Quality assurance for dynamic multileaf collimator modulated fields usinga fast beam imaging system

Lijun Ma, Paul B. Geis, and Arthur L. Boyera)

Division of Radiation Physics, Department of Radiation Oncology, Stanford University, Stanford,California 94305-5105

~Received 12 December 1996; accepted for publication 28 May 1997!

A quality assurance procedure was developed for x-ray beam intensity modulated conformal radio-therapy~IMCRT! using dynamic multileaf collimators~MLC!. The procedure verifies a prescribedintensity modulated x-ray beam pattern in the beam eye’s view~BEV! before the treatment proce-dure is applied to a patient. It verifies that~a! the leaf sequencing computer files were transferredcorrectly to the linac control computer;~b! the treatment can be correctly executed without machinefaults. A fast beam imaging system~BIS! consisting of a Gd2O2S scintillation screen, a charge-coupled device~CCD! camera, and a portable personal computer~Wellhofer Dosimetrie,Schwarzenbruck, Germany! was commissioned for this purpose. Measurements for the BIS perfor-mance are presented in this work. Reference images were derived from MLC leaf sequencing filesthat were used to drive a dynamic MLC system~Varian Oncology Systems, Palo Alto, CA!. Acorrelation method was developed to compare the BIS measurements with the calculated referenceimages. A correlation coefficient calculated using 26 correct intensity modulated fields was shownto be a reliable threshold to identify inaccurate treatment delivery files. The study has demonstratedthe feasibility of using the BIS and the correlation method to carry out on-line quality assurancetasks for IMCRT treatment fields in the BEV. ©1997 American Association of Physicists inMedicine.@S0094-2405~97!01708-2#

Key words: quality assurance, portal image, intensity modulation, conformal therapy, multileafcollimators

ingc

edrth

thoisa

th

mg

seice

taertatetodiho

thickateec-ut-llyg xthe

astionaf-ed

.

r

Aeylar

g ahe

I. INTRODUCTION

Intensity modulated conformal radiotherapy~IMCRT! usingdynamic multileaf collimators~MLC! aims to improve dosedistributions and increase local tumor control by conformseveral x-ray beams to a specific tumor shape. The teniques of using MLC in IMCRT have been actively studiand reviewed.1–5 To implement IMCRT on a Varian lineaaccelerator equipped with a MLC system, we adoptedcommonly used ‘‘sliding window’’ technique.6–8 The leafsequencing file produced by the leaf sequencing algoriwas saved by the MLC control computer to control the mtion of the MLC leaves during a treatment delivery. Itessential to verify a leaf sequence before an IMCRT trement procedure is applied to a patient.9,10 Our goal was todevelop a reliable and convenient procedure to ensurethe beam fluence patterns in the beam eye’s view~BEV!match those specified by an IMCRT plan. A fast beam iaging system~BIS! consisting of a metal/phosphor detectinscreen and a CCD camera was employed for this purpo

The BIS is a video-optical electronic portal image dev~EPID!. Various EPID devices have been reviewed.11 Formegavoltage x rays, theoretical modeling of the mephosphor system have been investigated by sevauthors.12–14 Advantages of using a metal/phosphor porimage device for IMCRT lies in its high data acquisition raeasy handling, high detecting efficiency, good signal-noise ratio, high image resolution, and its resistance to ration damage. The metal buildup layer that precedes the p

1213 Med. Phys. 24 (8), August 1997 0094-2405/97/24(8)

h-

e

m-

t-

at

-

.

l/all,-a-s-

phor screen serves as a signal enhancer. The metal isenough to allow incoming x rays to interact and to generCompton electrons within the layer. These Compton eltrons, in turn, produce the majority of the detected light oput from the phosphor. Since the light output monotonicacorresponds to the energy fluence carried by the incominrays, the energy fluence under BEV can be measured bysystem.

In the present work, an image correlation method wdeveloped for comparing the measured fluence distribuwith a reference fluence distribution calculated from the lesequencing file. A similar correlation method was employto compare digitally reconstructed radiographs~DRR! withelectronic portal images for detecting patient setup errors15

II. INSTRUMENTS AND TECHNIQUES

A. Wellho fer Beam Imaging System (BIS)

The schematic of the Wellho¨fer BIS device is shown inFig. 1. The device consists of a G2O2S fluorescent phosphoscreen preceded by a Cu screen.16 The Cu sheet is 1.0 mmthick. The G2O2S phosphor is 0.6 mm in thickness.charge-coupled device~CCD! video camera captures thphosphorescent beam image viewed through a 45° mmirror. The CCD camera has an array of 5123512 lightsensitive elements, which produce a digital image usincharge integration time adjustable from 120 ms to 1 s. Tcamera can view an image on the screen up to 30.1 cm330.7

1213/1213/8/$10.00 © 1997 Am. Assoc. Phys. Med.

