P

MareptplramRb

OdE

cP

Int. J. Radiation Oncology Biol. Phys., Vol. 68, No. 3, pp. 903–911, 2007Copyright © 2007 Elsevier Inc.

Printed in the USA. All rights reserved0360-3016/07/$–see front matter

doi:10.1016/j.ijrobp.2007.02.033

HYSICS CONTRIBUTION

MAGNETIC RESONANCE–BASED TREATMENT PLANNING FOR PROSTATEINTENSITY-MODULATED RADIOTHERAPY: CREATION OF DIGITALLY

RECONSTRUCTED RADIOGRAPHS

LILI CHEN, PH.D., THAI-BINH NGUYEN, M.S., ÉLAN JONES, ZUOQUN CHEN, PH.D., WEI LUO, PH.D.,LU WANG, PH.D., ROBERT A. PRICE, JR., PH.D., ALAN POLLACK, PH.D., M.D., AND

C.-M. CHARLIE MA, PH.D.

Department of Radiation Oncology, Fox Chase Cancer Center, Philadelphia, PA

Purpose: To develop a technique to create magnetic resonance (MR)-based digitally reconstructed radiographs(DRR) for initial patient setup for routine clinical applications of MR-based treatment planning for prostateintensity-modulated radiotherapy.Methods and Materials: Twenty prostate cancer patients’ computed tomography (CT) and MR images were usedfor the study. Computed tomography and MR images were fused. The pelvic bony structures, including femoralheads, pubic rami, ischium, and ischial tuberosity, that are relevant for routine clinical patient setup weremanually contoured on axial MR images. The contoured bony structures were then assigned a bulk density of 2.0g/cm3. The MR-based DRRs were generated. The accuracy of the MR-based DDRs was quantitatively evaluatedby comparing MR-based DRRs with CT-based DRRs for these patients. For each patient, eight measuring pointson both coronal and sagittal DRRs were used for quantitative evaluation.Results: The maximum difference in the mean values of these measurement points was 1.3 � 1.6 mm, and themaximum difference in absolute positions was within 3 mm for the 20 patients investigated.Conclusions: Magnetic resonance–based DRRs are comparable to CT-based DRRs for prostate intensity-modulated radiotherapy and can be used for patient treatment setup when MR-based treatment planning isapplied clinically. © 2007 Elsevier Inc.

Radiotherapy, MRI treatment planning, Prostate cancer, MR-based DRRs.

aFdtbtkrtplprp

faAef

INTRODUCTION

agnetic resonance imaging (MRI) provides superior im-ge quality for soft tissue delineation over computed tomog-aphy (CT) and is widely used for target and organ delin-ation in radiotherapy for treatment planning (1–4). Therostate clinical target volume (CTV) seems to be overes-imated with CT images compared with MRI (3, 5). Therostate volume on CT seems to be approximately 40%arger than on MRI, as reported by Rasch et al. (5). Theseesults were consistent with those reported by Krempien etl. (6). Sannazzari et al. (3) also showed a mean overesti-ation of CTV of 34% with CT compared with MRI.esults showed that radiation doses received by the penileulb and corporal bodies may cause erectile dysfunction

Reprint requests to: Lili Chen, Ph.D., Department of Radiationncology, Fox Chase Cancer Center, 333 Cottman Ave., Phila-elphia, PA 19111. Tel: (215) 728-3003; Fax: (215) 728-4789;-mail: lili.chen@fccc.eduPresented in part at the Annual Meeting of the American Asso-

iation of Physicists in Medicine, July 25–29, 2004, Pittsburgh,A.

