Int. J. Radiation Oncology Biol. Phys., Vol. 71, No. 1, Supplement, pp. S28–S32, 2008Copyright � 2008 Elsevier Inc.

Printed in the USA. All rights reserved0360-3016/08/$–see front matter

doi:10.1016/j.ijrobp.2007.08.082

QA FOR RT SUPPLEMENT

DEVELOPING QUALITY ASSURANCE PROCESSES FOR IMAGE-GUIDED ADAPTIVERADIATION THERAPY

DI YAN, D.SC.

Department of Radiation Oncology, William Beaumont Hospital, Royal Oak, MI

Quality assurance has long been implemented in radiation treatment as systematic actions necessary to provideadequate confidence that the radiation oncology service will satisfy the given requirements for quality care. Theexisting reports from the American Association of Physicists in Medicine Task Groups 40 and 53 have providedhighly detailed QA guidelines for conventional radiotherapy and treatment planning. However, advanced treat-ment processes recently developed with emerging high technology have introduced new QA requirements thathave not been addressed previously in the conventional QA program. Therefore, it is necessary to expand the ex-isting QA guidelines to also include new considerations. Image-guided adaptive radiation therapy (IGART) isa closed-loop treatment process that is designed to include the individual treatment information, such as pa-tient-specific anatomic variation and delivered dose assessed during the therapy course in treatment evaluationand planning optimization. Clinical implementation of IGART requires high levels of automation in image acqui-sition, registration, segmentation, treatment dose construction, and adaptive planning optimization, which bringsnew challenges to the conventional QA program. In this article, clinical QA procedures for IGART are outlined.The discussion focuses on the dynamic or four-dimensional aspects of the IGART process, avoiding overlap withconventional QA guidelines. � 2008 Elsevier Inc.

Image guidance, Adaptive radiotherapy, Quality assurance.

INTRODUCTION

Advanced treatment processes recently developed with

emerging high technology have introduced new QA require-

ments that have not been addressed previously in conventional

QA programs (1, 2). It is therefore necessary to expand exist-

ing QA programs to include these new considerations.

The major difference between conventional radiotherapy

practice and image-guided adaptive radiotherapy (IGART)

is the use of individual patient dynamic or time-serial

(four-dimensional [4D]) treatment history. In an IGART pro-

cess, treatment planning and modification decisions are typ-

ically made on the basis of patient-specific anatomic and

biological variations determined from multiple image mea-

surements of patient anatomy obtained at various times dur-

ing the course of treatment delivery (3, 4). Therefore, QA

tests should be performed with respect to these variations be-

fore and during the treatment process. In the following sec-

tions, a clinical IGART QA program developed at William

Beaumont Hospital is discussed. The focus is on the 4D as-

pects of the IGART process, avoiding overlap with conven-

tional QA guidelines.

S

DATA TRANSFER AND STORAGE IN THE IGARTPROCESS

The IGART process links patient image and therapy ma-

chine output obtained at the treatment delivery back to treat-

ment evaluation and planning modification. The patient

image and machine output first enter into multiple computer

servers for image registration and segmentation, dose recon-

struction, and treatment evaluation before reaching the adaptive

planning modification stage. In our clinical implementation,

all measurement devices and computer servers are on the net-

work with a maximum transfer speed 1 Gb/sec. Each server

automatically performs new tasks distributed by a process

task controller. A central data-storage with a hard disk space

of 30 Tb is used to store all patient data (Fig. 1). It includes

patient reference images for pretreatment planning (1.5 Gb

maximum per patient), daily cone beam CT (CBCT; 23 Gb

maximum per patient), and portal images (0.2 Gb maximum

per patient), organ displacement vector field and segmenta-

tion (9 Gb maximum per patient), daily patient treatment

dose distribution (0.15 Gb maximum per patient), and patient

plan information (1 Gb maximum per patient). Maximum

Reprint requests to Di Yan, D.Sc., Department of RadiationOncology, William Beaumont Hospital, 3601 West Thirteen MileRoad, Royal Oak, MI. Tel: (248) 551-6589; Fax: (248) 551-3784;E-mail: dyan@beaumont.edu

Conflict of interest: none.Acknowledgement—This short article provides a summary of clini-

28

cal QA development for image-guided adaptive radiotherapy per-formed by the research and clinical staffs in the Department ofRadiation Oncology, William Beaumont Hospital.

Received Feb 16, 2007, and in revised form Aug 4, 2007.Accepted for publication Aug 7, 2007.

Fig. 1. Hardware storage and connection for image-guided adaptive radiation therapy process.

