You are on page 1of 10

Int. J. Radiation Oncology Biol. Phys., Vol. 64, No. 4, pp.

12651274, 2006
Copyright 2006 Elsevier Inc.
Printed in the USA. All rights reserved
0360-3016/06/$see front matter

doi:10.1016/j.ijrobp.2005.11.008

PHYSICS CONTRIBUTION

CLINICAL EXPERIENCE WITH A 3D SURFACE PATIENT SETUP SYSTEM


FOR ALIGNMENT OF PARTIAL-BREAST IRRADIATION PATIENTS
CHRISTOPH BERT, M.S.,* KATHERINE G. METHEANY, B.S., KAREN P. DOPPKE, M.S.,
ALPHONSE G. TAGHIAN, M.D., PH.D., SIMON N. POWELL, M.D., PH.D.,

AND GEORGE T.Y. CHEN, PH.D.


*Abteilung Biophysik, Gesellschaft fr Schwerionenforschung, Darmstadt, Germany; and Department of Radiation Oncology,
Massachusetts General Hospital and Harvard Medical School, Boston, MA
Purpose: To assess the utility of surface imaging on patient setup for accelerated partial-breast irradiation
(APBI).
Methods and Material: A photogrammetry system was used in parallel to APBI setup by laser and portal
imaging. Surface data were acquired after laser and port-film setup for 9 patients. Surfaces were analyzed in
comparison to a reference surface from the first treatment session by use of rigid transformations. The surface
model after laser setup was used in a simulated photogrammetry setup procedure. In addition, breathing data
were acquired by surface acquisition at a frame rate of 7 Hz.
Results: Mean 3D displacement was 7.3 mm (SD, 4.4 mm) and 7.6 mm (SD, 4.2 mm) for laser and port film,
respectively. Simulated setup with the photogrammetry system yielded mean displacement of 1 mm (SD, 1.2 mm).
Distance analysis resulted in mean distances of 3.7 mm (SD, 4.9 mm), 4.3 mm (SD, 5.6 mm), and 1.6 mm (SD, 2.4
mm) for laser, port film, and photogrammetry, respectively. Breathing motion at isocenter was smaller than 3.7
mm, with a mean of 1.9 mm (SD, 1.1 mm).
Conclusions: Surface imaging for PBI setup appears promising. Alignment of the 3D breast surface achieved by
stereo-photogrammetry shows greater breast topology congruence than when patients are set up by laser or portal
imaging. A correlation of breast surface and CTV must be quantitatively established. 2006 Elsevier Inc.
Partial-breast irradiation setup, Surface imaging, Radiotherapy.

INTRODUCTION

Video-based patient setup has been used for patient setup


for a number of years (4 12). Multiple approaches are used
to obtain data on patient position at various treatment sites.
Not all techniques make use of 3D information. Johnson et
al. (10) describe the use of 2D video imaging with orthogonal cameras. On the basis of difference images between a
reference and a daily setup image, the mispositioning can be
determined and, if necessary, adjusted. This 2D system has
been used to analyze setup for head-and-neck treatments
(11). Three-dimensional optical acquisition can be facilitated by photogrammetry and laser interferometry. In laser
interferometry, fringes are projected on the patients skin
from which a height map can be generated. Moore et al. (5)
have applied this technique clinically to set up rectal cancer
patients and measure intratreatment changes. MacKay et al.

Accelerated partial-breast irradiation (APBI) by use of 3D


conformal external-beam radiation is a new technique that
has been recently suggested for selected patients with early
breast cancer (13). In this approach, only the resection site
is treated as clinical target volume (CTV), with appropriate
margins to generate the planning target volume (PTV). This
method requires more precise positioning than for conventional whole-breast treatment, particularly because the total
dose is usually delivered in 6 to 10 fractions. To facilitate
this treatment, portal images are taken before each treatment session to guide setup by alignment of the chest
wall in the portal images to digitally reconstructed radiographs (DRRs).

pists, especially Kathy Bruce and Gidget Manning. Special thanks to


Sashi Kollipara (MGH), Eike Rietzel, Ph.D. (MGH), Norman
Smith, Ph.D. and Ivan Meir, Ph.D. (VisionRT) for software tools
and insightful discussions. KGM acknowledges the Center for
Subsurface Sensing and Imaging Systems (CenSSIS) at Northeastern University.
Received Aug 8, 2005, and in revised form Nov 2, 2005.
Accepted for publication Nov 3, 2005.