Conflict of interest: none. A

903

fter prostate treatment (7). Studies are being performed atox Chase Cancer Center (FCCC) to use MRI to limit theose to the erectile tissue with intensity-modulated radiationherapy (IMRT). A Phase III randomized trial at FCCC iseing conducted to investigate the clinical significance ofhe erectile tissue–sparing technique. Recently, Steenbak-ers et al. (8) also reported that the dose delivered to theectal wall and bulb of the penis was significantly reduced inreatment plans based on MRI-delineated prostates com-ared with CT-delineated prostates, allowing a dose esca-ation of 2.0–7.0 Gy for the same rectal wall dose. Therescribed dose to the planning target volume could beaised from 78 to 85 Gy when using the MRI-delineatedrostate for treatment planning.Recent studies suggest that dose escalation with three-

Supported in part by grants from the U.S. Department of De-ense (PC030800), the National Institutes of Health (CA78331),nd the Howard Hughes Medical Institute.cknowledgments—The authors thank David Abraham for hisxcellent technical assistance; and Dr. Gerald E. Hank for hisoresight in initiating MRI simulation at Fox Chase Cancer Center.

Received Dec 1, 2006 and in revised form March 6, 2007.

ccepted for publication March 7, 2007.

dttbdcbampetsmidedtidbptw

I

Moae3aad

apdCdtbitpfwiws

C

MpvtwiapaCteaa

stbtss

view o

904 I. J. Radiation Oncology ● Biology ● Physics Volume 68, Number 3, 2007

imensional conformal radiotherapy or IMRT can poten-ially improve local control and toxicity (9–18). At presenthe gold standard for prostate treatment planning is to useoth CT and MR imaging to take advantage of the electronensity information with CT and the superior soft tissueapabilities with MRI. Ideally, treatment planning would beased solely on a clinically useful diagnostic MR study. Thedvantages of MR-based treatment planning are (1) to re-ove potential errors due to image fusion, (2) to reduce

atient cost by avoiding redundant CT scans, and (3) toliminate unnecessary radiation exposure to the patient ando save patient, staff, and machine time. However, there areeveral challenges to directly using MR images for treat-ent planning, including (1) the lack of electron density

nformation that is needed for heterogeneity corrections inose calculation, (2) image distortion that affects patientxternal contour determination and therefore introducesose calculation uncertainty, and (3) lack of bony structureso derive effective DRRs for patient setup. The first twossues (the lack of electron density information and imageistortions in MR-based prostate treatment planning) haveeen discussed in great detail by Chen et al. (19). Theurpose of the present study was to investigate a techniqueo create MR-based DRRs for prostate IMRT patient setuphen MR-based treatment planning is applied clinically.

METHODS AND MATERIALS

mage collectionMagnetic resonance images were obtained from a 0.23-T openR scanner (Philips Medical Systems, Cleveland, OH). A series

f 48 axial slices (3-mm thickness) covering the whole pelvisccording to the guidance image were acquired using turbo spincho, three-dimensional sequence, time to repetition/time to echo000/140 ms, field of view 45–50 cm (depending on patients’natomic dimensions), matrix 256 � 256, echo train length 32, flipngle 90°, number of excitations 1, bandwidth 39.5 kHz, frequency

Fig. 1. Digitally reconstructed radiographs from (a) magnfigures are coronal view and bottom figures are sagittal

irection horizontal, and 9-min scan time. The significant motion t

rtifacts were not observed from MRI. The MR images wereost-processed for image distortion correction using the gradientistortion correction (GDC) software provided by the vendor.omputed tomography–magnetic resonance fusion with CT-basedose calculation has been a routine procedure for prostate IMRTreatment planning at FCCC since 2001. Computed tomography–ased DRRs together with BAT (B-mode acquisition and target-ng; NOMOS, Sewickley, PA) ultrasound or the CT-on-rails sys-em (Siemens Medical Solutions, Concord, CA) are used for initialatient setup and target localization. On-line images, obtainedrom either the CT-on-rails system or the BAT ultrasound systemere used for daily target localization to correct for prostate

nterfraction motion. Twenty patients with fused CT-MR imagesere randomly selected from our patient database and used in this

tudy.