QA for IGART d D. YAN S29

35 Gb data storage per patient is necessary for the individual

enrolled in the IGART process. Therefore, our data storage

has the capacity to handle about 800 patients simultaneously

treated with the IGART process. In addition, a secondary

data-storage server (15 Tb) is placed at Hospital Information

Service for daily backup. Data for patients who have com-

pleted treatment are sent to the hospital archive at 3-month

intervals.

QA ROLE OF IGART TEAM MEMBERS

Radiation oncologists, physicists, dosimetrists, and thera-

pists/image specialists are the core personal performing qual-

ity control and assurance for the IGART process. Radiation

oncologists are responsible for targeting delineation and

dose prescription including both pretreatment planning and

adaptive plan modification. Dosimetrists delineate normal or-

gans and organ-confined targets, such as the prostate and

seminal vesicles. However, a radiation oncologist must re-

view and provide the final approval. The radiation oncolo-

gists are responsible for weekly review of patient organ and

treatment dose variation to decide whether treatment plan

modification is necessary.

Radiation dosimetrists are responsible for manual image

registration, organ delineation on the treatment images, and

creation of a modified treatment plan. In addition, dosimet-

rists are also responsible for verifying organ contours.

Radiation therapists/image specialists are responsible for

onboard image localization tests and patient localization

and correction. The therapists/image specialists perform on-

line CBCT-guided patient position correction based on

a physicist’s or physician’s approval for hypofractionated

treatment. For standard fractionated treatment, a physicist

or physician is consulted if the therapist/image specialist is

uncertain of the image registration.

Radiation oncology physicists are responsible to define

and manage all technical QA procedures during the IGART

operation. The physicists certify all hardware and software

tools for image registration, segmentation, parameter assess-

ment, dose reconstruction/evaluation, and adaptive planning.

The physicists are also responsible for providing baseline ref-

erence and tolerance for QA verifications and for educating

the clinical staff to use these standards properly. In addition,

the physicists should identify the source of error when clini-

cal staff members observe a large discrepancy during a rou-

tine QA process.

UNCERTAINTIES IN THE CLINICAL IGARTPROCESS

The IGART process reduces uncertainty in the patient’s

treatment position but shares the uncertainties that exist in

conventional treatment simulation, planning, and delivery.

Additionally, the specific and most significant sources of un-

certainties are patient image acquisition and registration,

couch or multileaf collimator (MLC) correction, estimation

of the systematic and random position variations, treatment

dose construction and evaluation, and adaptive planning

modification, which are outlined in the following sections.

Uncertainties in treatment position localizationThe IGART process involves frequent measurements of

patient anatomic position obtained by comparing treatment

images with a reference image of similar modality, that is,

the onboard CBCT image to the reference planning CT image

or the portal image to the digitally reconstructed radiography.

In addition, 4D CBCT (5) or radiographic motion view have

S30 I. J. Radiation Oncology d Biology d Physics Volume 71, Number 1, Supplement, 2008

been used to monitor and determine tumor motion with re-

spect to the 4D reference CT (6) or digitally reconstructed

fluoroscopy (7, 8). Each of these processes contains uncer-

tainties in imaging patient treatment position, image transfer

and registration, and determination of radiation beam accu-

racy. A number of studies (9–22) have been performed to as-

sess these uncertainties, determine their influence on

treatment, and establish the corresponding clinical QA toler-

ances. Integrated system calibration has been performed us-

ing a phantom with embedded radio markers to establish

baseline accuracy of the patient localization system including

CT simulation, planning, localization on the treatment ma-

chine, cone beam imaging, correction, beam delivery, and

multivoltage portal imaging. The results (11, 12, 16, 18)

have confirmed that the achievable system accuracy with re-

spect to beam isocenter can be within 1 mm, referring to both

the systematic and random errors. Both 2D and 3D image

rigid-body registration uncertainties have been evaluated us-

ing the Rando head phantom (Phantom Laboratory, Salem,

NY) (18). It is not surprising that image registration uncer-

tainties are greater than those obtained from the phantom

with radio markers, specifically in the longitudinal direction

because of the limited CT slice resolution.

In contrast to phantom study of rigid-body registration, un-

certainty studies on deformable image registration have been

limited because of a lack of proper QA phantoms. Deform-

able organ registration and segmentation have been evaluated

using patient images with or without insert radio markers

(22–24). However, the purpose of these studies was to vali-

date registration methods, rather than establish clinical QA.

It will be helpful to have a deformable phantom to test regis-

tration method and verify treatment dose construction

discussed subsequently.