ISnag

heitoneISranthe

tahsem

ag

nlhedhelyim

S-Ithesi

D.as-eri-be-nteachondr oflotnge-m-

alld10

, as

ingd

av-luelue,luegeedin

rrorncest-tionter-on

orldarry

1214 Ma, Geis, and Boyer: Dynamic multileaf collimator modulated fields 1214

cm. Through a two-way data line, the control unit of the Bis interfaced with a frame-grabber board within a persocomputer. The frame-grabber board is capable of acquirinfixed number of image frames, averaging them, and tgenerating a ten-bit beam image on the computer monThe total number of the summed image frames is betweeand 2000 and is user selectable prior to each measurem

To obtain a BEV image for an IMCRT treatment, the Bis fastened to the blocking tray holder of the linear acceletor. Once secured, the BIS rotates together with the gawithout detectable shifts in its position. We found that tBIS position is repeatable within 1.0 mm.

The measured BIS image was calibrated against a sdard background image to compensate for the image inmogeneity that may have been caused by nonuniformitiethe thickness of the Gd2O2S phosphor layer, warps on thmirror surface, or other imperfections in the imaging systeA calibrated image was also a background-subtracted imThe calibration formula for a measured image was

P~xi ,yj !5P0~xi ,yj !2Pb~xi ,yj !

r~xi ,yj !,

i 51, 2,..., 512,j 51, 2,..., 512, ~1!

where P(xi ,yj )is the calibrated pixel value at the positio(xi ,yj ) on the detection screen,P0 is the measured pixevalue,Pb is the background or dark current pixel value at tsame position (xi ,yj ), r is a calibration factor measurefrom a well-designed, uniform x-ray beam geometry. Tr(xi ,yj )value was provided by the manufacturer. Typicalthe P0 value was obtained by averaging more than 100age frames for a measurement.

B. Output and spatial linearity

Output linearity is a prerequisite for converting a BImeasured image into an x-ray beam fluence distribution.important for real-time monitoring of beam images in tBEV. Measurements of output response were made uexposures delivered by a linear accelerator~Clinac 2300

FIG. 1. A schematic of the Wellho¨fer Beam Imaging Device~BIS!. The dataline from the control unit interfaces via a frame-grabber board with a ptable personal computer~Dolch Computer Systems, Fremont, CA!.

Medical Physics, Vol. 24, No. 8, August 1997

lanr.1

nt.

-ry

n-o-in

.e.

,-

is

ng

C/D, Varian Oncology Systems, Palo Alto, CA! throughsquare and rectangular 6 MV x-ray fields at 100 cm SSThe linearity of the transmission ionization chamber andsociated electronics that controlled the exposures was vfied independently. Background images were measuredfore and after x-ray irradiation. The calibration dark-curreimage was the average of these two measurements. Forradiation field, an average pixel value within a fixed regiof interest ~ROI! was determined. The ROI was definearound the center of the field, such that the statistical errothe total counts within the region was less than 0.5%. A pof averaged pixel values within the ROI versus irradiatimonitor units is shown in Fig. 2. The result of a linear rgression fit is also shown. The errors for the fitting paraeters were evaluated at the 95% confidence level andfitted lines exhibitedr 2 . 0.99. From the slopes of the fittelines, the BIS output per monitor unit increases from a3 2 cm2 field to a 103 10 cm2 field. This indicates that theBIS output is directly related to the beam energy fluenceis typical for a metal/phospher system.13