alibration of bony structures on MRIFigure 1 shows an example of the DRR directly created fromR images. It is clearly demonstrated that the DRRs do not

rovide enough bony structure information for patient positioningerification. Therefore, it is necessary to develop a technique forhe creation of useful MR-based DRRs for initial treatment setuphen CT is not involved. Clinical bony landmarks such as pubis,

schium, and ischial tuberosity on CT-based DRRs are used todjust the patient treatment position by comparing with anterior–osterior (AP) and lateral portal images. Our goal was to developtechnique to create MR-based DRRs that are comparable to

T-based DRRs for the clinical implementation of MR-basedreatment planning for prostate IMRT. Our approach was to gen-rate MR-based DRRs by contouring the relevant bony structuresnd assign them with a bulk density. We performed this study onn AcQsim system (Philips Medical System, Cleveland, OH).

On T2-weighted MR images, bones appear hyperintense (white),urrounded by narrow hypointense (dark) regions. The marrow insidehe bone or soft bones shows high intensity (white), whereas compactone shows lower intensity (dark). Therefore, the boundary betweenhe bone and the soft tissue is not always distinguishable on MRI. Totudy the correlation between CT number (electron density) and MRignal intensity, CT and MR images were fused manually according

sonance (MR) and (b) computed tomography (CT). Topf the pelvic region.

etic re

o bony anatomy using the interactive method available on the Ac-

QiibctetM

G

g

patctscoepr

c(waiamrTdpbd“

Fbioa

905MR-based DRRs for prostate cancer ● L. CHEN et al.

sim system. Computed tomography images were loaded as primarymages, and MR images were loaded as secondary images. The twomages were shown side by side on the same computer screen. Theones were contoured on the basis of the CT images while theontours were also shown simultaneously on MRI. The calibration ofhe bones on MRI that are relevant to prostate treatment setups wasstablished according to the CT images. On the basis of this calibra-ion, the boundary of each relevant bone can be easily determined on

RI alone.

uidelines for bone contouringThe technique that we describe here is intended to serve as a

uide for the clinical implementation of MR-based treatment

ig. 2. Delineation of bony structures for magnetic resonance-ased digitally reconstructed radiographs showing contours start-ng at the level of (a) the femoral head to (b) the lesser trochanterf the femur. The pelvic bones are divided into three groupsccording to the similarity of the bone shapes.

Fig. 3. Correlations of the relevant bone contours drawn

Detailed descriptions are given in the text.

lanning for prostate cancer. We outline contours of the bonest the level of the top of the femoral head (a) to the lesserrochanter of the femur (b), as shown in Fig. 2. To describe ourontour technique clearly we categorize the pelvic bones intohree groups according to the similarity of the bone shape ashown in Fig. 2. Figure 3 shows the correlation of the bonyontours drawn only on the axial MR image and contour linesn the coronal and sagittal MR images. Critical points forach group are determined according to the relation of theoints on the DRR that are used for checking patient positionsoutinely.

For convenience, we have defined a few technical terms for ourontouring technique, including white side (strict), white sidenonstrict), dark side (strict), and dark side (nonstrict). The termhite indicates the hyperintense region on the T2-weighted image,nd dark indicates the hypointense region on the T2-weightedmage. Thus, white side (strict) means to draw a contour immedi-tely outside the hyperintense region, and white side (nonstrict)eans the contour is drawn slightly away from the hyperintense

egion (approximately 2 mm per pixel away), as shown in Fig. 4.he same concept can be applied to the terms dark side (strict) andark side (nonstrict). As mentioned previously we divided theelvic region into three groups. For each group the left and rightones are mirror images on the axial view. Therefore we onlyescribe our method for the right-side bones. To start contouring,virtual fluoroscopy” is selected on the tool menu, which displays

axial and coronal views and (b) axial and sagittal views.