Reprint requests to: Christoph Bert, M.S., GSI, Biophysik,


Planckstr. 1, 64291 Darmstadt, Germany. Tel: (49) 6159-712947; Fax: (49) 6159-71-2106; E-mail: c.bert@gsi.de
This paper was presented at the 46th Annual Meeting of the
American Society for Therapeutic Radiology and Oncology
(ASTRO), October 37, 2004, Atlanta, GA.
C.B. received a scholarship (Doktorandenkurzstipendium) from
the German Academic Exchange Service (DAAD).
AcknowledgmentThe authors acknowledge the help of the thera1265

1266

I. J. Radiation Oncology

Biology Physics

(12) used such a technique for prostate setup. The principles


of photogrammetry are described in the literature (13, 14).
Briefly, the position of markers placed on the patients skin
(4, 6, 7) or the entire skin surface can be measured by
optical methods (15). Rogus et al. (6) focused on a headand-neck alignment method, in which markers are attached
to the immobilization mask. Soete et al. (7) used the ExacTrac system (BrainLAB AG, Heimstetten, Germany) to
assess positioning of prostate cancer patients. Baroni et
al. (16) analyzed the utility of photogrammetry for breast
setup and attached multiple markers to the breast and
abdomen (to measure the breathing cycle and enable
gated video capture).
A commercially available 3D surface-imaging system
(AlignRT; Vision RT, London, UK) was installed at Massachusetts General Hospital (MGH) in the treatment room
to facilitate image-guided breast radiotherapy. The imaging
system is designed as a patient-setup device (classified as a
Class II device; 510 (k) clearance received from FDA) and
calculates the couch coordinate adjustment necessary to
register the breast surface model (SM) taken before each
treatment with a reference surface model (SMR).
We acquired and analyzed surface-model data of patients
undergoing partial-breast irradiation (PBI) to assess the
potential utility of close-range photogrammetry 3D surface
imaging in target alignment for PBI. The analysis focuses
on (1) a comparison of conventional and surface techniques
for patient setup on the basis of breast surface models, (2)
quantification of respiration during breast setup, and (3)
how reference surface models may be generated.
METHODS AND MATERIALS
Treatment planning and patient setup
This protocol is an institutional approved protocol, and all
patients consented to the treatment. All patients underwent CTbased simulation with immobilization on a breast board. The
clinical breast borders and lumpectomy scar were marked with
radiopaque catheters. CT images were obtained at 1.5-mm to 3-mm
intervals from the level of the mandible through the lung bases. The
lumpectomy cavity, ipsilateral breast, ipsilateral and contralateral
lungs, and heart were contoured. The CTV was assumed to be the
tissue within 1 or 1.5 cm of the excision cavity (chosen at the
discretion of the treating physician), except for anteriorly and
posteriorly, where it was limited by the skin and anterior chest wall
or pectoralis muscles. An additional 0.5 cm was felt adequate to
allow for setup error, patient motion, and respiration to create the
PTV. The PTV was generated by expansion of the excision cavity
in all directions by 1.5 to 2.0 cm; this PTV was then edited so that
it came no closer than 5 mm to the skin surface and no deeper than
the anterior chest wall/pectoralis muscles. Additional margin (usually 0.7 cm) was added to the field size to account for beam
penumbra.
Beam arrangements were left to the discretion of the treating
physician. Any combination of photon beams of 4 MV or higher
and electrons was permitted, so long as a minimum of 95% of the
PTV received 100% or more of the prescribed dose. The prescribed dose was 32 Gy, given in 4-Gy fractions twice daily, with
a minimum interfraction interval of 6 hours, over a maximum

