps are similar to those used for conventional treatment
planning. These steps simply establish patient data, generate a geometric model of the patient,
and identify the targets and sensitive structures. The image fusion can be performed on CT or
MRI acquired in any of the standard orthogonal views.
The prescription step is worth discussing in detail. Parameters describing the expected
immobilization uncertainty (a value of spatial uncertainty associated with each of the three
orthogonal axes) and localization uncertainty (additional spatial increments along each of the
three axes) are entered and used to determine the planning target volume (PTV) from the target
and structure contours (ICRU 1976) entered at the target and normal structure entry step. A
dosimetric panel allows entry of partial volume prescription information for both target and
sensitive structures. Specifically, one enters the desired dose to the target(s) along with the
maximum and minimum doses to the target(s) and the percentage(s) of the target volume(s)
which are allowed to receive a dose less than the desired value(s). Corvus allows physicians
freedom to prescribe not just to target dose, but also to specify the tolerance dose to the normal
structures.
The degree to which a plan is optimized is determined in part by constraints placed on
the planning algorithm. The planner has direct control over these constraints, which includes
dose limits to the sensitive structures and the number of fixed gantry positions in the treatment
plan. Using a computer engine with eight processors, it usually takes 30 min to 1 h to finish the
calculation, depending on the number of fields, the extent of the target volume and the size of
the dose matrix. Although this may seem to be a long time, the operator can leave the system
unattended or work on other plans while the optimization continues on the compute engine.
The optimization of a treatment plan is accomplished through the use of a simulated
annealing algorithm (Webb 1994). After the plan is approved, the beam intensity profiles are
converted into MLC leaf positioning sequences as a function of monitor units (Bortfeld et al
1994, L Ma et al 1998). The operator may choose either a dynamic MLC (sliding-window) or
 Verification of commercial treatment planning system 465

a static (step and shoot) delivery. The electronic files containing the leaf sequences are written
onto a floppy disk after the approval of the treatment plan. The file is transferred from the
floppy disk to the accelerator MLC workstation for dynamic delivery.
The purpose of this article is to report on a preliminary survey of the Corvus inverse
planning system. In particular, we will discuss the dosimetric properties and some of
the practical aspects of the system, which include the performance of the inverse planning
optimization algorithms, the accuracy of the dose calculation and the efficiency of the system.
The results indicated that the dosimetry of the Corvus system is clinically acceptable and that
the system has potential for clinical IMRT application.

2. Methods and materials

The verification of the inverse treatment planning system consisted of three separate but related
investigations. The first study was concerned with the system’s ability to accurately compute
broad beam data. For this purpose, dose distributions computed by Corvus for a series of open
fields were calculated and compared with water phantom scan data. This is common to all
treatment planning systems, including conventional 3D treatment planning systems. In order
to examine the accuracy of the dose calculation model in the case of intensity-modulated beams
and the dynamic MLC delivery system, the behaviour of the system for several special intensity
patterns was then studied. Finally, a preliminary survey of the type of dose distributions that
Corvus might achieve was carried out using a few hypothetical phantom cases. The accuracy
of plan delivery at a few points was assessed using ion-chamber measurements. The last two
studies are unique to IMRT.
We will start by briefly discussing the validation of the dose calculation. The dose
calculation in Corvus uses a finite pencil beam algorithm (Bourland and Chaney 1992). Each
treatment field is decomposed into a set of 1 cm × 1 cm beamlets, each of whose intensity can
vary from 0–100% independent of all other beamlets. Parameters used by the beamlet models
are extracted from film measurements of 1 cm wide fan beams characterized by their depth
dose and off-axis properties. The beamlets are directed along the ray lines starting at the x-ray
source. To test the dose calculation algorithm, dose distributions of open beams with varying
field sizes were calculated for a cubic phantom. The calculated per cent depth dose (PDD)
curves, beam profiles and their field size dependence were then compared with water phantom
scan data. Neither the optimizer nor the dynamic MLC delivery system were involved with
this part of the commissioning so that only the dose calculation model was being tested. This
comparison yielded information on the accuracy of the dose model in reproducing the dose
distribution for static rectangular fields.
Verification of conventional 3D treatment planning systems is achieved by demonstrating
agreement between computed and measured PDD curves, isodose distributions, beam profiles
and their field size dependence for a set of rectangular open and wedged fields. Detailed
guidelines have been recommended by the International Committee of Radiation Units and
Measurements (ICRU) and described in other publications (Van Dyk et al 1993, Fraass and
McShan 1995). Accurate modelling of static fields does not guarantee that the calculated dose
will agree with the delivered dose in the case of intensity-modulated beams. In the second
part of the study, measurements of several specially designed intensity-modulated beams were
made and compared with the calculated dose distribution.
It is impractical to validate the system by experimentally checking every possible intensity-
modulated beam. We have checked 15 representative beam patterns to ensure that the calculated
and measured dose are in accordance with each other. Figure 1 shows a few examples of
the intensity maps (the grey-scale figures were computed using the given intensity profiles).
466 L Xing et al

