P

Tollitripht

m

S5P

faAy

Int. J. Radiation Oncology Biol. Phys., Vol. 65, No. 5, pp. 1579–1584, 2006Copyright © 2006 Elsevier Inc.

Printed in the USA. All rights reserved0360-3016/06/$–see front matter

doi:10.1016/j.ijrobp.2006.04.038

HYSICS CONTRIBUTION

GEOMETRIC ACCURACY OF A REAL-TIME TARGET TRACKING SYSTEMWITH DYNAMIC MULTILEAF COLLIMATOR TRACKING SYSTEM

PAUL J. KEALL, PH.D.,* HERBERT CATTELL, B.E.,† DAMODAR POKHREL, M.S.,*SONJA DIETERICH, PH.D.,‡ KENNETH H. WONG, PH.D.,§ MARTIN J. MURPHY, PH.D.,*

S. SASTRY VEDAM, PH.D.,*� KRISHNI WIJESOORIYA, PH.D.,* AND RADHE MOHAN, PH.D.�

*Department of Radiation Oncology, Virginia Commonwealth University, Richmond, VA; †Varian Medical Systems,Palo Alto, CA; ‡Department of Radiation Medicine, Georgetown University Hospital, Washington, DC; §Department

of Radiology, Georgetown University Hospital, Washington, DC; and �Department of Radiation Physics,The University of Texas M.D. Anderson Cancer Center, Houston, TX

Purpose: Dynamically compensating for target motion during radiotherapy will increase treatment accuracy. Alaboratory system for real-time target tracking with a dynamic MLC has been developed. In this study, thegeometric accuracy limits of this DMLC target tracking system were evaluated.Methods and Materials: A motion simulator was programmed to follow patient-derived tumor motion paths,parallel to the leaf motion direction. A target attached to the simulator was optically tracked, and the leafpositions adjusted to continually align the DMLC beam aperture to the target. Analysis of the tracking accuracywas based on video images of the target and beam alignment. The system response time was determined and thetracking error measured. Response time–corrected tracking accuracy was also calculated to investigate theaccuracy limits of an improved system.Results: The response time of the system is 160 � 2 ms. The geometric precision for tracking patient motion is0.6 to 1.1 mm (1�) for the 3 patient datasets tested, with tracking errors relative to the original patient motionof 35, 40, and 100%.Conclusions: A DMLC target tracking system has been developed that can account for detected motion parallelto the leaf motion direction. The tracking error has a negligible systematic component. Reducing the responsetime will further increase the overall system accuracy. © 2006 Elsevier Inc.

Tumor tracking, Geometric accuracy, Dynamic multileaf collimator, Dynamic motion compensation.

p((

hmct

D

tw

taWsMt

A

INTRODUCTION

he multileaf collimator (MLC) is a widely available technol-gy for radiation therapy delivery. The technology to adjusteaf positions during beam delivery based on a predeterminedeaf sequence facilitating intensity modulated radiation therapys also quite mature (1). Another obvious application, thoughechnically challenging, is to use the MLC to continuouslyealign the radiation beam to the target during therapy, dynam-cally compensating for any detected target motion. Proof-of-rinciple studies of this approach have been performed, (2–6)owever to date no studies have reported on the implementa-ion of dynamic MLC (DMLC) target tracking.

Other options for aligning the radiation beam with aoving target during radiotherapy with dynamic motion com-

Reprint requests to: Paul J. Keall, Ph.D., Radiation Oncology,tanford University, 875 Blake Wilbur Dr., Stanford, CA 94305-847. Tel: (650) 723-5549; Fax: (650) 498-5008; E-mail:aul.Keall@stanford.eduSupported by Grant Nos. R01CA93626 and R21CA119143

rom the National Institutes of Health and by a sponsored researchgreement between Varian Medical Systems and VCU.cknowledgments—Dorin Todor helped to write the image anal-

sis software. Hassan Mostafavi and Sergey Povzner gave

1579

ensation are robotic control of the linear accelerator (7, 8)clinically available), block motion (9), and couch motion10) (investigated, though not clinically available).

A system for real-time tracking of targets with DMLCas been developed in a laboratory setting at Virginia Com-onwealth University. The aim of the current work was to

haracterize the limits of geometric accuracy of a DMLCarget tracking system.