Volume 64, Number 4, 2006

elapsed time of 1 week. The patient setup is initially defined by


tattoos marked at the time of CT and denoted by radiopaque bbs
during the CT scan. Three tattoos on the abdomen define rotation
about the longitudinal axis. Two tattoos, one on the sternum and
one lateral to the sternum, define CT isocenter. The sternum tattoo
also provides a point at which the SSD can be measured.
To align the patient to isocenter, a shift is applied with the
respect to the CT isocenter. Patient alignment is verified by AP and
lateral port films, taken at each treatment session. The films are
compared visually to DRRs reconstructed from the treatmentplanning CT. If needed, the patients position is adjusted (typically
done if a mispositioning is greater than 3 mm in one dimension).
Positional shifts of 5 mm or more are subject to verification by an
additional set of portal images after shift.

3D surface acquisition
The 3D surface-imaging setup analysis was performed on 9
patients over a total of 53 fractions (from 4 to 8 fractions per
patient). Patients were informed and gave written consent. Seven
patients had lesions in the left breast, and 2 patients had lesions in
the right breast. CTV volumes ranged from 10.8 to 88.4 cc, with a
mean standard deviation of 34.8 24.5 cc; breast volumes
ranged between 268.7 and 1,032.0 cc, with a mean standard
deviation of 672.8 275.7 cc. The patients were between 38 and
91 years of age, with a median age of 65 years.
The system accuracy was established by phantom studies and
reported in detail elsewhere (17). A brief summary of the process
is described as follows: surface information is obtained in approximately 10 seconds via 2 camera pods suspended from the treatment room ceiling. Each pod is equipped with a stereovision
camera (2 CCD cameras separated by a known baseline), a texture
camera, a clear flash, a flash used for speckle projection, and a
slide projector for speckle projection. Speckle refers to an optically
projected pseudorandom gray-scale pattern to enable 3D reconstruction of the surface. The setup in the treatment room is shown
in Fig. 1. Each pod acquires 3D surface data over approximately
120 in the axial plane, from midline to posterior flank. On the
basis of a proprietary calibration process, the data are merged to
form a single 3D surface image of the patient. In the overlap region
near midline, the surfaces from the 2 pods merge smoothly with
less than a 1-mm RMS discontinuity. The consistency of this
overlap region is recommended by the vendor to be checked by a
daily calibration verification procedure. This verification step is
easily performed during the linac warm-up period.
The surface model is defined by 3D vertices (vertex spacing was
set to 6 mm at MGH), which form the triangular surface faces (Fig.
2a). Each face is colored with a gray-level video image of the skin
surface acquired from the texture cameras. Typical surface images
are shown in Figs. 2b and 2c.
The system includes software designed to facilitate patient
setup, principally by surface-model acquisition and alignment by
surface matching with a reference. In the clinical workflow, a
reference surface model (SMR) is first acquired. This acquisition
can be done at the time of first treatment session on the linac, in the
simulator room by use of a second imaging system with appropriate intersystem calibration, or by extraction of the reference surface from CT-scan data. At subsequent treatment sessions, the 3D
system can then be used as an alignment tool by comparison of the
SMR with a surface model acquired after initial laser setup (SML) or
any other surface-model acquisition. During the alignment process,
which takes approximately 10 seconds (17), the software calculates
the optimal rigid-body transformation (couch translation and rota-

3D surface imaging for PBI setup

C. BERT et al.

1267

Fig. 1. Schematic outline of the surface-imaging system setup in the treatment room.

tion) that brings the SML of the daily treatment fraction into
congruence with the reference surface within a user-defined region
of interest (ROI). The use of an ROI excludes extraneous surface
parts that are not relevant to alignment of the breast to be treated
(e.g., gown). Alignment accuracy is better than 0.8 mm and 0.1
RMS, as shown in phantom studies (17). The system evaluation
used a phantom that assumes the patient is a rigid body. In some
sites, this assumption is appropriate; for other sites, breathing
motion or tissue deformation may alter surface topology.
To minimize breathing-motion artifacts, the system can be used
in a gated-acquisition mode. In this mode, the system monitors
patient respiration for several seconds at a frame rate of 7 Hz,
during which a sequence of surface images is acquired (50
frames). A SM is calculated for one of the frames at a userspecified respiratory phase. Phantom studies on a moving surface
indicate that the system accuracy in determination of amplitude is
better than 0.15 mm RMS (17). Data reported in this study were
acquired at random respiratory phase. We did not use gated acquisition, which was only available toward the end of the study.