To measure spatial linearity, standard field sizes rangfrom 1 3 1 cm2 to 253 25 cm2 at 100 cm SSD were measureusing the BIS. A small 0.63 0.6 cm2 ROI was defined sym-metrically around the center of a calibrated image. Theerage pixel value within the ROI was calculated. This vawas taken to be the 100% fluence value. Using this vafield edge positions corresponding to the 50% fluence vafor the entire field image were determined. From the edpositions, radiation field sizes in image pixels were plottagainst their nominal physical sizes. The result is givenFig. 3. The error bars on the data were computed from epropagation of the standard deviation of the 100% fluevalue. The error on the data was typically 0.5 pixel. A leasquare fit was performed on the data. The spatial resoluin the object plane of the BIS CCD camera lens was demined from the slope of the fitted line. The spatial resolutiwas determined to be 0.60260.014 mm for a horizontal di-

-FIG. 2. The Wellho¨fer BIS output response curves for two radiation fiesizes. The solid lines are the linear regression fits. The curves all cr 2.0.99.

efele

suinwrea

el

w

en

ept

waseond

estre-ard

e

t

n,

-of

00

nu-

Tati

1215 Ma, Geis, and Boyer: Dynamic multileaf collimator modulated fields 1215

rection, and 0.60060.011 mm for the vertical direction. Therrors corresponded to the 95% confidence level. The difence between the horizontal and vertical directions wasthan 0.5%.

C. BIS response deadtime

BIS response deadtime is the time lapse between twocessively captured images by the frame grabber. It is macaused by electronic switching processes associatedstoring and transmitting images. A correction for a largesponse deadtime would be necessary when converting impixel values to beam fluence values across a radiation fi

To measure the BIS deadtimet, a monitor unit, MU1, wasfirst selected for beam irradiation, such that

MU1[m DT,n DT, ~2!

where MU1 was the total monitor units,D was the averageddose rate in MU/min,T was the BIS integration time,n wasthe total frame counts for the BIS measurement, andm wasan arbitrary integer that satisfied Eq.~2!.

Using these values, an averaged pixel valueC8 for a 203 20 cm2 field image was measured. The measurementcarried out by starting the BIS frame grabberbeforethe ini-tiation of x-ray beam irradiation of the BIS phosphor scre

FIG. 3. The results of the BIS spatial resolution and field size response.solid lines are the linear regression fits. The vertical and horizontal spresolutions are determined from the slopes of the fitted lines.

Medical Physics, Vol. 24, No. 8, August 1997

r-ss

c-lyith-ged.

as

.

The total time for the BIS picture-capture process was klonger than the total beam-on time from Eq.~2!. Typically,m 5 0.75n was selected.

After the above measurement, a second measurementcarried out to acquireC9 by setting the total beam-on timmuch larger than the total BIS capture time. For the secmeasurement, the BIS frame grabber was startedafter theinitiation of the x-ray beam irradiation.

To carry out data analysis, one identical region of inter~ROI! was set within the two images from the two measuments. The average pixel value, together with its standdeviation, was calculated for the ROI.

From the first measurement, we obtained

C85FD@mT2~m21!t#/n5FDTS m

n2

m21

n

t

TD~n.m>2!, ~3!

whereT was the BIS integration time, i.e.,mT was the totalbeam-on time.C8 was the average pixel value within thROI, D was the dose rate for the beam,F was the dose topixel value conversion factor,n was the chosen frame counfor the measurement, andt was the BIS deadtime.

From the second measurement, we obtained

C95n8F DT/n85F DT, ~4!

whereC9 was the average pixel value andn8 was the BISframe count for the measurement. Combining Eqs.~3! and~4!, the BIS deadtimet was computed using

t5S m

n2

C8

C9D n

m21T ~n.m>2!. ~5!

The statistical error ont was calculated by error propagatioi.e.,

Dt5AS ]t

]C8D2

DC821S ]t

]C9D2

DC92

5nT

~m21!C9ADC821S C8

C9D 2

DC92 ~n.m>2!,

~6!

whereDC8 and DC9 were the standard deviations~s.d.! ofC8 and C9, respectively. By substitutingn, m, T, C8, andC9 values into Eqs.~5! and ~6!, the deadtime and the statistical error of the BIS were calculated. The measurementst were carried out at two dose rates of 400 MU/min and 6MU/min.

