IMAGE-GUIDED RADIOTHERAPY IN NEAR REAL TIME WITH INTENSITY-
MODULATED RADIOTHERAPY MEGAVOLTAGE TREATMENT BEAM IMAGING
WEIHUA MAO, PH.D.,*yANNIE HSU, PH.D.,* NADEEM RIAZ, M.D.,* LOUIS LEE, PH.D.,*
RODNEY WIERSMA, PH.D.,* GARY LUXTON, PH.D.,* CHRISTOPHER KING, M.D.,* LEI XING, PH.D.,*
AND TIMOTHY SOLBERG, PH.D.y
*Department of Radiation Oncology, Stanford University, Palo Alto, CA; andyDepartment of Radiation Oncology, University of
Texas Southwestern Medical Center, Dallas, TX
Purpose: To utilize image-guided radiotherapy (IGRT) in near real time by obtaining and evaluating the online
positions of implanted fiducials from continuous electronic portal imaging device (EPID) imaging of prostate
intensity-modulated radiotherapy (IMRT) delivery.
implanted fiducial markers are obtained, and their expected two-dimensional (2D) locations in the beam’s-
eye-view (BEV) projection are calculated for each treatment field. During IMRT beam delivery, EPID images
of the megavoltage treatment beam are acquired in cine mode and subsequently analyzed to locate 2D locations
of fiducials in the BEV. Simultaneously, 3D positions are estimated according to the current EPID image, informa-
measured 2D and 3D positions of each fiducial are compared with their expected 2D and 3D setup positions,
respectively. Any displacements larger than a predefined tolerance may cause the treatment system to suspend
the beam delivery and direct the therapists to reposition the patient.
Results: Phantom studies indicate that the accuracyof 2D BEVand 3D tracking arebetter than 1 mm and 1.4 mm,
displacement of 6.7 mm and a maximum 3D displacement of 6.9 mm over 34 fractions.
Conclusions: This EPID-based, real-time IGRT method can be implemented on any external beam machine with
portal imaging capabilities without purchasing any additional equipment, and there is no extra dose delivered to
? 2009 Elsevier Inc.
IGRT, Real-time tracking, Fiducial marker, Prostate, EPID.
Modern conformal radiotherapy techniques, such as three-
dimensional (3D) conformal radiotherapy and intensity-
modulated radiotherapy (IMRT), can provide radiation doses
that closely conform to a tumor volume while sparing
surrounding sensitive structures (1, 2). However, a critical
component that is missing in prostate radiotherapy is real-
time monitoring of intrafraction tumor motion. Several stud-
cm due to bladder and rectum filling and emptying (3–6).
Because of the dynamic nature of human anatomy, image-
guided radiotherapy provides the most benefit when it can
be performed in real time, to ensure an accurate delivery of
the planned conformal dose distribution (7, 8). Kitamura
et al. (9) reported that the amplitude of 3D prostate move-
ment was 0.1–2.7 mm in the supine and 0.4–24 mm in the
prone positions by using multiple kilovoltage (kV) imaging
sets to track implanted markers continuously for 2 min. Xie
et al. (10) evaluated stereoscopic X-ray images of implanted
fiducials during hypofractionated radiotherapy with the
CyberKnife. They found that prostate motion in excess of 2
mm in 30 s exists in approximately 5% of data sets and
that this percentage increases in a larger time interval.
Kupelian et al. (6) reported on continuous electromagnetic
tracking of 35 patients undergoing radiotherapy for prostate
cancer, whereby intrafraction motion of 3 mm and 5 mm
was observed for 30 s or more in 41% and 15% of the frac-
tions, respectively. Compared with using multiple kV imag-
ing sets or electromagnetic system to track fiducials,
megavoltage (MV) treatment beam imaging does not expose
the patient to additional dose and does not require any addi-
tional equipment because most modern treatment machines
Reprint requests to: Weihua Mao, Ph.D., Department of
Radiation Oncology, UT Southwestern Medical Center at Dallas,
5801 Forest Park Road, Room NE3.232, Dallas, TX 75390-9183.
Tel: (214) 645-8565; Fax: (214) 645-2885; E-mail: weihua.mao@
Conflict of interest: none.