Figure 1. Intensity-modulated beam patterns designed for testing the dose calculation algorithm
and dynamic MLC delivery of an inverse treatment planning system.

Figure 1(g) is five MLC-collimated stripes produced by three consecutive movements of an

MLC, and figure 1(h) is an open field produced by moving a 2 cm×10 cm slit in five steps. Other
beam patterns in figure 1 are self-evident. This procedure involved both the dose calculation
and the dynamic MLC delivery of the system. These beams were measured using radiographic
films (XV2 Verification Film, Kodak, Rochester, NY) to record the dose delivered to a cubic
solid water phantom using a 100 cm source–surface distance (SSD). The processed films
were scanned with a computer-controlled densitometer (Vidar Systems Corporation, Herndon,
 Verification of commercial treatment planning system 467

VA) and the optical densities were converted to dose using a calibration curve obtained from
irradiation of a series of verification films in the same pack. The measurements were then
compared with the Corvus calculations.
A cylindrical water phantom containing a movable ion chamber holder was designed for
accurate dose measurement of the system. A photo of the phantom is shown in figure 2. The
diameter of the phantom was 30 cm. During a measurement, the phantom was positioned
with its cylindrical axis perpendicular to the couch top and the gantry was fixed to direct the
beam horizontally. The phantom was supported by a bearing allowing for rotation about the
cylindrical axis of the phantom, which simulates gantry rotation and allows one to measure
the dose of a multifield plan without rotating the gantry. By using the cylindrical rotation of the
phantom to simulate gantry rotation, we have implicitly ignored gantry sag. The acceptance
test data on the machines, from the commissioning stage, indicated a gantry axis run-out less
than ±1 mm. This was typical of the gantry sag reported for linear accelerators (Curran et al
1992) and any error in dose caused by ignoring this effect should be negligible. The ion-
chamber holder in this phantom was designed to move accurately along the radius and length
of the phantom. It was found that the overall uncertainty of the holder movement was within
±1.3 mm. A set of computerized tomography (CT) images was created for this phantom.
Fiducial marks were used to place the isocentre of the treatment plan at the centre of the
cylindrical phantom.

Figure 2. A photo of the cylindrical water phantom for spot check of the absolute dose.

In this study, a hypothetical S-shaped target (see figure 3(a)) was created at the centre of
the phantom. The two areas inside the S-shaped target simulated the sensitive structures. The
target and sensitive organs were delineated using the Corvus system segmentation tools. Other
hypothetical cases examined included a cylindrical target (figure 3(b)), a doughnut-shaped
target (figure 3(c)) with a central sensitive structure and, a C-shaped target (figure 3(d)).
Plans using one, three and nine coplanar fields distributed at equal angular intervals were
produced using Corvus for the above hypothetical cases. No attempt was made to optimize the
beam angulation. Plan design balanced the need to deliver a uniform high dose to the target
volume while not exceeding sensitive structure dose limits. The inverse planning algorithm
468 L Xing et al

Figure 3. A few hypothetical study cases created in the cylindrical water phantom. These cases
were used to study the accuracy of the dose calculations in the Corvus inverse planning system.