METHODS AND MATERIALS

MLC target tracking systemA schematic diagram and photograph of the DMLC target

racking system are shown in Fig. 1. A motion simulator (11, 12)as programmed to follow patient tumor motion paths. Details of

imely technical support and advice. Chris Bartee, Mark Hile,nd Matthew Schaefer provided engineering support. Elisabetheiss and Michelle Svatos offered useful critique on the manu-

cript. Many others from VCU Radiation Oncology and Varianedical Systems have also contributed to the VCU 4D Radio-

herapy project.Received Feb 8, 2006, and in revised form April 18, 2006.

ccepted for publication April 18, 2006.

tta(rpwbtandc

ucwlmwtlpeT

rult

octdttafNoea

●

●

●

itddtasa

Ftmicc

F(b

1580 I. J. Radiation Oncology ● Biology ● Physics Volume 65, Number 5, 2006

he patient data are given below. The motion was optically de-ected using the Varian (Palo Alto, CA) Real-Time Position Man-gement (RPM) system in which images acquired with a cameralabeled as camera 1 in Fig. 1a) of a marker box with two infraredeflecting markers are segmented and positions calculated. Theseositions were sent via serial port to a computer. Software wasritten to calculate the required leaf positions to track the targetased on the incoming position signal. These leaf positions werehen communicated via ethernet to the MLC controller, whichctuated the mechanical leaf motion of the MLC (Varian Millen-ium 120-leaf). For the studies described herein leaf sequencesescribing a circle were used, though the choice of leaf shapeould be chosen to fit an arbitrary target.

A separate beam’s eye view camera (camera 2 in Fig. 1a) wassed to synchronously measure the target and leaf motion. Theamera-MLC-target distances (18 cm and 17 cm respectively)ere a similar ratio to the actual source-MLC-target distances in a

inear accelerator. Video images of the target motion, and DMLCotion were recorded and analyzed. The ‘target’ for these studiesas a reflective marker. To facilitate obtaining the DMLC posi-

ion, reflective markers were placed on 2 of the leaves, 1 on eacheaf bank, and the 2 marker positions averaged. The DMLCositions and the target motion were both measured by segmentingach image frame from the camera above the motion simulator.

ig. 1. Experimental set-up of the dynamic multileaf collimatorDMLC) target tracking system. (a) Schematic diagram; (b) la-eled photograph.

hus the 2 signals (DMLC and target motion) were simultaneously m

ecorded by the same device. Image analysis software was writtensing Matlab (Mathworks, Natick, MA) to segment the target andeaf reflective markers. Example images of the video files analyzedo obtain the data are shown in Fig. 2.

A motion simulator was programmed to reproduce either peri-dic sinusoidal motion or nonperiodic patient motion. A sinusoidalurve with a 2 cm range of motion and a 6 s period was used forhe response time calculation, motion calibration, and accuracyetermination under known conditions. Three representative pa-ient motion examples, covering a range of breathing types andumor locations, were simulated. The tumor motion data werecquired under IRB protocol at Georgetown University Hospitalrom Cyberknife (Accuray, Sunnyvale, CA) Synchrony treatments.ote that the motion data are periodically measured using orthog-nal X-ray, and predicted between X-ray images based on anxternal optical signal (7). Brief details of the patient tumor motionre given below:

Patient 1 was a middle-aged man with a lower lobe lung tumor.The tumor motion was regular with a range of motion of 10 to15 mm.Patient 2 was a middle-aged woman with an advanced-stageright central lung tumor. The tumor motion was small and veryirregular, with a motion period of approximately 1 s.Patient 3 was an elderly woman with a lower lobe lung tumor.The tumor motion was fairly regular with a range of motion of6 to 10 mm.