Received Jan 19, 2009, and in revised form April 6, 2009.
Accepted for publication April 13, 2009.
Int. J. Radiation Oncology Biol. Phys., Vol. 75, No. 2, pp. 603–610, 2009
Copyright ? 2009 Elsevier Inc.
Printed in the USA. All rights reserved
0360-3016/09/$–see front matter
are equipped with electronic portal imagers. Kotte et al. (11)
obtained portal images at the beginning of every treatment
field in 427 patients and observed intrafraction prostate
motion in excess of 2 mm and 3 mm in 66% and 28% of
11,426 fractions, respectively. Despite these observations,
the standard of care for prostate patients at most institutions
relies on isocenter localization from two orthogonal MV
portal images (12, 13); real-time tumor tracking based on
treatment beam portal imaging has not yet been reported.
We propose a method to analyze portal MV images acquired
continuously throughout a treatment fraction to precisely
tested by phantom studies and then applied retrospectively to
a prostate patient as a feasibility study.
METHODS AND MATERIALS
After initial patient setup using orthogonal AP (anterior–poste-
rior) and lateral portal images, all fiducials on both images were
detected by means of a pattern-matching algorithm (12, 13).
Because every fiducial was accurately located on two orthogonal
projections, their 3D spatial positions were computed by using
a methodology reported previously (12, 13). During IMRT beam
delivery, electronic portal imaging device (EPID) images of the
treatment MV beam were acquired in cine mode and subsequently
analyzed. All visible fiducials were identified automatically using
the pattern-matching algorithm (12, 13). Three-dimensional and
two dimensional (2D) tracking was performed to guide the radio-
therapy in near real time (Fig. 1).
The pseudo-3D position of each fiducial was estimated by using
two or more projections at different gantry angles; the term pseudo-
3D position is used to indicate that the 3Dposition is calculated here
according to two images obtained at different times. Every portal
image gives accurate 2D information but is very insensitive to the
spatial change along the beam direction, owing to the small diver-
gence angle (<5?) of the cone beam. Two or more projections of
the same fiducial at different directions are needed to provide suffi-
cient3Dspatial information.However,simultaneous imagingatdif-
ferent angles is usually unavailable on a conventional linear
accelerator. One could estimate motion along the beam direction
by using information from previous images, including images
from all previous treatment fields and the setup images (as shown
in Fig. 1). Figure 2 illustrates the coordinate systems for the linear
accelerator and EPID imaging system. The room coordinate system
has its origin at the isocenter. The X axis is along the couch lateral
direction (right–left for a supine and head-first patient), the Y axis is
in the couch vertical direction (anterior–posterior/posterior–anterior
for a supine and head-first patient), and the Z axis is along the lon-
gitudinal couch direction (or the patient superior–inferior direction).
Figure 3 shows four portal images obtained at different gantry an-
gles. On the basis of one projection ui;viat a gantry angle of fi
and all previous projections ðuj;vjÞ at other gantry angles fjas
long as jsi, the 3D coordinates can be calculated as:
x ¼ SADSDD$?ujsinðfiÞ ? uisin?fj
y ¼ SADSDD$?uicos?fj
?? ujcosðfiÞ?? uiuj
?þ SDD$?uj? ui
SDD½SAD þ ycosðfiÞ ? xsinðfiÞ?
whereSDD and SAD arethe source-to-detector distance andsource-
to-axis distance, respectively. The detected fiducial 3D position is
Fig. 1. Flow chart of a proposed two-dimensional (2D) and
AP = anterior–posterior; BEV = beam’s eye view.
Fig. 2. Schematic diagram of coordinate systems. VRT = couch verti-
cal direction; AP/PA = anterior–posterior/posterior–anterior direction
for a supine and head-first patient; Lat = couch lateral direction; R/L
= right–left direction for a supine and head-first patient; LNG = couch
longitudinal direction; SI = superior–inferior direction for a supine and
head-first patient; EPID = electronic portal imaging device.