The measured deadtime for the BIS was 0.763.7 ms atthe dose rate of 600 MU/min, and 4.165.3 ms at the doserate of 400 MU/min. Therefore, within a 3s confidencelevel,t was smaller than 20 ms, which agrees with the mafacturer’s estimated value of zero.

heal

the

ld

talolurontena

e

inISeres

ldin

cal

hes of

e

eodhe

tedCe.

bp

ion

ofhe

one

1216 Ma, Geis, and Boyer: Dynamic multileaf collimator modulated fields 1216

III. METHODS

A. BIS image registration

To correlate a BIS image with a calculated image,center and the orientation for the reference coordinates wdetermined from a measured ‘‘cross-hair’’ radiation fieThe field pattern is presented in Fig. 4~a!. The photon jawswere used to form six narrow 0.53 20 cm2 fields centered a0, 11, and21 cm from the field center along the horizontand vertical directions. The accuracy of the digital contrused to set the jaws was verified by mechanical measments. The maxima points at the nine intersecting locatiof the six narrow fields were extracted from the integraBIS image. The reference center was calculated as the ceof the mass of the nine maxima positions. Once the BIS wattached to the blocking holder, positional errors for the rerence center were determined within60.6 mm in transla-tional shift and60.1° in rotational shift.

The MLC leaf width was measured in units of pixelsorder to determine each MLC leaf pair’s position in a Bimage. The field pattern in Fig. 4~b! was measured using thsame BIS setup. Leaf-edge positions separated by diffeleaf widths were determined within the field. The edge po

FIG. 4. The top part is the BIS-measured ‘‘cross-hair’’ pattern as formedsecondary collimators for determining the reference center. The bottomis the BIS-measured MLC field pattern for single leaf-width determinatwith the BIS attached to the gantry.

Medical Physics, Vol. 24, No. 8, August 1997

ere.

se-s

dters

f-

nti-

tions corresponded to the 50% pixel values of the fiemaxima. The results for the average leaf width was givenFig. 5. The leaf width was determined to be 12.4360.23pixels at 74.5 cm SSD, which agreed with the mechanicalibration of 1 cm at 100 cm SSD for the Varian MLCsystems.

To obtain the beam fluence distribution modulated by tMLC leaves, the BIS image was represented as a serievectorsCk(xi) at xi along the leaf motion direction, wherek denotes thekth leaf pair, andi represents thei th elementof theCk vector.Ck(xi) was determined by averaging threbeam profiles along the middle line of thekth leaf pair plustwo adjacent ones, i.e.,

Ck~xi !5 13@Ck~xj1d!1Ck~xi !1Ck~xj2d!#, ~7!

whereCk(xi)is the pixel value scanned along the middle linof the kth leaf pair, andd is the separation between twadjacent scans. Typically,d was selected to be 0.6 mm, anCk(xi) was normalized to the pixel maximum across timage.

B. Reference image calculation

Reference images for BIS measurements were calculafrom dynamic MLC leaf sequencing files that dictated MLleaf motion through the MLC controller computer interfacWe define a reference fluence vectorFk(xi) for the BIS mea-surements at a given positionxi alongkth leaf pair motion,i.e.,

Fk~xi !5E0.0

1.0

Fk~xi , f !d f , ~8!

whereFk(xi , f )was the calculated fluence at positionxi forthe kth leaf pair when cumulative-fraction-dose~CFD! val-ues specified in the leaf sequencing file changed fromf to

yart

FIG. 5. The determined leaf-width from the different leaf edge positionsthe MLC fields, as shown in the bottom part of Fig. 4. The solid line is tmean value. The region enclosed by two dashed lines represents thestandard deviation from the mean.

ee

o

-t

tityhe

n

eh

a

the

ter-

1217 Ma, Geis, and Boyer: Dynamic multileaf collimator modulated fields 1217

f 1 d f . The CFD value is taken to be proportional to thaccumulated monitor units, and increased from a normalizvalue of 0.0–1.0.

The Fk(xi , f ) in Eq. ~8! was modeled as the convolutionbetween a source distribution function in the target planS(x,y) and a collimator transmission functionT( f :x,y).16

The whole expression was corrected by the off-axis factOAR(x,y). It is given as

Fk~x, f !5OAR~x,yk!E E S~x8,y8!

3T~ f :x2x8,yk2y8!dx8 dy8. ~9!

In Eq. ~9!, the OAR(x,yk) was directly measured by theBIS using a large 403 40 cm2 MLC open field. It wasobtained by normalizing each pixel at position (x,yk) withthe pixel value of the reference center. The measured OA

FIG. 6. The multiparameter fitting result of the penumbra region of thBIS-measured MLC field. The profile is extracted from the scan along tmiddle line of a MLC leaf pair.