computed beam intensity profiles that achieved the best compromise dose distributions. Upon
approval of these plans, the corresponding leaf sequence files were generated. A Varian
2300 C/D with 6 MV photons was used to deliver the intensity-modulated fields. The leaf
sequence files were read by the Varian MLC software and used to control the position of
leaves as a function of accumulated monitor units. The doses at a few spatial points were
measured using a 0.147 cm3 IC-10 ionization chamber with diameter of 4 mm and a height
of 6 mm (Wellhöffer Dosimetrie, Schwarzenbruck, Germany) using recommendations of the
AAPM protocol (AAPM 1983). No corrections were made for the variation in the chamber
replacement effect, which depends on the dose gradient at the measurement point and the
chamber diameter. This procedure examined the optimization, dose calculation algorithm and
the delivery algorithm, providing a systematic test of the inverse planning system.
The film measurements were performed in either coronal planes (beams perpendicular to
the film plane) or axial planes (beams parallel to the film plane and directed toward the film
edges) of a 30 × 30 × 30 cm3 cubic phantom. While a film measurement in a coronal plane can
yield accurate results if careful calibration and quality control over the processing procedure is
ensured, it can be used to study only one beam at a time. The axial film measurement may be
less accurate because of the depth dependence of the film response calibration. Nevertheless,
the axial film measurement is convenient as a preliminary review of the dose distribution,
especially in the case of multifield treatments. In order to use measurements in a cubic or
cylindrical phantom as a surrogate for measurements in the patient, it is required to replace the
CT data acquired from the patient or another phantom by the CT data acquired from the cubic
phantom and to recalculate the SSDs and the dose distribution. The calculations made using the
 Verification of commercial treatment planning system 469

cubic film phantom were then directly comparable with the film measurements. The calculated
and measured dose distributions were compared by manually overlaying the measured dose
distribution on the calculated distributions.
Further study included using a beam-imaging system (BIS) (Wellhöffer Dosimetrie,
Schwarzenbruck, Germany) to measure the intensity-modulated x-ray beam patterns in the
beam’s-eye view (BEV), which were then compared with the reference images derived from
the Corvus MLC leaf sequencing files. The BIS is a video-optical electronic portal image
device (Boyer et al 1992, L Ma et al 1997). Various characteristic properties of the device
have been established in an earlier study (L Ma et al 1997). The device was used to verify that
the leaf sequence files were transferred correctly to the linac MLC control computer and were
accurately executed without machine faults. The correlation between the BIS measurements
with calculated reference images were evaluated for all incident beams.

3. Results and discussion

The dose calculation algorithm was first examined by using a series of rectangular uniform
fields ranging from 2 × 2 to 35 × 35 cm2 . In figure 4, the calculated PDD curves for a series
of fields are compared with measured results obtained using a computer-controlled water
phantom scanner. Differences of less than 2% were demonstrated between the measured and
the calculated PDDs for these and all other field sizes studied. The calculated in-plane and
the cross-plane beam profiles at various depth are also plotted as broken curves in figure 4,
for comparison with the water phantom scan data that are plotted as full curves. The general
agreement between the two was better than 3%. However, the calculation underevaluated the
dose at the shoulders of the profiles. At the field edges, it is more appropriate to compare
spatial displacement of the isodose curves and the agreement was found to be within 3.8 mm
between the 40% isodose lines. These results indicated that the finite pencil beam algorithm
in Corvus produced clinically acceptable dose distributions for conventional treatment fields.
The dynamic delivery test started from a set of specially designed intensity modulated
patterns shown in figure 1. Dose distributions of these intensity modulated beams in a
30×30×30 cm3 cubic phantom with 100 SSD were computed and are shown in Figure 5. A few
isodose distributions, obtained using radiographic films in the coronal plane at 2dmax = 2.4 cm
depth in the cubic phantom, are shown in figure 5 together with the calculation results. It was
found that the agreement between measured and calculated isodose lines was less than 3 mm
for all beam patterns tested, except near the interleaf leakage regions where displacement of
up to 4.3 mm was found. It is expected that the interleaf leakage will be partially smeared out
for multifield treatments.
The system was further tested by creating a hypothetical S-shaped PTV in the cylindrical
water phantom (figure 3(a)). Isodose curves computed using one and nine coplanar beams at
equally spaced angles are shown in figure 6. The doses were normalized to the maximum dose.
The dose distribution calculated for a single-field plan is given in figure 6(a). The purpose
of running a single-field plan here was to examine the performance of the inverse planning
algorithm in the limit of a single field, where the solution can be intuitively conceived. For a
single field, no phantom rotation (to simulate the gantry rotation) was needed to measure the
absolute dose, and thus any error caused by the phantom rotation was avoided. The fluence
for this plan shows less intensity passing through the sensitive structures and much more
passing through the PTV (see figure 6(a)). The relative weights of the two types of fluence
were determined by the prescribed DVHs to the PTV and the sensitive structures. As with
all inverse planning optimization, dosimetric compromises were made between delivering full
dose to the target or sparing the sensitive structures because the two are immediately adjacent
470 L Xing et al