604 I. J. Radiation Oncology d Biology d Physics Volume 75, Number 2, 2009
ment larger than a preset 3D tolerance can be used to trigger the sys-
tem to suspend the beam delivery and inform the therapist to
reposition the patient.
2D beam’s-eye-view tracking
The expected 2D projection locations of fiducials in the beam’s
eye view (BEV) were calculated for every treatment field gantry
angle according to their 3D setup positions. Any fiducial ðx;y;zÞ
is projected to ðui;viÞ on the imager at a gantry angle of fi:
cosðfiÞx þ sinðfiÞy
SAD ? sinðfiÞx þ cosðfiÞy;
SAD ? sinðfiÞx þ cosðfiÞy:
The detected fiducial projection is then compared with the
expected projection in the 2D BEV. Any significant displacement
between detected and expected fiducial projection indicates that
a portion of the planned treatment volume might have moved out
of the treatment field.
To evaluate the reliability and accuracy of this method, a pelvic
phantom was used. Three cylindrical gold fiducials, each with
a diameter of 1.2 mm and a length of 3 mm, were inserted into
the phantom. A typical prostate treatment IMRT plan with seven
fields was delivered to this phantom. The patient couch was inten-
tionally moved by (10, 10, 10) mm, and delivery was repeated,
with images analyzed according to the original initial setup images
delivery without motion indicates the reliability of this method,
whereas the displacement detected in the shifted phantom provides
an estimate of the accuracy of the method.
All experiments were carried out on a Trilogy stereotactic system
(Varian Medical Systems, Palo Alto, CA) with an EPID, which had
a dimension of 1024 ? 768 and an effective pixel size of 0.392 mm.
Setup was performed by comparing fiducial locations on orthogonal
(AP and lateral directions) portal MV image pairs with those on
digitally reconstructed radiographs. Sequential MV imaging was
scheduled, and cine mode portal images were obtained during
irradiation with a SAD of 1000 mm and a SDD of 1500 mm. Geo-
metric calibration had been performed for both kV and MV cone
beam imaging systems to ensure geometric accuracy (14).
Images from a prostate cancer patient who previously had three
gold fiducials implanted in a configuration similar to that of the
phantom study were retrospectively evaluated. An IMRT plan
identical to that used for the phantom study was used to treat
the patient. Imaging and delivery were performed on the Trilogy
system in a manner analogous to that described above. Cine
mode portal images were taken throughout each of the 34 treat-
Figure 4 shows the 3D and 2D BEV results of tracking
a single fiducial in the phantom before and after the couch
shift. Before the couch shift, the maximum 3D displacement
from setup position is approximately 1.5 mm. After the shift
Fig. 3. Sample portal images at different gantry angles as labeled. Fiducials were highlighted in circles.
EPID-based IGRT in near real time d W. MAO et al. 605
are less than approximately 0.6 mm.
Two-dimensional BEV analyses of the same fiducial are
displayed separately in Fig. 4b and c. Before the couch
shift, the detected fiducial locations (‘‘x’’ symbols) are
very close to their expected locations (lines) based on initial
setup results. The maximum 2D displacement from
expected positions is 0.8 mm, whereas the average 2D dis-
placement is approximately 0.3 mm. After the phantom
shift of (10, 10, 10) mm, the 2D displacements vary with
gantry angle owing to the relative orientation. For example,
the displacement in u is approximately 10 mm in the AP
direction (180?gantry angle) and is approximately 1 mm
and 14 mm at gantry angles of 129?and 232?, respectively.
After correcting for the gantry angle effects and comparing
with the actual 3D position (after the shift), the average 2D
displacement is 0.6 mm.