The DMLC tracking system currently only accounts for motionn the superior-inferior (SI) direction, parallel to the leaf motion;he RPM system only tracks motion in the anterior-posterior (AP)irection. For these reasons only the SI component of the patientata were used (this has the largest magnitude of the 3 dimensions)o simulate both SI and AP motion so that the RPM system couldcquire the motion data, and the DMLC tracking system compen-ate for it. For each of the cases, video images were acquired forpproximately 60 s.

ig. 2. Images of the dynamic multileaf collimator (DMLC) targetracking for (a) downward target motion (inhale to exhale), (b) nootion (end exhale), and (c) upward target motion (exhale to

nhale). The red circles/crosses are the segmented leaf and targetenter positions, the yellow circle/cross is the DMLC targetingenter position. The DMLC is seen to be lagging behind the target

otion because of the system response time.

eitltsatttctswpbwtb

Da

iMwmrmwc

C

msptcrtmodt

C

bco(atf

bi

tdrpystco

Da

lttlms

C

fiiwbtr

C

aorthrwrtetipr

dsdsD1sd

1581Dynamic multileaf collimator tracking system ● P. J. KEALL et al.

EXPERIMENTAL PROCEDURES

First, the uncertainty of the system needed to be established, tonsure the tracking errors found were credible. Next, we werenterested in determining the system response time to understando what extent the tracking error is limited by the present techno-ogic implementation. The response time (or system latency), i.e.,he time for the DMLC to respond to a given target motion, is theum of the time for the tasks of image acquisition, segmentationnd processing by the RPM system, leaf position calculation, dataransfer, and leaf motion execution. We passively tracked thearget first to determine the alignment between the DMLC and thearget. The response time was estimated and the response time–orrected average DMLC position was matched to the averagearget position. The subsequently recorded target and DMLC po-itions were used for analysis. This scenario is not dissimilar tohat would occur for patient applications—the DMLC wouldassively track the patient motion without radiation delivery untileam-target alignment occurred at which time radiation deliveryould be initiated. This approach would also be used for resump-

ion of an interruption during delivery (e.g., an interruption causedy a cough).

etermining the uncertainty of the tracking system in thebsence of target motionTo determine the uncertainty of the measurement system, video

mages were recorded with the target static. The target and leafLC positions were segmented in the images using custom-ritten software. The high contrast of the image of the reflectivearkers used made the segmentation process relatively trivial. No

elative motion of the two inputs was expected; any observedotion is an estimate of the system measurement uncertainty,hich was used to put the subsequently analyzed tracking errors in

ontext.

alibrating motion and calculating system response timeThe motion simulator was programmed to exhibit sinusoidalotion with a known range of motion (2 cm). Tracking the

inusoidal motion enabled the calibration of the requested leafosition change based on the incoming position information fromhe RPM system. The system response time is the time taken toomplete the feedback loop shown in Fig. 1a. The effect of theesponse time on the DMLC position can be seen in Fig. 2, wherehe DMLC position is lagging behind the target position when inotion. Figure 3 also shows the DMLC positions are slightly

ffset from the target positions. The system response time wasetermined by computing the phase difference from a sinusoidal fito the target motion and DMLC motion.

haracterizing geometric tracking accuracyTo determine the alignment between the DMLC and the target,

oth the DMLC and target motion were recorded for 6 s. Thehoice of 6 s is arbitrary, and is used to yield a reasonable estimatef the target position for alignment. The response time–correctedsee following) average DMLC position was matched to the aver-ge target position over this time period. Subsequent recordings ofhe target position (beyond 6 s) and the DMLC positions were usedor analysis.

The analysis included computing the tracking error (differenceetween DMLC and target position), response time–corrected track-

ng error (difference between response time–corrected DMLC and i

arget position) and the time integrated distributions (probabilityensity functions) of these parameters. The response time–cor-ected DMLC positions were computed by shifting the DMLCositions in the time axis by 2 image frames (130 ms). This dataields an estimate of the system accuracy if the feedback loophown in Fig. 1 were completed in �30 ms. It should be noted thathe response time–corrected data cannot be realized with theurrent system, however is included to estimate the accuracy limitsf an improved DMLC tracking system.