Uncertainties in couch or beam aperture correctionCouch position and MLC-based beam aperture adjustments

have been the most common means to correct patient treat-

ment position. However, the uncertainties associated with

each correction method should be determined before a routine

application in the clinic. Couch correction accuracy has been

studied using a rigid body phantom (16, 18). Translational cor-

rection accuracy has been achieved within 1 mm. However,

treatment position corrections using an MLC-based beam

aperture shift are nontrivial because of the limited leaf width.

A recent study (25) of online image guidance and rigid-body

correction using MLC has shown that a maximum 2% dose

discrepancy in the target and a much higher dose discrepancy

in normal tissues could occur in both prostate and head/neck

cancer treatment. Beam aperture correction has also been pro-

posed and tested to compensate for target deformation (26) in

conformal radiotherapy, and this was later expanded to IMRT

(27). However, the dosimetric uncertainty caused by multiple

leaf segment adjustment has not been fully explored thus far.

Uncertainties in determining patient variation parametersA unique feature of the IGART process is the estimation

of the systematic and random errors of individual treatment

positions. A few methods (4, 26, 28) have been applied to

provide the estimation. Because the estimation is performed

on the basis of limited samples of measurements, uncertainty

in parameter estimation is always accompanied by statistical

residuals. The systematic and random errors of treatment

setup position have been systematically evaluated for pros-

tate and head/neck cancer treatment. Following the estima-

tion, performed at the end of the first treatment week,

patient setup position has been continuously monitored for

additional 3 to 5 treatment days. If the residual systematic er-

ror after the first correction in the prostate cancer treatment

was 1 mm (the predetermined cutoff value) larger than the es-

timated value, a second correction was performed. Among

our current patient data (a total of 983 patients who had com-

pleted treatment from October 1999 to July 2006), 9.8% had

a second correction, and 0.7% had third correction. For head/

neck cancer treatment, additional position correction was

performed if the residual systematic error was >2 mm. In

this case, 18% of 89 patients had a second correction, and

2% had a third correction. The difference in the frequency

of the second correction directly reflected the decision rule

applied in the correction process. In addition to patient setup

position, the patient-specific internal target volume to com-

pensate for internal target motion (26) has been evaluated

by examining weekly CT images for the first 269 prostate

cancer patients treated in our clinical IGART process. Our re-

sults show that 98% of patients received excellent treatment

coverage (<2% target dose discrepancy).

Uncertainties in treatment dose construction andevaluation

The Department of Radiation Oncology at William Beau-

mont Hospital has a major task in implementing individual

patient dose-tracking and feedback adaptive processes in

the routine clinic. At present, the dose-tracking and feedback

process is undergoing clinical testing and evaluation, rather

than routine use. Treatment dose—daily as well as cumula-

tive—in organs of interest has been constructed using volu-

metric image-based deformable organ registration (29, 30).

Two major factors that have been considered in treatment

or 4D dose construction are organ subvolume position dis-

placement and variation of patient global density distribution.

A study conducted at our facility (31) has shown that ignor-

ing patient global density variation in treatment dose con-

struction may cause dose evaluation discrepancy. The

discrepancy is commonly small for dose evaluation in the

pelvic region (<3%), noticeable in the head/neck region

(5%–10%), and significant in the lung region (up to 25%).

The study has also shown that significant improvement in

dose construction accuracy could be achieved if the mean

CT image weighted by organ position probability density

was used in dose calculation. Treatment dose in organs of in-

terest can be updated frequently in the IGART process for

treatment evaluation and planning modification decisions.

However, uncertainties in image registration and organ delin-

eation directly affect the reliability of the dose in treatment

QA for IGART d D. YAN S31

evaluation. Study of this consideration must therefore be sys-

tematically explored.

Uncertainties in adaptive planning modificationTwo typical adaptive planning methods have been ex-

plored. One is to determine patient-specific clinical target

volume to planning target volume margins (26) and then

design the corresponding dose distribution. The other is to

design dose distribution to compensate directly for patient-

specific motion with adaptive inverse planning (32). Sources

of uncertainty in adaptive planning are similar to those in

conventional planning as reported by American Association

of Physicists in Medicine (AAPM) Task Group 53, although

their influences may be different. The influence of organ de-

lineation uncertainty on the accuracy of dose evaluation may

be reduced by a potential random effect in multiple organ de-

lineations. However, the accuracy of dose evaluation be-

comes more critical in the selection of the individual

prescription dose. Therefore, IGART quality control should

include selection of the most robust planning parameters,

such as selecting dose–volume constraints on the relatively

stable portion of the dose–volume histogram (DVH) curve

for planning evaluation.