on (a)

tds

C

w“c

aibstFwo

C(

sPPchDhhtgb

C

dTtg

C

g

Fwptm

F(r

906 I. J. Radiation Oncology ● Biology ● Physics Volume 68, Number 3, 2007

he three-dimensional views of the MR images. The contours arerawn on the axial images slice by slice; the contours will showimultaneously on both coronal and sagittal views.

ontouring bony structures for the first group (Region 1)The points of interest (POIs) here are “A” and “B” (Fig. 5),

hich will determine points “a” and “b” on the coronal DRR anda” on the sagittal DRR (see Fig. 9). Point “C” was found notritical for patient setup and appears as “b” on the sagittal DRR.

ig. 4. Demonstration of the contouring technique. (a) Contourhite side (strict) means contouring immediately outside the hy-erintense region; (b) contour white side (nonstrict) means con-ouring slightly away from the hyperintense region (by approxi-ately 1 pixel).

To contour POI A, one can imagine that there are two lines “a” t

nd “b” on two sides of the peak. The intersection of “a” and “b”s the POI A. Lines “a” and “b” are determined according to theoundary between hyper- and hypointense regions using the whiteide (nonstrict) technique. Point of interest B is the intersection ofhe straight lines between black side and white side, as shown inig. 6. It should be mentioned that the exact locations of the POIsill affect the accuracy of the resulting DRR, whereas the volumesf the contoured structures will not.

ontouring bony structures for the second groupRegion 2)

There are two relevant bones in Region 2, including pubicymphysis and ischial tuberosity. As shown in Fig. 7, POI A andOI C are critical for the positioning check clinically, whereasOIs B, E, and D are not critical. For point A the vertical line isontoured with white side (strict). Attention should be paid to theorizontal line here because its height will appear on the sagittalRR that will be used to judge the patient AP position. Theorizontal line is contoured with white side (strict). We draw theorizontal line starting from where the white (hyperintense) beginso blur. For POI C we use the same method as for POI B of the firstroup. For POIs B, E, and D we contour with white side (nonstrict)ecause they are not critical to patient setups.

ontouring bony structures for the third group (Region 3)For the third group, only POI A is the critical point. The way to

etermine this point is similar to that for POI A for the first group.his means that POI A is the intersection of two lines: one is on

op (white side), and the other is the dark side (nonstrict) of theap, as shown in Fig. 8.

reation of MR-Based DRRsAfter contouring the three bone groups, proper names were

iven to these bones in the “organ selection” menu (e.g., symphy-

ig. 5. An example from the first bone group. Point of interestPOI) A is correlated to both the coronal and sagittal digitallyeconstructed radiographs (DRRs), whereas POI B is correlated to

he coronal DRR and POI C is correlated to the sagittal DRR only.

sigrcdt

D

tfsiscavt

Fos

907MR-based DRRs for prostate cancer ● L. CHEN et al.

is). A bulk electron density was assigned to these bones. Accord-ng to our experience, an electron density between 1.8 and 2.0/mm3 is adequate. After assigning electron densities to all theelevant bones, the DRRs were generated. The quality of theoronal and lateral DRRs can be improved by adjusting the win-ow and level parameters. Figure 9 shows the MR-based DRR andhe CT-based DRR for 1 of the 20 patients investigated.

Fig. 6. An example to delineate point of interest (POI)“a” and “b.” (b) The volume defined by the contours awreconstructed radiograph.

tructed radiographs. POI � point of interest.

Fts

ata analysisPaired MR-based and CT-based DRRs were compared quanti-

atively for all the patients investigated. Contoured bony structuresrom MRI were superimposed on CT-based DRRs using Photo-hop software (Adobe Systems, San Jose, CA) with the samesocenter (Fig. 10). The eight points (relevant to patient treatmentetup) from both coronal and sagittal DRRs were used for theomparison. The difference for each position point between CTnd MRI was calculated according to the pixel value. Negativealues mean the position points on MR-based DRRs were insidehe outlines of the corresponding bony structures on the CT-based

POI B. (a) They are the intersection points of the linesm the POIs will not affect the accuracy of the digitally

A anday fro

ig. 7. An example of the second bone group and the correlationf the axial contours with the coronal and sagittal digitally recon-

ig. 8. An example of the third bone group and the correlation ofhe axial contours with the coronal and sagittal digitally recon-

tructed radiographs. POI � point of interest.