Imaging protocol
To assess the performance of the system as a patient-alignment
device, 3D surface models were acquired in parallel to conven-

tional setup and treatment of 9 PBI patients. Because the technology does not use ionizing radiation, multiple surface models were
obtained during each treatment session. The reference surface
model SMR was selected to be the surface model acquired at the
first treatment fraction after portal film alignment. For the remaining fractions, surface data were acquired after alignment by
lasers (SML) and after portal-imaging alignment (treatment
surface model SMT). During development of the portal films, a
dynamic breathing study was recorded to quantify surface motion during light respiration. Figure 3 shows the protocol schematically.

Couch shift analysis


The goal of the 3D surface-imaging process is to provide reproducible breast positioning for PBI. For each treatment fraction,
the system software calculates couch coordinate shifts needed to
align a SM to the SMR within a user-defined ROI. The calculation
is based on surface topology only; texture information is not used.
To analyze the setup accuracy and reproducibility of the modalities
laser, port film, and surface during the course of treatment, measured surface models (e.g., SMT) were aligned offline to the
patients SMR, which resulted in a set of recommended couch
shifts for each model. A typical ROI is shown in Figs. 2b and 2c

Fig. 2. Surface models of a left female breast. (a) Surfaces are created by triangular tiles with approximately 6 mm of
spacing between intersections (the cube of size 1 cm3 is added for comparison). Texture is achieved for part of this
model by coloring each tile with the appropriate part of a bitmap image. Because of the three-dimensional
representation, each model can be viewed from any direction. For example, a textured patient-surface model is viewed
from (b) frontal and (c) left lateral. The thick black lines enclose the ROI chosen for surface registration. White surface
tiles are caused by missing data, for example, because the breast blocks one camera view (c). The texture data allow
visualization of skin marks, scar, and nipple. In all images, the thin black lines represent virtual lasers.

1268

I. J. Radiation Oncology

Biology Physics

Fig. 3. Schematic outline of the imaging protocol. The protocol in


clinical use is in solid-line boxes, the surface images are in
dashed-line boxes, and the virtual protocol for setup by surface
imaging is in the circles.

(this ROI comprises approximately 2,300 surface tiles) and is


defined to encompass the treated breast completely.
Three dimensional alignment by use of the rigid-body transformation calculated by the system could not be measured clinically.
The proposed couch shift could not be applied, because the system
was used in parallel with the conventional positioning protocol.
Instead, a virtual alignment was performed (circles in Fig. 3), by
transformation of the SML according to the transformation parameters obtained by comparison with the SMR. To be consistent with the degrees of freedom on a standard linac couch, only
couch translation and rotation about the vertical axis were used
for transformation. This limitation is also considered in the
registration algorithm. The transformation was applied by transformation of each 3D surface vertex; the connections of the triangles were retained (Fig. 4a). Texture data could not be transformed. The resulting transformed surface (virtual 3D alignment
surface model, SMV) was analyzed as the laser/treatment surface
model (SML/T) (i.e., aligned to the SMR that yields the couch shifts
required for alignment.)

Volume 64, Number 4, 2006

is farther from isocenter than the reference surface. The mean,


absolute mean, and standard deviation of the distance measurements within the ROI were computed. In addition, distance histograms were calculated (1212 mm, 100 bins) as part of the
analysis protocol.
The visualization of distances between 2 surfaces was achieved
in 3 ways: (1) The distance over the ROI can be projected onto the
textured reference surface as a color-wash representation as shown
in Fig. 5a. The green areas represent alignment within 2 mm, blue
areas represent distances between surfaces that range from 2 to 4
mm, red areas represent distances between surfaces that range
from 4 to 6 mm, and yellow areas represent distances between
surfaces that are greater than 6 mm, with a cutoff of 2 standard
deviations. The shades of color encode positive and negative
values for the magnitude of distance. All other surface tiles are
colored gray. Underneath the color are the texture data. This
visualization is useful in identifying surface landmarks such as
nipple or scar, even though they are sometimes obstructed by color
changes in that area. (2) By automated analysis, axial and sagittal
body contours through isocenter and at planes 40 mm from
isocenter were computed for each SM in comparison to the SMR.
Contour plots permit classic evaluation of alignment accuracy in a
specified plane and are helpful in understanding distance distributions. (3) Overlaid images or movie loops of several SMs proved
to be useful when patient setup is compared over several fractions.