FIG. 7. A BIS-measured MLC open field profile compared with the calculated reference profile. The triangular symbol is the measurement datathe solid line is the calculated profile.

Medical Physics, Vol. 24, No. 8, August 1997

d

e

r

R

varied as much as 15% at the outer edge of the field.T( f :x 2 x8,yk 2 y8) in Eq. ~9! is the fluence transmis

sion function when CFD5 f in a leaf moving sequence. Iwas calculated as

T~ f :x2x8,yk2y8!5$H~ f :xL2x8!2H~ f :xR2x8!%

3T~yk2y8!, ~10!

whereH is the Heaviside step function,xL is the left-sideleaf-tip position of thekth MLC leaf pair,xR is the right-sideleaf-tip position of thekth leaf pair.T(yk 2 y8) 5 1, wheny8 falls within the projection of thekth leaf pair centered ayk and is zero otherwise. This is because MLC intensmodulation is realized along the horizontal leaf motion in tx direction. Over the verticaly direction, Fk(xi , f )is ap-proximated to be the same across the width of a leaf.

S(x8,y8) in Eq. ~9! was calculated as a linear combinatioof two Gaussian distributions,17 i.e.,

S~x8,y8!5A1 expS 2x821y82

2s12 D 1A2 expS 2

x821y82

2s22 D ,

~11!

e

-nd

FIG. 8. A typical BEV intensity modulated treatment field measured byBIS. The in-plane orx direction is along the MLC leaf or lower jaw motiondirection. The cross plane ory direction is along the upper jaw motiondirection.

TABLE I. The best-fit parameters for source distribution functions demined from the penumbra region of a 6 MV MLC open field.

s1 ~mm! s2 ~mm! A1 /A2

0.15 2.65 3.45

a

mn

-

aa.

f

ee

Se

s-

6se

e

r-

-

1218 Ma, Geis, and Boyer: Dynamic multileaf collimator modulated fields 1218

where A1 , A2 , s1 , and s2 are empirical parameters. Thefirst Gaussian represents the focal distribution of the x-rsource. The second Gaussian represents extrafocal scattecomponents from the source plus other contributions suchelectron transport uncertainties associated with the BIS iage processing. The empirical parameters were determifrom the minimized-x2 fitting of the penumbra region of aBIS-measured MLC open field using Eq.~9!. The fitting re-sults are shown in Fig. 6 for a penumbra region and Fig.for an open field profile. The best-fitting parameters are tablated in Table I.

C. Global correlation coefficient

A global correlation coefficientr was used to comparecalculated reference imageFk(xi) from Eq. ~7! with the BISmeasurementsCk(xi) from Eq. ~8!. The correlation coeffi-cient r is defined as

r 5( i ,k„Fk~xi !2F…„Ck~xi !2C…

A( i ,k„Fk~xi !2F…2A( i ,k„Ck~xi !2C…

2, ~12!

FIG. 9. The top of the figure is the isovalue contour of the BIS-measurimage shown in Fig. 7. The bottom part is the calculated reference imafrom the leaf sequencing file.

Medical Physics, Vol. 24, No. 8, August 1997

yringas-

ed

7u-

whereF is the average ofFk(xi) over i andk, andC is theaverage ofCk(xi) over all i andk. For a fixed valuek,Eq.~12! becomes

r k5( i„Fk~xi !2Fk…„Ck~xi !2Ck…

A( i„Fk~xi !2Fk…2A( i„Ck~xi !2Ck…

2. ~13!

The resultingr k represented the correlation between the reference image and the BIS measurements for thekth leaf pair.

Calculation of the correlation coefficientr was suited forour purpose because it tested the linear predictability ofBIS measurement from its reference image, and vice versIn practice, if the global correlation coefficientur u in Eq. ~12!or the minimum of theur ku in Eq. ~13! failed to satisfy apreset quality assurance value, the location of the faulty leamotion will be flagged to diagnose the cause.

IV. RESULTS

A typical BIS measurement for an intensity modulatedfield is shown in Fig. 8. The isovalue contours for the fieldand the calculated reference image are given in Fig. 9. Thfluence distribution patterns in the beam eye’s view werconveniently identified from the figures. The global correla-tion coefficient between the reference image and the BImeasurement was 97% for this case. It confirmed that thMLC control file for driving the leaf motion had been cor-rectly transferred and loaded into the linear accelerator’control unit. Moreover, the leaf sequencing program for generating the MLC control files had also functioned correctly.