(a)

(b)

Figure 4. Per cent depth dose curves and in-plane and cross-plane beam profiles for a series of
field sizes (10 × 4, 4 × 4, 10 × 10 and 20 × 20 cm2 ). The broken curves were obtained using the
Corvus and the full curves are water phantom scan data.

to each other. For the nine-field plan, shown in figure 6(b), the gradient at the boundary of the
target and the sensitive structures was steep and the resulting dose distribution demonstrated
a sensible compromise between treating the PTV and sparing the sensitive structures. Of
course, a visual examination does not provide the quantitative measure of the ‘goodness’ of an
optimizer. It would be useful to introduce some conformality indices based on the volumetric
information of a dose distribution for the evaluation of IMRT plans.
 Verification of commercial treatment planning system 471

(c)

(d)

Figure 4. (Continued)

The dose distributions shown in figure 6 were verified experimentally using an ion chamber
and films. The ion chamber measurements and the calculation at various points of interest for
the one- and nine-field plans are summarized in figure 6. In these two particular cases, the
measurements for all points were within 3% of the predicted values at both high-dose and low-
dose points. Similar studies were done for IMRT plans using one, three and seven fields for
doughnut-shaped and C-shaped targets at the centre of the cylindrical phantom (see figure 3).
For all cases, the measurements agreed with the calculation to within 4%. It is useful to
472 L Xing et al

(a)

(b)

Figure 5. Isodose curves in the coronal plane at depth 2.4 cm obtained using radiographic films
for intensity-modulated beams shown in figures 1(a), 1(d), 1(e) and 1(h). The calculation results
of the Corvus are plotted along with the measurements (right-hand side).

mention that the monitor unit calculation used for these studies partially accounted for MLC
transmission (Holmes et al 1997). In an earlier version of the Corvus system, the transmission
was ignored, and it was found that the overall agreement was about 5% to 9%, indicating
the importance of accounting for the leaf transmission in the delivery of IMRT. Because of
the relatively large monitor units involved in the treatment, the leakage should be taken into
account as accurately as possible in any IMRT planning system.
Ion chamber measurements were time consuming, as anticipated, and could only measure
the dose point-by-point. It is important to verify the spatial dose distribution. Unfortunately,
for dynamic fields this cannot be done without using unconventional instrumentation such as
an array of ion chambers or TLD detectors. However, it was useful to recalculate the dose
distribution in a solid water phantom and to perform film measurements to indirectly confirm
 Verification of commercial treatment planning system 473

(c)

(d)

Figure 5. (Continued)

the spatial dose distribution. The film measurement was primarily concerned with investigating
the relative dosimetric agreement between the planned and delivered dose distributions because
of the limitations of using film for absolute dosimetry. The uncertainty inherent in the film
dosimetry can be partially compensated for by ion chamber measurement. For the film
measurements, the intensity-modulated fields used to compute the dose distributions shown
in figure 6 were used to recalculate the dose distribution for the cubic film phantom. Upon
the completion of the dose calculations, leaf sequence files were generated using a step-and-
shoot leaf sequencing approach (Bortfeld et al 1994, L Ma et al 1998). Figure 7 provides
film measurements of the nine-field dose distributions in the central axial plane of the cubic
phantom (corresponding to the calculations given in figure 6(b)). The main features of the
calculated dose distribution were seen in the delivered dose distribution. The measured 90%
isodose curves were found to lie within 3 mm of the calculated isodose curves. At the 60%
isodose level, consistent spatial agreement was found to within 4 mm, except in the interleaf
474 L Xing et al

Figure 6. One- and nine-field plans (on the central axial plane) for an S-shaped target located at
the centre of the cylindrical water phantom. The doses at the circular points were measured using
an ion chamber. The discrepancy between the calculation and measurement is shown in the figures
as a percentage.