For the prostate cancer patient, 7330 images were acquired
during 34 fractions. All images were analyzed, and all fidu-
cials not obscured by the leaves or jaws were detected. As
a demonstration, results from 2 fractions (Fractions 28 and
12) are presented, because small and large displacements
occurred in Fractions 28 and 12, respectively. Figure 5 illus-
trates the 3D and 2D BEV tracking results for the 2 fractions,
and the resulting histograms are shown in Fig. 6. Two-
dimensional displacements of approximately 1 to 2 mm
and as large as 6 mm are apparent in Fractions 28 and 12,
In addition to performing posttreatment analysis for
a single fraction, multiple-fraction analyses were carried
out in a similar manner. Figure 7 shows the average and max-
imum 2D BEV displacements over all fractions; the average
of the average 3D displacement relative to the initial setup
position; the maximum and average displacements are 6.9
mm and 2.9 mm, respectively. Whereas the agreement in
the lateral (X) and superior–inferior (Z) directions is gener-
that the prostate tends to move posteriorally in the few min-
utes between initial setup and the commencement of treat-
ment. This was a common observation through out most of
the 34 fractions. Figure 8 compares average 3D positions
of one fiducial during treatment with its setup 3D positions
in all fractions. Significant displacements occur in the Y
(AP) directions occasionally. This posterior drift, often cou-
pled with superior–inferior drift, is consistent with the obser-
vations of other research groups (6, 9, 10, 15). Additionally,
prostate motion is highly patient dependent, and the motion
of the same patient may be different and unpredictable
from day to day.
Although the pseudo-3D position is calculated according
to two images obtained at different times, we claim that the
pseudo-3D position is accurate on the basis of two facts:
(1) the fiducials in all previous images and the current image
are close to their setup positions because any significant
variation from setup position would result in stopping the
treatment and repositioning the patient, and (2) results from
all previous images are used to estimate the fiducial 3D
position based on the current image. Every set of pseudo-
3D positions at any time point is calculated according to
the current image and all previous images, usually dozens
or even hundreds of images, so that the errors due to random
small motions are statistically small.
It is true that the tumor may move or deform between the
two images used for a set of 3D position calculations. All 3D
positions were calculated according to the current image and
one image from a previous field or a setup field. Depending
on the availability of previous fields and setup fields, various
sets of 3D positions were calculated for the same time point.
If the fiducial has moved after initial setup, images of the
Fig. 4. Two-dimensional beam’s-eye-view and three-dimensional
results for a single fiducial implanted in a phantom before and after
couch shift of (10, 10, 10) mm. (a) Three-dimensional coordinates;
(b) u coordinates; (c) v coordinates. Lines are expected fiducial
locations based on setup position. Symbols are detected fiducial
before and after couch shift. Results at different treatment fields
are separated and marked by gantry angles.
606 I. J. Radiation Oncology d Biology d Physics Volume 75, Number 2, 2009
fiducial at different locations will be used for the calculation,
so that different sets of results may be significantly different.
As shown in Fig. 5f, the Y coordinate oscillates between two
positions, indicating that the fiducial moved after setup. It
should be noted that with the proper implementation this
would be prevented from occurring in an actual treatment.
Although it is known that the prostate moved in the poste-
rior–anterior direction after setup, the motion can not be
caught in the first treatment field at a gantry angle of 26?
because of the small 2D BEV displacements (Fig. 5b).
However, the 2D BEV displacement is large enough in the
second treatmentfieldatagantryangle of77?. Thistreatment
would be suspended immediately and the patient reposi-
tioned in an actual image-guided procedure.
Another concern isthat the method isinsensitive tomotion
along the beam direction, for example, in the first treatment
field in Fig. 4. As shown in Fig. 9, motion along the beam
direction only leads to limited effects on the dose delivery
to the target and surrounding sensitive structures. For exam-
ple, a patient displacement of 1 cm along the beam direction
leads to an approximately 2% change in X-ray intensity due
change in the field size due to small beam divergence. Even
for a large treatment field with a size of 10 cm, the field size
difference may be only 1 mm owing to the displacement of
1 cm along the beam direction. Providing the 2D BEV
displacements are small, there is no clinical significance to
motion along the beam direction.