The above correlation procedure was carried out for 2intensity modulated fields for different cancer sites such aprostate, head and neck, and cervix. A scatter plot of thglobal correlation coefficients is shown in Fig. 10~a!. For all

dge

FIG. 10. The top part~a! is the global correlation coefficientr for variousintensity modulated fields. The dot–dashed line indicates a 95% global corelation coefficient value. The bottom part~b! is the obtainedr value be-tween the BIS measurements and the mirror reflection of the reference images.

.

ac

-

y

e

c

y

co-om-

tedtsorsion0.5

edas

ref-d by, acu-re-ap-

ageork-site

toentso thent

di-

i,nataas

w-n-on-the

n,

al

n,o-

l.

--

ty-,’’

a

1219 Ma, Geis, and Boyer: Dynamic multileaf collimator modulated fields 1219

cases, the global correlation coefficients were above 95%To demonstrate the effectiveness of the procedure for d

tecting simple errors in leaf sequencing files, global correltion coefficients between BIS images and the mirror refletion of their reference images were calculated. The result wpresented together with the originalr value in Fig. 10~b!. Forall cases, global correlation coefficients were reduced to btween210% and215%. For each individual MLC leaf pair,the correlation coefficientsr k in Eq. ~13! was also consis-tently within 210% to215%.

To demonstrate the sensitivity of the procedure, randoerrors were intentionally introduced into the MLC leaf motion sequence. Thexi in Fk(xi , f ) of Eq. ~8! was replaced bya random variant following the Gaussian distribution whoscentroid wasxi .The magnitude of the error was governed bthe standard deviations of the Gaussian distribution. Theerror distribution for each leaf motion instance was generatsubject to conditions of continuous and valid leaf motionContinuous leaf motion forbade discontinuity in a leaf’stravel path. Valid leaf motion ensured no leaf collision.

The calculated global correlation coefficients as a funtion of s for various intensity modulated fields are given inFig. 11. With a thresholdr value of 92% in the figure,s wasscreened to less than 1 mm. Such a thresholdr value couldbe used to detect faults in a leaf sequencing file for the Qprocedure.

V. CONCLUSIONS AND DISCUSSION

A simple and convenient quality assurance procedure wdeveloped to verify intensity modulated conformal therapThe procedure employed a fast beam image system~BIS!capable of measuring and examining intensity modulatebeam fluence patterns in the beam eye’s view~BEV!. In theprocedure, a comparison was made between BIS measuments and reference images generated from the MLC le

FIG. 11. The global correlation coefficientsr as a function of randomlyintroduced leaf motion errors for intensity modulated fields. The horizontdashed line represents the 92%r value. The vertical dashed line indicatesthe s value of 1 mm.

Medical Physics, Vol. 24, No. 8, August 1997

e---

as

e-

m

e

d.

-

A

as.

d

re-af

sequencing files. A parametric test employing correlationefficients was developed and demonstrated to score the cparison results.

The procedure was carried out for 26 intensity modulafields. It was found that the global correlation coefficienwere above 95% for all cases. By introducing random errinto the reference image calculation, the global correlatcoefficient was able to detect uncertainties of less thanmm in the MLC leaf motion. A gross error such as a flippreference image from an erratic leaf sequencing file wreadily detected by the procedure.

Further improvements in the agreement between theerence image and the BIS measurements are anticipateincluding photon transport effects in the optical systemmore realistic MLC transmission function, and a more acrate extrafocal scattering model. A compromise between pcision and calculation speed was also considered in theproximations made in this work. Presently, a reference imtook, on average, 40 s to calculate on a Sun SPARC10 wstation. The whole procedure could be performed onwith other quality assurance tasks.

Since the global correlation technique was insensitivethe absolute normalization between the BIS measuremand the reference images, this technique can be applied tverification of dose distributions in an optically transparemedium that provides clinically equivalent scatter contions.