Figure 7. Measured (a) and calculated (b), the calculation phantom has a uniform tissue density
and the circular region inside the phantom is an artefact of creating the cubic phantom on top
of the existing circular phantom scan) isodose curves in the central axial plane of the cubic film
phantom. The isodose curves for the cubic phantom was obtained by using the nine optimal
intensity-modulated beams for the cylindrical water phantom (figure 6(b)).

leakage region. It is expected that the agreement could be improved by further research into
how the leakage dose and head scatter should be incorporated into the dose calculation.
Given a prescribed dose and dose rate, the delivery time of a conventional isocentric
treatment depends mainly on the depth of prescription point, field size and the wedge filter
if there is one. In addition to the depth of prescription point, the delivery time of an IMRT
field also depends on the fractional contribution of the field to the prescribed point (similar
to the concept of a wedge factor) and the overall intensity profile. A complex beam profile,
 Verification of commercial treatment planning system 475

appearing in the case of complex target- and sensitive-structure geometries, usually results in
a larger number of field segments and a longer delivery time. In table 1 we list the delivery
time along with the monitor units and number of field segments for each of the nine fields in
the plan shown in figure 6(b). The dose rate used for delivery was 300 MU min−1 and 1 Gy
was prescribed to the point of maximum dose for each incident beam.

Table 1. Delivery time of the nine-field plan shown in figure 6(b).

Gantry angle Delivery time

(deg) Monitor units Field segments (min)
340 58 39 0.62
300 65 66 0.77
260 69 41 0.65
220 47 67 0.70
180 29 33 0.42
140 54 40 0.67
100 76 38 0.65
60 61 47 0.65
20 40 54 0.68

Finally, a BIS was used to measure the fluence distribution of the incident intensity-
modulated fields. These measurements were similar to a film measurement on a plane
perpendicular to the incident beam. But using the BIS was more efficient and the images
obtained could be easily compared quantitatively with the theoretical calculation obtained
from the MLC leaf sequence files. In figure 8 the measured fluence patterns of the nine
incident beams in figure 6(b) were presented along with the theoretical calculations. It is seen
that the two sets of images appear identical to each other. The correlation coefficients for the
beams with gantry angle 340◦ , 300◦ , 260◦ , 220◦ , 180◦ , 140◦ , 100◦ , 60◦ and 20◦ were found to
be 0.97, 0.96, 0.99, 0.97, 0.96, 0.99, 0.96, 0.99 and 0.96 respectively, indicating an excellent
agreement for each beam (L Ma et al 1997). For testing purposes, some random errors were
introduced into the leaf sequence files (reference images) and the global correlation coefficient
was able to detect uncertainties of less than 0.6 mm in the MLC leaf motion. Gross errors
such as a flipped or rotated reference image from an erratic leaf sequencing file were readily
detected by the BIS system.
In addition to the S-shaped target, other simply shaped targets, such as cylindrical,
doughnut-shaped and C-shaped targets, were tested. Similar agreements were found in the
measurements using ion-chamber, films and the BIS. The simple yet non-trivial shapes of the
phantom and targets provided useful tests of the system. It is important to use some simple,
intuitive cases in commissioning and testing an inverse treatment planning system, otherwise
the results could be complicated by the geometries of clinical patient anatomy and target
volumes, and much useful physics could be buried. Once the results for these cases are fully
understood, one can go further to deal with more complex cases.
In the above discussion, the plans were generated and delivered in a cubic or cylindrical
phantom. In general, a patient contour is not regularly shaped. The treatment, however, can be
verified by using simple phantoms: the optimized intensity profiles for a patient can be used to
compute surrogate test dose distribution in a simple phantom (the phantom does not have to be
regularly shaped) by using software made available in Corvus. The plan can then be verified
using film, TLD or ion chamber measurements in the simple phantom. Of course, it is not
totally satisfactory to transfer the dose distributions to cubic or cylindrical phantoms. Futher
study should include measurements in an anthropomorphic phantom. Finally, we mention that
476 L Xing et al