Another challenge is that is it possible for a portion of the
dose to have been delivered in an IMRT field while all fidu-
cials are blocked by the multileaf collimator (MLC), with the
motion undetected until some fiducials are detected later in
the field. This can be compensated for by having the largest
segment delivered first in each IMRT treatment field (11),
so that any significant interbeam motion can be detected
before a significant portion of the dose is delivered.
Fig. 5. Patient results on two-dimensional beam’s-eye-view in 2 fractions. (a) u coordinates of three fiducials in Fraction
28; (b) u coordinates of three fiducials in Fraction 12; (c) v coordinates of 3 fiducials in Fraction 28; (d) v coordinates of
three fiducials in Fraction 12; (e) three-dimensional coordinates of Fiducial 2 in Fraction 28; (f) three-dimensional coor-
dinates of Fiducial 2 in Fraction 12. Lines are expected fiducial locations based on setup position. Symbols are detected
fiducial locations. Results at different fields are separated and marked by gantry angles.
EPID-based IGRT in near real time d W. MAO et al. 607
A more general problem is that there will always be
segments where fiducials are blocked partially or completely
by the MLC in many IMRT plans. This makes the reported
information into consideration during the IMRT inverse plan-
ning process. With the development of segment-based dose
optimization methods (16, 17), it should be feasible to ensure
process. Of course, the addition of this type of constraint in
inverse planning may compromise the achievable dose distri-
bution.Any tradeoff willlikely beminimal,however,because
the fiducials are all inside the tumor target volume and repre-
sent high-dose regions. We are currently studying this issue,
and the results will be reported elsewhere. Continuous tumor
tracking in real time is possible in the future.
In dynamic MLC delivery, the ‘‘sliding’’ portion of the
field may lead to some blurring of the image when an
EPID ‘‘catches’’ the sliding in action. This may degrade
the image quality and make the fiducial detection more diffi-
culty. This is particularly true in the region close to a field
boundary. For fiducials located at the center of a segment,
however, there is no difference between a dynamic MLC
and a step-and-shoot MLC technique.
Although this study has been performed retrospectively,
this method of evaluating 2D BEV and pseudo-3D positions
in near real time is ready for online application, providing
manufacturers can provide real-time access to the EPID
Fig. 6. Two-dimensional (2D) beam’s-eye-view and three-dimensional (3D) displacement analysis for 2 fractions. (a, c)
Fraction 28; (b, d) Fraction 12.
Fig. 7. Average (Mean) and maximum (Max) two-dimensional
beam’s-eye-view (2D BEV) displacement analysis results for all
608 I. J. Radiation Oncology d Biology d PhysicsVolume 75, Number 2, 2009
image data. Until such a time, an alternative method of
accessing the portal image through a video capture card has
been reported (18). In addition, fully automated analysis
software is capable of tracking and analysis in real time.
This tracking strategy is applicable to any conventional
linear accelerator with a portal imager. Although the Trilogy
yses were also carried out on a Clinac 21EX (Varian Medical
Systems). Access to a kV imager is helpful in assisting initial
patient setup and can provide complementary information
but is not required for the process. Unlike kV imaging, no
extra dose is delivered to the patient. Additionally, applica-
tion of this tracking method in the clinic does not require
the purchase of additional equipment.
Finally, this method is not limited to tracking prostate
motion: any target with implanted fiducial markers, includ-
ing certain types of surgical clips, can be tracked similarly.
Application to tracking pancreas and liver tumors is in
A fully automated method to track prostate movement
in near real time has been proposed and evaluated. It can
provide accurate 2D tumor positions at every treatment field
through the entire treatment process and can provide accurate
3D positions. This makes online monitoring and offline dose
calculation more reliable. This method should also be appli-
cable to other anatomic sites. In addition, this method utilizes
information ignored previously, and it monitors prostate
movement during treatment without extra cost. It does not
require additional equipment, and it does not expose the
patient to additional radiation. This method can be easily
applied to track prostate and other tumor motions in most
1. IMRT Collaborative Working Group. Intensity-modulated
radiotherapy: Current status and issues of interest. Int J Radiat
Oncol Biol Phys 2001;51:880–914.