RESULTS

etermining the uncertainty of the tracking system in thebsence of target motionWith the target static, the system detected target motion of

ess than 0.1 mm (1�), and recorded a tracking error of lesshan 0.15 mm (1�). This measurement error is an estimate ofhe accuracy attainable with the current set-up, and is muchower than the tracking errors measured in their presence ofotion (described below) indicating that the following re-

ults are reasonable estimates of their true values.

alculating system response timeBy computing the phase difference based on a sinusoidal

t to the target motion and DMLC motion for repeat track-ng of sinusoidal motion, the average system response timeas computed to be 160 � 2 ms. This average value variesecause of discrete events such as the RPM image acquisi-ion (33-ms period), leaf position calculations (20-ms pe-iod) and the MLC cycle period (50-ms period).

haracterizing geometric tracking accuracyThe target tracking accuracy of the DMLC for sinusoidal

nd patient motion can be seen in Fig. 3. The DMLC isbserved to lag behind the target (as expected based on theesponse time measurements). Because of this responseime, the tracking error is largest when the target velocity isighest. If this response time were significantly reduced, theesponse time–corrected DMLC curve matches very closelyith the target motion. The resultant response time–cor-

ected tracking errors are very small. By integrating overime, probability density functions (pdfs) of these trackingrrors were generated and are displayed in Fig. 4. Theracking error distribution is significantly lower than thenitial target motion distribution; however, still has an ap-reciable width. Reducing the response time would furthereduce the tracking error.

The pdfs were quantified based on their means and stan-ard deviations. The results are shown in Table 1. There areeveral interesting features. In the first column, the meanisplacements are small, though nonzero, indicating that ahift had taken place since the initial alignment of theMLC and target in the first few seconds. For Patient 3 a-mm mean difference was observed. Had the data acqui-ition been for longer than a minute, larger systematicifferences may have been observed. The systematic track-

ng error is very small in all cases. For Patient 2, though the

thetwil

t

ttwweetirlt

atient

1582 I. J. Radiation Oncology ● Biology ● Physics Volume 65, Number 5, 2006

arget motion was on average small, because of the extremelyigh frequency and irregular breathing motion the trackingrror was the same as the target motion indicating no benefit ofracking for this patient. The tracking error for patients 1 and 3ere 40% and 35% of the target motion respectively. Correct-

ng for the response time of the system yields significantlyower tracking errors, of the order of 0.3 mm (1�).

DISCUSSION

The geometric accuracy of a DMLC target tracking sys-

Fig. 3. Position vs. time plots of the target position (red(blue dashed line) and the response time–corrected DML(c) Patient I data. Position vs. time plots of DMLC traDMLC tracking error for the sinusoidal curve (b) and P

em to track sinusoidal and patient motion has been inves- e

igated. The observed tracking errors for the patient data, 0.6o 1.1 mm (1�) are encouraging, and give an estimate ofhat geometric errors may be expected of such a systemhen clinically implemented. The magnitude of tracking

rror is approximately equal to that calculated by Vedamt al. (13) for a response time of 160 ms (�1 mm), thoughhe patient data were different in these studies. The lack ofmprovement in the tracking results for Patient 2 are aeminder to be careful; indeed if the response time wereonger it is likely the use of tracking would have reducedargeting accuracy over not tracking. Overall, the tracking

ne), the dynamic multileaf collimator (DMLC) positiontion (black dash–dot line) for (a) a sinusoidal curve andrror (blue solid line) and the response time–corrected

1 data (d).

solid liC posicking e

rrors are random in nature with a small systematic com-

peftabttc

siTAsospiiqi

twadnitqases

tftmrmr

c

ash–do

SPPP

1583Dynamic multileaf collimator tracking system ● P. J. KEALL et al.

onent. According to published margin formulas, randomrrors are less deleterious than systematic errors as shown,or example, in the table in van Herk et al. (14). However,heir magnitude should still be quantified and accounted forppropriately. The dynamic motion compensation offeredy DMLC-based target tracking means that if variations inhe patient’s initial treatment position are observed duringreatment then the tracking system can automatically ac-ount for the change in position.