DpbeTm

s

Mtgpt(doF

908 I. J. Radiation Oncology ● Biology ● Physics Volume 68, Number 3, 2007

RRs, whereas positive values mean the MR-determined positionoints were outside the outlines of bony structures on the CT-ased DRRs (Fig. 10). The mean and 1 standard deviation of therror for each measurement point were calculated for 20 patients.he maximum differences between CT and MR for each measure-ent point were also calculated.

RESULTS

We investigated MR-based DRRs using 20 randomlyelected prostate cancer patients. Figure 9 demonstrates that

Fig. 9. An example of (a) magnetic resonance (MR)-basetomography (CT)-based DRRs. The MR-based DRRs wassigned to a bulk density of 2.0 g/cm3.

Fig. 10. Measurement positions on coronal and sagittaaccuracy of magnetic resonance (MR)-based DRRs (char4 are used for the left–right position setup, points 1 and

and points 7 and 8 are used for the anterior–posterior position

R-based DRRs are comparable to CT-based DRRs andhat it is reasonable to use a bulk electron density of 2.0/cm3 for the DRR creation. In fact, one can also select torint MR-based DRRs with the outlined bony contours onop of the gray-scale image. This way, the contour linesinstead of the gray-scale image) on the DRRs can beirectly used to compare the bony structures on a portal filmr an electronic portal imaging device (EPID) image (seeig. 10).The differences in the position points between MR-based

lly reconstructed radiographs (DRRs) and (b) computederated after the relevant bones had been contoured and

ally reconstructed radiographs (DRRs) to evaluate theed by the MR-based outlines/contour lines). Points 3 andand 6 are used for the inferior–superior position setup,

d digitaere gen

l digitacteriz5 or 2

setup.

ariwaefmtTpw2a8As

fupsvfpib

bg

MM

d

M uctures

BPPPP

BPPPP

Ias

pb

909MR-based DRRs for prostate cancer ● L. CHEN et al.

nd CT-based orthogonal DRRs for 20 patients are summa-ized in Table 1. The contours were generated by threenvestigators. Because the patients’ CT and MR imagesere directly taken from our clinical database and were

lready fused for treatment planning, the positional differ-nces shown here include the positional errors due to theusion process, which is estimated to be approximately 2m, and the interobserver variations. The agreement be-

ween MR-based DRRs and CT-based DRRs was excellent.he mean difference was within 0.8 � 1.5 mm except foroint 7 (1.3 � 1.6 mm), and the maximum difference wasithin 3.0 mm for all 180 position points evaluated for the0 patients except for point 7 of Patient 19 (3.6 mm). Fromclinical point of view, point 7 is less important than point(pubis), which is used for the determination of the patientP positions. These results demonstrate that there is no

ystematic biasing in our contouring method.Table 2 shows the mean shift values of the 38 treatment

ractions for the first 4 prostate IMRT patients plannedsing MRI alone and 4 other prostate IMRT patientslanned using CT alone, being treated at the same time. Thehift values in the three major axes were all within expectedalues. Occasionally larger shifts (�1 cm) were observedor both CT- and MRI-based planning when large gasockets occurred in the rectum. The combined target local-zation accuracy for routine prostate treatment is expected to

Table 1. Distance (in millimeters) of measured posit

Patient no. 1 2 3

1 1.3 �0.9 �1.72 �1.0 �1.3 0.43 0.0 �0.9 0.44 0.4 0.4 �2.25 0.4 0.4 0.96 1.3 0.9 �1.77 0.9 �1.3 1.38 0.4 0.4 �0.99 0.4 �0.9 0.4