Quantifying respiratory motion


Three-dimensional surface models can be affected by respiration, and, thus, breathing motion must be considered during the
alignment process, if clinically relevant. If surfaces move significantly, gated acquisition of the surface model would be advisable.
Assessment of the influence of respiration on surface acquisition
was possible by postprocessing data from continuous-mode surface capture, which was available for all patients. Patients were not
instructed to breathe in a specific pattern; data were acquired
during normal light respiration.
The sequence of surface models acquired during a continuous-

Distance measurements
In addition to the software provided by the commercial system,
software was developed in-house to analyze system performance.
These programs were written in C by use of the public-domain
VTK libraries (18). A prerequisite is the ability to import patient
surface data, which was straightforward because all surfaces are
stored in a graphical standard format (Wavefront OBJ format,
http://www.dcs.ed.ac.uk/home/mxr/gfx/3 d/OBJ.spec). For each
patient, an automated analysis was performed. To evaluate the
characteristics of a surface match between surfaces, two general
approaches proved useful, quantitative analysis and visualization.
For quantitative analysis, the distance between SMR and a
second SM within the entire ROI was calculated (Fig. 4b). This
distance metric differs from the proprietary commercial metric
used in the registration cost function. Our distance function is
defined as the distance from the SMR to the SM along the normal
of the reference surface and measured at each vertex of the
reference surface. A positive distance value means the test surface

Fig. 4. (a) Transformation of the 3D vertices xLi of the surface


model after laser setup (SML) into the vertices xV
i yields the virtual
surface of 3D surface-imaging setup (SMV). The transformation
parameters Rz and tx,y,z are the parameters the surface-imaging
software calculates as required couch shift. (b) Distance measurement from a reference surface SR (black) to a test surface
ST (gray). The visible part of the surfaces is in solid lines, and
covered surface parts are in dotted lines. The transition visible/
covered is indicated by the dash-dotted line. Distances are measured at surface-tile intersections of SR normal to SR to intersection
with the ST. Positive measurements (vector p) encode intersection
with ST further away from isocenter (IS) than SR, and negative
measurements (vector n) refer to ST closer to isocenter.

3D surface imaging for PBI setup

C. BERT et al.

1269

Fig. 5. (a) Distance data for 4 fractions and 3 modalities in color-wash representation for a typical patient. Underneath
the color wash is the textured reference surface model from the first fraction. The white lines represent virtual lasers,
and the white crosses indicate positions for distance measurements obtained by use of the axial contours through
isocenter shown in (b). A quick congruence check with respect to the reference surface model is possible if the distance
data are presented in the form of histograms (c).
mode surface capture in principle permits motion analysis at any
surface point. This feature is useful for a qualitative analysis, but
quantitatively, too much data are generated. Distance maps were
used for visualization of the qualitative analysis. They provide an
overview of surface motion during respiration. The distance is
calculated (relative to the first frame) for each of the following 35
to 50 frames of a breathing capture sequence. At each vertex of
the first frame, the mean, absolute mean, and standard deviation

of the distance measurements are calculated, and typically, the


absolute mean is projected onto the surface of the first frame. This
value represents the mean motion amplitude. For sinusoidal motion, it is 2/ of the amplitude.
Quantitative measurement of respiratory-induced surface motion can be performed offline by use of the gated-capture mode of
the control software. The software allows measurement of the
motion amplitude along the normal of the first frame as a function

1270

I. J. Radiation Oncology

Biology Physics

Volume 64, Number 4, 2006

Fig. 6. (a) Mean distance to the first frame of a typical breathing sequence (50 frames at 7 Hz). For the more quantitative
analysis, the distance values were measured at or in the vicinity of the indicated positions (R). (b) Overlay of 8 surface
models in treatment position (SMT). The arm-position change can be seen in this representation.

of time at any specified surface point. Breathing motion is assessed


at treatment isocenter and at a point in the abdomen (Fig. 6a.
Because of missing surface patches or view obstructed by clothing,
identification of the same coordinates for all treatment fractions of
a specific patient frequently was not possible. In such cases, the
closest possible point was chosen.