An important component of any dynamic compensationystem is the target position monitoring system. Any errorsn this process will manifest themselves as tracking errors.he motion detection system used here was optically based.s the target was visible for the experiments, the input

ignal was highly correlated with the target motion. The usef optical motion signals as a surrogate for internal motionhould be used with caution, as variations in correlation andhase shifts (15–20) have been observed between monitor-ng systems and internal structures. An ideal position mon-toring system would have high accuracy, high update fre-uency, low processing time and give a large volume ofnformation about the target (and normal structure) posi-

Fig. 4. Probability density functions for (a) a sinusoidalinput target motion (red solid), the dynamic multileaf cresponse time–corrected DMLC tracking error (black d

Table 1. Mean (x�) and standard deviation (�) of the target moa 60-s tr

Data source

Target displacement

x� (mm) � (mm)

ine curve 0.12 7.22atient 1 �0.36 2.77atient 2 0.02 0.61

atient 3 1.00 2.34 �0.0

ions. The motion detection system used in the current workas limited to one dimension, thus the target tracking was

lso limited to accounting for motion along the leaf motionirection. Tumor motion has been observed to be predomi-antly in one direction, (21) however hysteresis was observedn some cases. For DMLC tracking it would be prudent to alignhe DMLC with the major axis of tumor motion. High-fre-uency (relative to target motion) 3D motion detection systemsre available or becoming available, based on dual fluoro-copic, (22) combined fluoroscopic and optical, (7, 8) andlectromagnetic technology (23). These 3D motion detectionystems would integrate well with a DMLC tracking system.

The results of the hypothetical response time–correctedracking error scenario show a clear pathway for where toocus future development efforts. An alternative to reducinghe response time to improve accuracy is to incorporateotion prediction algorithms (13, 24–26). On reviewing

esults published by Vedam et al. (13), at the response timeeasured here (160 ms) motion prediction is likely to

educe the tracking error by 30%.The patient motion data used was well within the me-

hanical velocity and acceleration constraints of the DMLC

nd (b) Patient 1 data. Each plot shows three curves: thetor (DMLC) tracking error (blue dashed line) and thet line).

acking error and response-time–corrected tracking error overperiod

Tracking errorTracking error

(response-time–corrected)

) � (mm) x� (mm) � (mm)

5 1.22 0.04 0.411 1.12 �0.01 0.301 0.61 �0.02 0.31

curve aollima

tion, tracking

x� (mm

0.00.0

�0.0

7 0.84 �0.07 0.31

fattct

acgs

ctgmeac

1

1

1

1

1

1

1584 I. J. Radiation Oncology ● Biology ● Physics Volume 65, Number 5, 2006

or measurements of the same MLC type, (27) and theddition of a beam hold during extremely rapid target mo-ion would not have significantly increased the accuracy ofhe system. Such a beam hold would be desirable for thelinical implementation of DMLC tumor tracking to avoidreating during times of rapid target motion, e.g., coughing.

The DMLC tracking system investigated here was onlypplied to lung applications, since this is one of the mosthallenging sites for tracking. However the technology iseneral enough to account for detected motion for other

ites, e.g., prostate, pancreas, and liver. c

REFEREN

5. Vedam SS, Kini VR, Keall PJ, et al. Quantifying the

1

1

1

1

2

2

2

2

2

2

2

2

CONCLUSIONS

A DMLC target tracking system has been developed thatan account for detected motion parallel to the leaf direc-ion. The response time of the current system is 160 ms. Theeometric precision for tracking patient motion is 0.6 to 1.1m (1�) for the three patient datasets tested, with tracking

rrors relative to the original patient motion of 35%, 40%,nd 100%. The tracking error has a negligible systematicomponent. Reducing the response time will further in-

rease the overall system accuracy.

CES

1. Webb S. IMRT delivery techniques. In: Bortfeld T, Schmidt-Ullrich R, De Neve W, et al., editors. Image-guided IMRT.Heidelberg: Springer-Verlag; 2006. p. 73–90.

2. Keall PJ, Kini VR, Vedam SS, et al. Motion adaptive x-raytherapy: a feasibility study. Phys Med Biol 2001;46:1–10.

3. Suh Y, Yi B, Ahn S, et al. Aperture maneuver with compelledbreath (AMC) for moving tumors: a feasibility study with amoving phantom. Med Phys 2004;31:760–766.

4. Neicu T, Shirato H, Seppenwoolde Y, et al. Synchronizedmoving aperture radiation therapy (SMART): average tumourtrajectory for lung patients. Phys Med Biol 2003;48:587–598.

5. Papiez L, Rangaraj D. DMLC leaf-pair optimal control formobile, deforming target. Med Phys 2005;32:275–285.

6. Papiez L, Rangaraj D, Keall P. Real-time DMLC IMRTdelivery for mobile and deforming targets. Med Phys 2005;32:3037–3048.