10 2.2 �0.9 2.211 1.3 2.2 1.712 0.4 0.9 0.913 0.9 �0.9 �0.914 0.4 0.4 �2.215 �2.0 1.7 1.716 0.4 1.6 2.017 �1.2 �1.6 �2.018 �2.0 �1.6 �0.819 �1.6 �2.4 �1.220 �1.2 �0.8 1.2

ean � SD 0.1 � 1.2 �0.2 � 1.3 �0.0 � 1.5 �0ax. diff. 2.2 2.4 2.2

Abbreviations: MR � magnetic resonance; DRR � digitally recifference.

The eight points of measurement are illustrated in Fig. 10. ThR-based DRRs were inside and outside the outlines of bony str

e better than 5 mm at a 95% confidence level. s

We also investigated the time required to generate MR-ased DRRs. Although manual contouring is needed toenerate MR-based DRRs, the time required to contour all

ints between MR-based DRRs and CT-based DRRs

of measurement

5 6 7 8

�1.7 �1.0 �2.2 �2.2�1.7 �2.0 0.9 0.4�0.4 0.4 �1.3 �1.3�0.9 �1.0 0.9 �1.7�0.9 2.2 0.4 �0.9�2.6 �3.0 0.9 0.9

0.4 0.4 3.0 3.0�1.3 0.9 1.3 1.3�1.7 �2.0 0.4 1.7�2.2 �1.0 1.3 0.0

0.9 0.9 2.6 0.0�2.2 �2.2 2.2 1.7

1.7 �1.3 0.9 �1.3�1.3 2.2 3.0 0.4

2.2 1.7 2.2 �1.72.0 2.8 2.4 1.2

�1.6 �1.6 2.8 �2.0�2.4 �1.6 �1.2 �2.8�1.2 �0.8 3.6 2.0�1.6 �0.4 1.6 0.8

.2 �0.8 � 1.5 �0.3 � 1.6 1.3 � 1.6 �0.0 � 1.62.6 3.0 3.0 3.0

ted radiograph; SD � standard deviation; Max. diff � maximum

tive and positive values indicate whether the position points onon the CT-based DRRs, respectively.

Table 2. Mean (� standard deviation) shift values along thethree major axes based on the BAT ultrasound target

localization procedures for 38 IMRT treatments

Patient no.

Shifts (cm)

RL AP SI

ased on MR simulationatient 1 �0.19 � 0.25 0.07 � 0.39 0.51 � 0.43atient 2 �0.20 � 0.32 0.31 � 0.53 0.10 � 0.48atient 3 0.15 � 0.25 0.39 � 0.23 0.78 � 0.23atient 4 0.12 � 0.29 0.53 � 0.29 0.14 � 0.29

ased on CT simulationatient 1 0.09 � 0.42 �0.33 � 0.30 �0.03 � 0.34atient 2 0.28 � 0.25 0.03 � 0.42 0.48 � 0.41atient 3 �0.01 � 0.25 �0.06 � 0.27 0.32 � 0.32atient 4 0.05 � 0.24 0.04 � 0.17 0.40 � 0.40

Abbreviations: BAT � B-mode acquisition and targeting;MRT � intensity-modulated radiotherapy; RL � right–left; AP �nterior–posterior; SI � superior–inferior; DRR � digitally recon-tructed radiograph.

The shift values were calculated as the differences between thelanned target locations and the actual target locations measuredy the BAT system prior to each treatment. The initial treatment

ion po

Points

4

�1.3�2.2�1.3

0.9�1.3�0.9�1.7

0.4�0.4�1.3�1.3�1.3�1.3

0.91.3

�0.81.6

�1.6�2.4

1.2.6 � 1

2.4

onstruc

e nega

etups were using either CT-based DRRs or MR-based DRRs.

rTbcidhD

MbnvfiMbgbBpr

pabsFbsmacc

ws5aDspTi“

epubrMftptEmTp

DfpDastt

910 I. J. Radiation Oncology ● Biology ● Physics Volume 68, Number 3, 2007

elevant bones for a patient was approximately 10–15 min.he time to assign a bulk electron density to the relevantony structures was minimal. In fact, it is not necessary toontour the femoral head because the pubis (symphysis),schium, and ischial tuberosity positions are sufficient toetermine the patient positions. We have contoured femoraleads to provide redundant checks and for more realisticRRs.