2. Galvin JM,Ezzell G,EisbrauchA,etal.Implementing IMRTin
clinical practice: A joint document of the American Society for
Therapeutic Radiology and Oncology and the American Asso-
ciation of Physicists in Medicine. Int J Radiat Oncol Biol
3. Balter JM, Sandler HM, Lam K, et al. Measurement of prostate
movement over the course of routine radiotherapy using im-
planted markers. Int J Radiat Oncol Biol Phys 1995;31:
4. Herman MG, Pisansky TM, Kruse JJ, et al. Technical aspects of
daily online positioning of the prostate for three-dimensional
conformal radiotherapy using an electronic portal imaging
device. Int J Radiat Oncol Biol Phys 2003;57:1131–1140.
5. Ghilezan MJ, Jaffray DA, Siewerdsen JH, et al. Prostate
gland motion assessed with cine-magnetic resonance imag-
ing (cine-MRI). Int J Radiat Oncol Biol Phys 2005;62:
6. Kupelian P, Willoughby T, Mahadevan A, et al. Multi-institu-
tional clinical experience with the Calypso System in localiza-
tion and continuous, real-time monitoring of the prostate
gland during external radiotherapy. Int J Radiat Oncol Biol
8. Xing L, Thorndyke B, Schreibmann E, et al. Overview of
9. Kitamura K, Shirato H, Seppenwoolde Y, et al. Three-dimen-
sional intrafractional movement of prostate measured during
real-time tumor-tracking radiotherapy in supine and prone treat-
ment positions. Int J Radiat Oncol Biol Phys 2002;53:
10. Xie Y, Djajaputra D, King CR, et al. Intrafractional motion of
the prostate during hypofractionated radiotherapy. Int J Radiat
Oncol Biol Phys 2008;72:236–246.
11. Kotte AN, Hofman P, Lagendijk JJ, et al. Intrafraction motion
oftheprostate duringexternal-beam radiation therapy:Analysis
Fig. 9. Schematic diagram illustrates a displacement of 1 cm along
beam direction. SDD = source-to-detector distance; SAD = source-
Fig. 8. Average three-dimensional positions (Tx Avg, hollow sym-
bols) of one fiducial during treatment in comparison with its setup
three-dimensional positions (Setup, solid symbols) for all sessions.
EPID-based IGRT in near real time d W. MAO et al. 609
of 427 patients with implanted fiducial markers. Int J Radiat Download full-text
Oncol Biol Phys 2007;69:419–425.
12. Mao W, Riaz N, Lee L, et al. A fiducial detection algorithm for
real-time image guided IMRT based on simultaneous MV and
kV imaging. Med Phys 2008;35:3554–3564.
13. Mao W, Wiersma RD, Xing L. Fast internal marker tracking
algorithm for onboard MV and kV imaging systems. Med
14. Mao W, Lee L, Xing L. Development of a QA phantom and
automated analysis tool for geometric quality assurance of
on-board MV and kV x-ray imaging systems. Med Phys
15. Langen KM, Willoughby TR, Meeks SL, et al. Observa-
tions on real-time prostate gland motion using electromag-
netic tracking. Int J Radiat Oncol Biol Phys 2008;71:
16. Cotrutz C, Xing L. Segment-based dose optimization using
a genetic algorithm. Phys Med Biol 2003;48:2987–2998.
17. Shepard DM, Earl MA, Li XA, et al. Direct aperture optimiza-
tion: A turnkey solution for step-and-shoot IMRT. Med Phys
18. Wiersma RD, Mao W, Xing L. Combined kV and MV imaging
for real-time tracking of implanted fiducial markers. Med Phys
610 I. J. Radiation Oncology d Biology d Physics Volume 75, Number 2, 2009