Figure 8. Intensity patterns for the nine beams in figure 6(b). The left of each beam is the calculated
reference image from the Corvus leaf sequence file. The right part is the BIS-measured image.

within the context of this paper we have not investigated how well the Corvus system models
the tissue density inhomogeneity. This is a very important issue and deserves a thorough study
using inhomogeneity phantom and Monte Carlo simulation (C Ma et al 1998).

4. Summary

The commissioning of conventional treatment planning systems has been extensively discussed
in the literature (Van Dyk et al 1993, Fraass and McShan 1995). The recommended procedures
check whether the output data from the system matches the input data, i.e. the beam data
equivalence. IMRT using an MLC is a new advancement of radiation therapy, and as yet there
are no detailed descriptions on commissioning an inverse treatment planning system in the
literature. The emergence of IMRT posts a new challenge to medical physicists. In this system
an open field is only a special case, and many new features, which include the optimization,
intensity modulation and the leaf sequence, need to be tested and validated.
In this paper, the dosimetric properties of the Corvus inverse planning system was reported.
The procedures we used also apply to the commissioning of any other inverse treatment
planning system. Three levels of testing have been proposed in order to ensure that the
system can be used safely and accurately. These procedures include: (i) checking the dose
calculation model for a series of open fields, (ii) testing the delivery system with several
specially designed intensity patterns and (iii) examining typical hypothetical phantom cases.
The first part of the study was concerned with the dose calculation model. The second part
dealt with the dose model and the delivery process. The third study considered all aspects
of the system, from planning and dose calculation to the plan delivery. The calculated dose
 Verification of commercial treatment planning system 477

distributions were compared with the measurements using ion chamber and radiographic films.
The clear correspondence between the calculated and delivered dose distributions demonstrated
the accuracy of the system. A BIS was employed to measure the fluence of incident beams,
which showed a strong correlation with the theoretical prediction based on the generated leaf
sequence files. The results indicated that the Corvus inverse treatment planning system is able
to produce sensible plans for complicated geometries, and to write leaf-sequence files and
monitor unit specifications capable of delivering the plan with acceptable accuracy.

Acknowledgments

This work was partially supported by a seed grant of the Radiological Society of North America
(RSNA), grant no CA43840 from the National Cancer Institute (NCI) and grant no IRG-58-
008-40 from the American Cancer Society (ACS). Colour illustrations of figures 2, 3, 5, 6 and
7 can be viewed at http://ipomoea.stanford.edu/lei/COMM/figures.html. The contents of this
work are solely the views of the authors and do not necessarily represent the official views of
the RSNA, NCI and ACS. The authors wish to thank S Brain, D Findley, G Dalbow, P Geis
and P Xia for many useful discussions.