QA TESTS IN THE CLINICAL IGART PROCESS

Clinical development of IGART QA should aim to address

the most significant sources of uncertainty. QA recommenda-

tions on image acquisition, anatomic description, dose calcu-

lation, and treatment planning are detailed in AAPM Task

Group Reports 40 and 53. In following these recommenda-

tions, we focus only on additional QA procedures that should

be applied to implement a clinical IGART process.

QA verification for patient treatment position assessmentand correction

Staff physicists should perform an integrated system test

on image localization and position correction accuracy before

starting an IGART protocol to determine a baseline reference

for routine monthly and daily QA tests. A therapist/image

specialist performs isocenter localization tests daily for on-

line image-guided hypofractionation and monthly for offline

image guidance. These tests should also be performed again

when software or hardware devices are updated. Staff phys-

icists perform integrated system tests from the CT simulation

to treatment delivery semiannually. A phantom with embed-

ded radio markers can be applied for these image localization

and position correction tests. In addition, a phantom mounted

on a motor-driven motion stage can be used for respiratory-

correlated CT or CBCT imaging QA.

QA verification for parameter estimation and treatmentdose construction

The systematic and random errors of the individual treat-

ment position are verified the week after the assessment by

a physicist or dosimetrist and evaluated weekly by a physician

throughout the course of the treatment. It is strongly recom-

mended that the changes in organ volume and the organ’s

center of mass be evaluated weekly. Treatment dose to the or-

gans of interest should be evaluated by a physicist daily for

hypofractionated radiotherapy and weekly for a normal frac-

tionated treatment. If a large dose discrepancy is observed,

the corresponding treatment image and organ contours

should be evaluated.

QA verification for treatment evaluation and adaptiveplanning

The planning CT image should be verified to identify and

eliminate the effect of contrast material and unusual events

(i.e., bladder contrast and large rectal gas filling) on the

dose calculation. The treatment volumetric images used in

adaptive planning are verified weekly to ensure that they

are appropriated for the 4D treatment dose construction.

When a treatment DVH parameter is used for treatment eval-

uation, weekly verification is recommended to ensure that the

parameter is not on the high gradient portion of the DVH

curve. When patient-specific PTV is used in adaptive plan-

ning, weekly CT image verification is recommended to avoid

unexpected deviation.

REFERENCES

1. Kutcher G, Coia L, Gillin M, et al. Comprehensive QA for ra-diation oncology: Report of AAPM Radiation Therapy Com-mittee Task Group 40. Med Phys 1994;21:581–618.

2. Fraass B, Doppke K, Hunt M, et al. Radiation Therapy Commit-tee Task Group 53: Quality assurance for clinical radiotherapytreatment planning. Med Phys 1998;25:1773–1829.

3. Yan D, Vicini F, Wong J, Martinez A. Adaptive radiation ther-apy. Phys Med Biol 1997;42:123–132.

4. Yan D. Image-guided/adaptive radiotherapy. In: Schlegel W,Bortfeld T, Grosu AL, editors. New technologies in radiationoncology. New York: Springer-Verlag; 2005. p. 321–336.

5. Sonke JJ, Zijp L, Remeijer P, van Herk M. Respiratory corre-lated cone beam CT. Med Phys 2005;32:1176–1186.

6. Ford EC, Mageras GS, Yorke E, Ling CC. Respiration-corre-lated spiral CT: A method of measuring respiratory-induced an-atomic motion for radiation treatment planning. Med Phys 2003;30:88–97.

7. Hugo G, Yan D, Watt L, et al. A method for portal verification

of 4D lung treatment. XIVth International Conference on the

Use of Computers in Radiotherapy, Seoul, Korea, May 10–

13, 2004. p. 82–85.8. Hugo G, Vargas C, Liang J, et al. Online portal fluoroscopic ver-

ification of adaptive radiotherapy for lung cancer [Abstract]. Int

J Radiat Onocol Biol Phys 2004;60(Suppl. 1):S335.9. Yan D, Ziaja E, Jaffray D, et al. The use of adaptive radiation

therapy to reduce setup error: A prospective clinical study. Int

J Radiat Onocol Biol Phys 1998;41:715–720.10. Pisani L, Lockman D, Jaffray D, et al. Setup error in radiother-

apy: On-line correction using electronic kilovoltage and mega-

voltage radiographs. Int J Radiat Onocol Biol Phys 2000;47:

825–839.11. Letourneau D, Martinez AA, Lockman D, et al. Assessment of

residual error for online cone beam CT guided treatment of

S32 I. J. Radiation Oncology d Biology d Physics Volume 71, Number 1, Supplement, 2008

prostate patients. Int J Radiat Onocol Biol Phys 2005;62:1239–1246.