ACKNOWLEDGMENTS

The authors would like to thank Rob Hill, Shidong LC.-M. Ma, Edward Mok, Martin Murphy, and ChristiaWeinrank for useful discussions and their assistance in dacquisition and instrumentation processes. This work wsupported in part by Stanford University C.F. Aaron Endoment fund and Grant No. CA43840 from the National Cacer Institute. The contents of this work are solely the respsibility of the authors and do not necessarily representofficial views of the National Cancer Institute.

a!Electronic-mail: boyer@reyes.stanford.edu1A. L. Boyer, T. Bortfeld, D. Kahler, G. Starkshall, and T. Waldro‘‘Multileaf collimation for 3D conformal therapy,’’Proceedings of the3D Radiation Treatment Planning and Conformal Therapy InternationSymposium, St. Louis, MO~Medical Physics Publishing, Madison, 1993!.

2S. Webb,The Physics of Three-Dimensional Radiation Therapy~Instituteof Physics Publishing, Bristol and Philadelphia, 1993!.

3T. Bortfeld, A. L. Boyer, W. Schlegel, D. L. Kahler, and T. J. Waldro‘‘Realization and verification of three-dimensional conformal raditherapy with modulated fields,’’ Int. J. Radiat. Oncol. Biol. Phys.30,889–908~1994!.

4T. Bortfeld, D. L. Kahler, T. J. Waldron, and A. L. Boyer, ‘‘X-ray fieldcompensation with multileaf collimators,’’ Int. J. Radiat. Oncol. BioPhys.28, 723–730~1994!.

5A. L. Boyer, T. R. Bortfeld, L. Kahler, and T. J. Waldron, ‘‘MLC modulation of x-ray beams in discrete steps,’’Proceedings of the XIth Conference on the Use of Computers in Radiation Therapy, Manchester, En-gland ~IOP Publishing, London, 1994!.

6D. J. Convery and M. E. Rosenbloom ‘‘The generation of intensimodulated fields for conformal radiotherapy by dynamic collimationPhys. Med. Biol.37, 1359–1374~1992!.

l

by

a

T,

a

P.al

a-

lev,ge

rlotage

ydiat.

Ators

1220 Ma, Geis, and Boyer: Dynamic multileaf collimator modulated fields 1220

7S. V. Spirou and C. S. Chui, ‘‘Generation of arbitrary intensity profilesdynamic jaws or multileaf collimators,’’ Med. Phys.21, 1031–1041~1994!.

8P. Geis, A. L. Boyer, and N. H. Wells, ‘‘Use of multileaf collimator asdynamic missing-tissue compensator,’’ Med. Phys.23, 1199–1205~1996!.

9B. Fraass, ‘‘Quality assurance for 3D treatment planning,’’Proceedingsof the 1996 Summer School, Teletherapy: Present and Future, edited byR. Mackie and J. R. Palta~Advanced Medical Publishing, Madison1996!, pp. 253–318.

10X. Wang, S. Spirou, T. LoSasso, J. Stein, C. S. Chui, and R. Moh‘‘Dosimetric verification of intensity-modulated fields,’’ Med. Phys.23,317–327~1996!.

11A. L. Boyer, L. Antonuk, A. Fenster, M. Van Herk, H. Meertens,Munro, L. E. Reinstein, and J. Wong, ‘‘A review of electronic portimaging devices,’’ Med. Phys.19, 1–16~1992!.

Medical Physics, Vol. 24, No. 8, August 1997

.

n,

12W. Swindell, E. Morton, P. Evans, and D. Lewis, ‘‘The design of megvoltage imaging systems: Some theoretical aspects,’’ Med. Phys.18,855–866~1991!.

13T. Radcliffe, G. Barnea, B. Wowk, R. Rajapakshe, and S. Sha‘‘Monte Carlo optimization of metal/phosphor screens at megavoltaenergies,’’ Med. Phys.20, 1161–1169~1993!.

14D. A. Jaffray, J. J. Battista, A. Fenster, and P. Munro, ‘‘Monte Castudies of x-ray energy absorption and quantum noise in megavoltransmission radiography,’’ Med. Phys.22, 1077–1088~1995!.

15L. Dong and A. L. Boyer, ‘‘An image correlation procedure for digitallreconstructed radiographs and electronic portal images,’’ Int. J. RaOncol. Biol. Phys.33, 1053–1060~1995!.

16Wellhofer Dosimetrie, Inc.~private communication, 1996!.17M. B. Sharpe, D. A. Jaffray, and P. Munro, ‘‘Extrafocal radiation:

unified approach to the prediction of beam penumbra and output facfor megavoltage x-ray beams,’’ Med. Phys.22, 2065–2067~1995!.