References

AAPM TG-21 1983 A protocol for the determination of absorbed dose from high energy photon and electron beams
Med. Phys. 10 741–71
Brahme A 1988 Optimization of stationary and moving beam radiation therapy techniques Radiother. Oncol. 12
129–40
Bortfeld T, Boyer A L, Schlegel W, Kahler D L and Waldron T J 1994 Realization and verification of three-dimensional
conformation radiotherapy with modulated fields Int. J. Radiat. Oncol. Biol. Phys. 30 899–908
Bortfeld T, Ürkelbach J, Boesecke R and Schlegel W 1990 Methods of image reconstruction from projections applied
to conformation therapy Phys. Med. Biol. 35 1423–34
Bourland J D and Chaney E L 1992 A finite-size pencil beam model for photon dose calculations in three dimensions
Med. Phys. 19 1401–12
Boyer A L, Antonuk L, Fenster A, Van Herk M, Meertens M, Munro P, Reinstein L E and Wong J 1992 A review of
electronic portal imaging device Med. Phys. 19 1–16
Carol M, Targovnik H, Smith D and Cahill D 1992 3D planning and delivery system for optimized conformal therapy
Int. J. Radiat. Oncol. Biol. Phys. 24 (suppl 1) 158
Curran B H, Tsai J S, Engler M J and Sternick E S 1992 Measurement and evaluation of linear accelerator isocenter
movement for stereotactic radiation therapy Int. J. Radiat. Oncol. Biol. Phys. 24 (suppl 1) 231
Fraass B A and McShan D L 1995 Three-dimensional photon beam treatment planning Radiation Therapy Physics
ed A R Smith (Berlin: Springer) pp 43–93
Holmes T W, Bleier A R, Carol M P, Curran B H, Kania A A, Lalonde R J, Larson L S and Sternick E S 1997 The
effect of MLC leakage on the calculation and delivery of intensity modulated radiation therapy Proc. XII Int.
Conf. on the Use of Computers in Radiation Therapy ed D Leavitt and G Starkschall (Madison, WI: Medical
Physics Publishing) pp 1215–25
Holmes T W, Mackie T R and Reckwerdt P 1995 An iterative filtered backprojection inverse treatment planning
algorithm for tomotherapy Int. J. Radiat. Oncol. Biol. Phys. 32 1215–25
ICRU 1976 Determination of absorbed dose in a patient irradiated by beams of x or gamma rays in radiotherapy
procedures ICRU Report 24 (Washington, DC: ICRU)
Ma L, Boyer A L, Xing L and Ma C 1998 An optimized leaf-setting algorithm for beam intensity modulation using
dynamic multileaf collimators Phys. Med. Biol. 43 1629–43
Ma L, Geis P B and Boyer A L 1997 Quality assurance for dynamic multileaf collimator modulated fields using a fast
beam imaging system Med. Phys. 24 1213–20
Ma C, Xing L, Mok E, Kapur A and Boyer A L 1998 Verification of IMRT dose calculation using Monte Carlo
simulation Int. J. Radiat. Oncol. Biol. Phys. 42 207
Mackie T R, Holmes T W, Swerdloff S, Reckwerdt P, Deasy J O, Yang J, Paliwal B and Kinsella T 1993 Tomotherapy:
a new concept for the delivery of dynamic conformal therapy Med. Phys. 20 1709–19
478 L Xing et al

Mohan R, Wang X, Jackson A, Bortfeld T, Boyer A L, Kutcher G J, Leibel S A, Fuks Z and Ling C C 1994 The
potential and limitations of the inverse radiotherapy technique Radiother. Oncol. 32 232–48
NEMA 1998 Digital Imaging Communications in Medicine version 3.1 (Washington, DC: National Electrical
Manufacturers Association)
Van Dyk J, Barnett R B, Cygler J E and Shragge P C 1993 Commissioning and quality assurance of treatment planning
computers Int. J. Radiat. Oncol. Biol. Phys. 26 261–73
Wang X, Spirou S, LoSasso T, Stein J, Chui C and Mohan R 1996 Dosimetric verification of intensity modulated
fields Med. Phys. 23 317–27
Webb S 1994 Optimizing the planning of intensity-modulated radiotherapy Phys. Med. Biol. 39 2229–46
Woo S Y, Grant W H, Bellezza D, Grossman R, Gildenberg P, Carpenter L S, Carol M and Butler E B 1996 A comparison
of intensity modulated conformal therapy with a conventional external beam stereotactic radiosurgery system
for the treatment of single and multiple intracranial lesions Int. J. Radiat. Oncol. Biol. Phys. 35 593–7
Xing L, Hamilton R J, Pelizzari C, Chen G T Y and Boyer A L 1998 Fast iterative algorithms for three-dimensional
inverse treatment planning Med. Phys. 25 1858–65
Xing L and Chen G T Y 1996 Iterative methods for inverse treatment planning Phys. Med. Biol. 41 2107–23
Yu C X, Symons M J, Du M N, Martinez A A and Wong J 1995 A method for implementing dynamic photon beam
intensity modulation using independent jaws and multileaf collimator Phys. Med. Biol. 40 769–87