RESULTS
Recommended couch shifts
The patient-alignment procedure provides the therapist
with couch shifts to bring the surfaces in an ROI into
alignment. These shifts are based on a rigid-body transformation that minimizes the distance between the SM acquired for setup and the SMR. In off-line analysis, recommended shifts were calculated for all SML, SMT, and SMV
acquisitions. Data from 9 patients and 44 fractions (plus 9 as
reference) are combined in Fig. 7. For each degree of
freedom, the mean, standard deviation, maximum, and minimum of recommended shift are plotted. The resulting 3D
displacements are summarized in Table 1.

Table 1. Three-dimensional displacement (in mm) as


recommended by the alignment procedure
Surface model
Fig. 7. Mean standard deviation with minimum () and maximum () of the couch shift required to bring the corresponding
surface model back to reference. Data are from 9 patients, and 44
fractions were analyzed.

Laser
Treatment
Virtual 3D
alignment

Mean

Standard
deviation

Minimum

Maximum

7.3
7.6

4.4
4.2

1
1.7

17.6
19.3

1.2

4.2

3D surface imaging for PBI setup

C. BERT et al.

1271

Table 2. Statistical analysis of distance measurements for the patient data shown in Figure 6
Mean (mm)

Absolute mean (mm)

Standard deviation (mm)

Fraction

Laser

Portal

3D

Laser

Portal

3D

Laser

Portal

3D

4
6
7
8

0.55
4.64
3.54
0.06

1.99
5.28
1.35
4.50

0.22
0.52
0.38
0.32

1.15
6.16
4.33
1.95

2.53
5.42
2.08
4.95

1.10
2.29
1.45
1.01

3.80
5.99
3.85
3.10

4.77
4.82
3.41
4.12

3.59
4.86
3.11
2.39

Distance measurements
An alternative analysis to recommended couch transformations involves distance calculations, where distances between SMR and SMT at triangle nodes are calculated. Distance maps, distance histograms, and 6 principle plane
contours were generated for all fractions imaged. The volume of image data is too large to be presented here. Instead,
data from a typical patient and combined results from all
patients are presented.
The SMR acquired at the first treatment fraction was used
as the basis of distance-measurement analysis. Distances
were measured over the ROI that included the treated breast.
Figure 5a shows the color-coded distance maps for a typical
patient (5 fractions acquired). A color wash overlays the
textured SMR. For the subsequent 4 fractions, images are
shown after laser alignment, port-film alignment, and virtual
3D surface setup relative to SMR. Each image contains
virtual crosshairs (white lines) to provide a general orientation. Both laser and port-film alignment lead to SMs that
show mispositioning relative to SMR by amounts greater
than 6 mm for some of the fractions (e.g., laser Fractions 6
and 7, port-film Fractions 6 and 8) but can also lead to very
good agreement (e.g., laser Fraction 4). The virtual setup
based on 3D surface matching, after virtual corrections in
couch translation and rotation about the vertical axis, mainly
led to surface congruence within less than 2 mm. The results
for this specific patient are typical for most other patients
analyzed.
Contour representations of patient position are useful for
understanding the distances calculated. Figure 5b shows
axial contours through isocenter of Fraction 7. The arrows
are approximately at the position of the white crosses indicated in Fig. 5a. Distance measurements are performed
normal to the reference surface and, therefore, normal to the
reference contour of Fig. 5b. The measurements show that,
for example, the yellow area in the laser-aligned surface is
slightly above 6 mm at S-I isocenter.
Another method of assessing surface-data congruence is
through distance histograms. Figure 5c displays the same
information as the color-coded images but yields quick
access to quantitative information of the alignment quality.
For example, the comparison of Fractions 4 and 7 for
port-film alignment shows that both are aligned better than
approximately 4 mm, that the Fraction 4 SMT is closer to
isocenter than is the SMR (negative values) in contrast to the
Fraction 7 SMT. Table 2 quantifies the statistical analysis of
the distance measurement.