DISCUSSION

In this work we have developed a method to generateR-based DRRs for the clinical implementation of MRI-

ased treatment planning for prostate IMRT. Magnetic reso-ance–based DRRs can be used for initial patient positionerification by comparing with images taken using portallm or an EPID image. In our clinical implementation ofRI-based treatment planning for prostate IMRT, MRI-

ased DRRs are used during initial treatment setups to-ether with either a BAT or a CT-on-rails system to set theaseline treatment positions and later as a backup for theAT or CT-on-rails systems if the systems are down or if aatient cannot be set up using BAT/CT-on-rails for variouseasons.

As can be seen clearly in Fig. 10, all the bones (andosition points) on the coronal DRR have mirror images,nd therefore their relative positions, which may be affectedy the contouring method used, will have little effects on theetup accuracy as long as they are contoured consistently.or example, if the symphysis is contoured consistentlyetween the left side and the right side, the midline of theymphysis (i.e., the left–right position of the patient) deter-ined by points 3 and 4 will be accurate even though the

bsolute positions of points 3 and 4 may be affected by theontouring method or by interobserver variations. The same

an be said about the patient superior–inferior position, n

REFEREN

6. Krempien RC, Schubert K, Zierhut D, et al. Open low-field

1

1

hich can be accurately determined by bony landmarksuch as pubis and ischium through paired points such as (1,) and (2, 6). The patient AP position is mainly determinedccording to the pubis position (i.e., point 8 on the sagittalRR). Again, the relative position determined by a bony

tructure is expected to be more accurate than the absoluteositions of the contour points defining the bony structure.he same phenomenon can be observed with gray-scale

mages, such as CT and MRI, when different “window” andlevel” parameters are used.

On the basis of this analysis and our results in Table 1, westimate that the accuracy of our MR-based DRRs is ap-roximately 3 mm or better. This does not include thencertainties introduced by the CT-MRI fusion processecause they will be eliminated when MRI alone is used foradiotherapy treatment planning. The 3-mm accuracy for

R-based DRRs is considered adequate for this purpose. Inact, for routine prostate IMRT treatment at FCCC, we usehe BAT ultrasound system or the CT-on-rails system as arimary means to relocate the soft tissue target before eachreatment. The initial patient setup based on portal film orPID only serves as a secondary means to place the treat-ent target within 0.5–0.7 cm of their treatment positions.his will facilitate the ultrasound/CT target localizationrocess in the subsequent daily treatments.In summary, we have explored the use of MR-based

RRs for initial radiotherapy treatment setup verificationor prostate IMRT by comparing with images taken usingortal film or EPID. Our results showed that MR-basedRRs using the outlines of relevant bony structures have an

ccuracy of approximately 3 mm, adequate for initial patientetup. This technique has been used, in combination withhe BAT/in-room CT daily target localization technique, forhe clinical implementation of MRI-based treatment plan-

ing for prostate IMRT at FCCC since November 2003.

CES

1. Khoo VS, Adams EJ, Saran F, et al. A Comparison of clinicaltarget volumes determined by CT and MRI for the radiother-apy planning of base of skull meningiomas. Int J Radiat OncolBiol Phys 2000;46:1309–1317.

2. Potter R, Heil B, Schneider L, et al. Sagittal and coronalplanes from MRI for treatment planning in tumors of brain,head and neck: MRI assisted simulation. Radiother Oncol1992;23:127–130.