7. Schweikard A, Glosser G, Bodduluri M, et al. Robotic motioncompensation for respiratory movement during radiosurgery.Comput Aided Surg 2000;5:263–277.

8. Schweikard A, Shiomi H, Adler J. Respiration tracking inradiosurgery. Med Phys 2004;31:2738–2741.

9. Uematsu M. CT-guided focal high dose radiotherapy. 4th S.Takahashi International Workshop on 3 Dimensional Confor-mal Radiotherapy. Nagoya, Japan; 2004. p. 81.

0. D’Souza WD, Naqvi SA, Yu CX. Real-time intra-fraction-motion tracking using the treatment couch: a feasibility study.Phys Med Biol 2005;50:4021–4033.

1. Zhou TJ, Dieterich S, Cleary K. A robotic 3-D motion simu-lator for enhanced accuracy in Cyberknife radiosurgery. In:Lemke HU, Inamura K, Doi K, Vannier MW, Farman AG,Reiber JHC, editors. Computer aided radiology and surgery.London: Elsevier; 2004. p. 323–328.

2. Tang J, Dieterich S, Cleary K. Respiratory motion tracking ofskin and liver in swine for CyberKnife motion compensation.In: Galloway RL Jr, editor. SPIE Med Imaging Proceedings;2004;5367:729–734.

3. Vedam SS, Keall PJ, Docef A, et al. Predicting respiratorymotion for four-dimensional radiotherapy. Med Phys 2004;31:2274–2283.

4. van Herk M. Errors and margins in radiotherapy. Semin RadiatOncol 2004;14:52–64.

predictability of diaphragm motion during respiration witha noninvasive external marker. Med Phys 2003;30:505–513.

6. Mageras GS, Pevsner A, Yorke ED, et al. Measurement oflung tumor motion using respiration-correlated CT. Int J Ra-diat Oncol Biol Phys 2004;60:933–941.

7. Ahn S, Yi B, Suh Y, et al. A feasibility study on the predictionof tumour location in the lung from skin motion. Br J Radiol2004;77:588–596.

8. Hoisak JD, Sixel KE, Tirona R, et al. Correlation of lungtumor motion with external surrogate indicators of respiration.Int J Radiat Oncol Biol Phys 2004;60:1298–1306.

9. Koch N, Liu HH, Starkschall G, et al. Evaluation of internallung motion for respiratory-gated radiotherapy using MRI:part I–correlating internal lung motion with skin fiducial mo-tion. Int J Radiat Oncol Biol Phys 2004;60:1459–1472.

0. Tsunashima Y, Sakae T, Shioyama Y, et al. Correlation be-tween the respiratory waveform measured using a respiratorysensor and 3D tumor motion in gated radiotherapy. Int JRadiat Oncol Biol Phys 2004;60:951–958.

1. Seppenwoolde Y, Shirato H, Kitamura K, et al. Precise andreal-time measurement of 3D tumor motion in lung due tobreathing and heartbeat, measured during radiotherapy. Int JRadiat Oncol Biol Phys 2002;53:822–834.

2. Shirato H, Shimizu S, Kunieda T, et al. Physical aspects of areal-time tumor-tracking system for gated radiotherapy. Int JRadiat Oncol Biol Phys 2000;48:1187–1195.

3. Balter JM, Wright JN, Newell LJ, et al. Accuracy of a wirelesslocalization system for radiotherapy. Int J Radiat Oncol BiolPhys 2005;61:933–937.

4. Sharp GC, Jiang SB, Shimizu S, et al. Prediction of respiratorytumour motion for real-time image-guided radiotherapy. PhysMed Biol 2004;49:425–440.

5. Murphy MJ. Tracking moving organs in real time. SeminRadiat Oncol 2004;14:91–100.

6. Isaksson I, Jalden J, Murphy MJ. On using an adaptive neuralnetwork to predict lung tumor motion during respiration forradiotherapy applications. Med Phys 2005;32:3801–3809.

7. Wijesooriya K, Bartee C, Siebers JV, et al. Determination ofmaximum leaf velocity and acceleration of a dynamic mul-tileaf collimator: implications for 4D radiotherapy. Med Phys

2005;32:932–941.