12. Oldham M, Letourmeau D, Watt L, et al. Cone-beam guided ra-diation therapy, Part II: A model for on-line application. RadiatOncol 2005;75:271–278.

13. Weed DW, Yan D, Martinez A, et al. The validity of clips asa radiographic surrogate for the lumpectomy cavity in imageguided accelerated partial breast irradiation. Int J Radiat OnocolBiol Phys 2004;60:484–492.

14. Lockman D, Yan D, Wong J, et al. Cone-beam in an offline im-age guidance strategy: Margin and efficiency gains [abstract].Int J Radiat Onocol Biol Phys 2004;60(Suppl.):S198.

15. Kim L, Vicini F, Yan D, et al. Reduction of PTV margin foraccelerated partial breast irradiation using on-line detection ofsurgical clips [Abstract]. Int J Radiat Onocol Biol Phys 2004;60(Suppl.):S336.

16. Kim L, Wong J, Yan D. Online localization of the lumpectomycavity using surgical clips. Int J Radiat Onocol Biol Phys 2007;69:1305–1309.

17. Hugo G, Liang J, Campbell J, Yan D. Online target positionlocalization in the presence of respiration: a comparison oftwo methods. Int J Radiat Onocol Biol Phys 2007;69:1634–1641.

18. Sebastian E, Kim L, Wloch J, et al. Comparison of 2D imagealignment techniques for head and neck [Abstract]. Int J RadiatOnocol Biol Phys 2006;66(Suppl.):S650.

19. Wu Q, Ivaldi G, Liang J, et al. Geometric and dosimetric eval-uations of an online image-guided 3D conformal radiation ther-apy of prostate cancer. Int J Radiat Onocol Biol Phys 2006;64:1596–1609.

20. Grills IS, Hugo G, Chao K, et al. Initial clinical experience withextracranial stereotactic lung radiotherapy: Advantage of conebeam CT image guidance over stereotactic body frame immobi-lization alone [Abstract]. Int J Radiat Onocol Biol Phys 2006;66(Suppl.):S469.

21. Hugo G, Yan D, Liang J. Population and patient-specific targetmargins for 4D adaptive radiotherapy to account for intra- andinter-fraction variation in lung tumor position. Phys Med Biol2007;52:257–274.

22. Ivaldi GB, Wu Q, Liang J, et al. Evaluation of two methods for

soft tissue registration in online CT guided prostate cancer ra-

diotherapy [Abstract]. Int J Radiat Onocol Biol Phys 2004;

60(Suppl.):S331.23. Zhang T, Chi Y, Meldolesi E, Yan D. Automatic delineation of

online head and neck CT images: Towards online adaptive ra-

diotherapy. Int J Radiat Onocol Biol Phys 2007;68:522–530.24. Chi Y, Meldolesi E, Yan D. Verification of deformable organ

registration for prostate using implant markers [Abstract]. IntJ Radiat Onocol Biol Phys 2006;66(Suppl.):S645.

25. Wu Q, Liang J, Ghilezan M, Yan D. A ‘‘Field to Target’’ (F2T)

online correction technique for image guided radiation therapy

[Abstract]. Int J Radiat Onocol Biol Phys 2006;66(Suppl.):S140.26. Yan D, Lockman D, Brabbins D, et al. An off-line strategy for

constructing a patient-specific planning target volume for image

guided adaptive radiotherapy of prostate cancer. Int J RadiatOnocol Biol Phys 2000;48:289–302.

27. Mohan R, Zhang X, Wang H, et al. Use of deformed intensity

distributions for on-line modification of image-guided IMRT

to account for interfractional anatomic changes. Int J RadiatOnocol Biol Phys 2005;61:1258–1266.

28. Sohn M, Birkner M, Yan D, Alber M. Modeling individual geo-

metric variation based on dominant eigenmodes of organ defor-

mation: Implementation and evaluation. Phys Med Biol 2005;

50:5893–5908.29. Yan D, Jaffray D, Wong J. A model to accumulate fractionated

dose in a deforming organ. Int J Radiat Onocol Biol Phys 1999;

44:665–675.30. Yan D, Lockman D. Organ/patient geometric variation in exter-

nal beam radiotherapy and its effects. Med Phys 2001;28:

593–602.31. Yan D. Treatment dose summation and estimation in image

guided adaptive radiotherapy [Abstract]. Int J Radiat OnocolBiol Phys 2006;66(Suppl.):S51.

32. Birkner M, Yan D, Alber M. Adapting inverse planning to pa-

tient and organ geometrical variation: algorithm and implemen-

tation. Med Phys 2003;30:2822–2831.