3. Sannazzari GL, Ragona R, Ruo Redda MG, et al. CT-MRIimage fusion for delineation of volumes in three-dimensionalconformal radiation therapy in the treatment of localized pros-tate cancer. Br J Radiol 2002;75:603–607.

4. Tanner SF, Finnigan DJ, Khoo VS, et al. Radiotherapy planningof the pelvis using distortion corrected MR images: The removalof system distortions Phys Med Biol 2000;45:2117–2132.

5. Rasch C, Barillot I, Remeijer P, et al. Definition of the prostatein CT and MRI: A multi-observer study. Int J Radiat OncolBiol Phys 1999;43:57–66.

magnetic resonance imaging in radiation therapy treatmentplanning. Int J Radiat Oncol Biol Phys 2002;53:1350–1360.

7. Buyyounouski MK, Horwitz EM, Uzzo RG, et al. The radia-tion doses to erectile tissues defined with magnetic resonanceimaging after intensity-modulated radiation therapy or iodine-125 brachytherapy. Int J Radiat Oncol Biol Phys 2004;59:1383–1391.

8. Steenbakkers RJ, Deurloo KE, Nowak PJ, et al. Reduction ofdose delivered to the rectum and bulb of the penis using MRIdelineation for radiotherapy of the prostate. Int J Radiat OncolBiol Phys. 2003;57:1269–1279.

9. Am AM, Mott J, Mackay RI, et al. Prediction of the benefitsfrom dose-escalated hypofractionated intensity-modulated ra-diotherapy for prostate cancer. Int J Radiat Oncol Biol Phys2003;56:199–207.

0. Hanks GE. Progress in 3D conformal radiation treatment ofprostate cancer. Acta Oncol 1999;38(Suppl. 13):69–74.

1. Hanks GE, Hanlon AL, Schultheiss TE, et al. Dose escalation

with 3D conformal treatment: Five year outcomes, treatment

1

1

1

1

1

1

1

1

911MR-based DRRs for prostate cancer ● L. CHEN et al.

optimization, and future directions. Int J Radiat Oncol BiolPhys 1998;41:501–510.

2. Hunt MA, Zelefsky MJ, Wolden S, et al. Treatment planningand delivery of intensity-modulated radiation therapy for pri-mary nasopharynx cancer. Int J Radiat Oncol Biol Phys 2001;49:623–632.

3. Pollack A. Preliminary results of a randomized radiotherapydose-escalation study comparing 70 Gy with 78 Gy for pros-tate cancer. J Clin Oncol 2000;23:3304–3911.

4. Pollack A, Zagars GK, Rosen II. Prostate cancer treatmentwith radiotherapy: Maturing methods that minimize morbid-ity. Semin Oncol 1999;26:150–161.

5. Pollack A, Zagars GK, Starkschall G, et al. Prostate cancerradiation dose response: Results of the M. D. Anderson phaseIII randomized trial. Int J Radiat Oncol Biol Phys 2002;53:

1097–1105.

6. Wachter S, Wachter-Gerstner N, Bock T, et al. Interobservercomparison of CT and MRI-based prostate apex definition.Clinical relevance for conformal radiotherapy treatment plan-ning. Strahlenther Onkol 2002;178:263–268.

7. Yeoh EE, Fraser RJ, McGowan RE, et al. Evidence forefficacy without increased toxicity of hypofractionated ra-diotherapy for prostate carcinoma: Early results of a PhaseIII randomized trial. Int J Radiat Oncol Biol Phys 2003;55:943–955.

8. Zelefsky MJ, Leibel SA, Gaudin PB, et al. Dose escalationwith three-dimensional conformal radiation therapy affects theoutcome in prostate cancer. Int J Radiat Oncol Biol Phys1998;41:491–500.

9. Chen L, Price RA Jr, Wang L, et al. MRI-based treatmentplanning for radiotherapy: Dosimetric verification for prostate

IMRT. Int J Radiat Oncol Biol Phys 2004;60:636–647.