The distance data from all patients and all fractions can
be globally combined to assess the quality of alignment
methods, as shown in Fig. 8. Laser-guided and port-film
guided alignment yield similar distance histograms. Absolute
mean standard deviations are 3.7 4.9 mm and 4.3 5.6
mm for laser and port-film based alignment, respectively.
Deviations greater than 5 mm were observed in 27.5% of the
vertices for laser alignment and 32.2% for portal-film alignment. In contrast, virtual alignment by 3D surface imaging
produces a narrower distance histogram, with 1.6 2.4 mm
absolute mean standard deviation and 5.0% of the vertices with greater than 5-mm deviation. In this statistical
analysis, the distance difference was cut off at greater than
30 mm to eliminate the effects of bad data, which can
appear when the SM exhibits a blank patch, and the distance
is measured from the SMR through the hole in the SM to the
patients gown further away.
Respiratory motion
Breathing can influence patient setup by introducing a
variation as large as the breathing amplitude of the surface
used. To analyze the effects of breathing motion, the measured-sequence raw data were used to extract breathing
amplitudes at treatment isocenter and at the abdomen (Fig.
6a). The number of fractions with breathing data are not
equally distributed among the total number of patients measured. The results are summarized in Table 3. At isocenter,
the peak-to-peak motion was 1.9 1.1 mm combined for

Fig. 8. The combined histograms from distance data of 9 patients


for the modalities laser, port film, and 3D surface alignment.

1272

I. J. Radiation Oncology

Biology Physics

Table 3. Analysis of the respiratory motion peak-to-peak


amplitude at isocenter and abdomen

Mean (mm)
Standard deviation (mm)
Minimum (mm)
Maximum (mm)
Number of patients
Number of measurements

Isocenter

Abdomen

1.9
1.1
0.6
4.4
7
37

5.7
1.3
3.4
8.8
6
29

all patients, with a range from 0.9 0.3 mm to 3.7 0.5


mm for the 7 patients (combined for all patient fractions).
An example for qualitative results is shown in Fig. 6a. The
enhanced motion at the abdomen can clearly be detected.
DISCUSSION
Partial-breast irradiation patient setup is more demanding
than setup for whole-breast irradiation because of the reduced field size (6 8 cm2 vs. 16 22 cm2) in
combination with the nonisotropic margin extension of the
CTV. To implement some level of image guidance in PBI
setup, we currently use AP and lateral portal films to align
the chest wall. Film processing requires approximately 3
minutes. During this interval, some patients can move considerably.
Furthermore, a visual alignment of portal film and DRR
is dependent on the experience and judgment of the radiation
technologist. An alternative approach is to use an electronic
portal-imaging device with computer-aided landmark alignment, which is commercially available. Electronic imaging
decreases image acquisition and processing time in comparison to radiographic film and, thus, reduces the possibility of
patient movement and potentially improves or maintains a
certain alignment quality.
Three-dimensional surface imaging, in addition to conventional setup, can perturb clinical workflow. To avoid
elongation of the treatment session, the SML was not taken
immediately after laser setup. Rather, it was acquired immediately after the portal images were taken but before any
positioning adjustment. Due to time constraints, not all
desired images could be obtained, in particular the breathing
study was not always acquired. These omissions do not
influence the data quality, but reduce the number of total
fractions and, therefore, affect statistical analysis.
Images from 9 patients are included in this study. These
data represent the clinical experience in several months of
operation. To definitively determine the overall impact of
surface imaging on PBI setup, a larger study is needed,
which would also allow the cohort of patients to be split up
for analysis (e.g., on the basis of breast size). Three-dimensional surface imaging for setup implicitly assumes a strong
correlation between the skin surface and the subsurface
location of the target volume. Analysis of this correlation
was not technically feasible on our linac, because on-board
diagnostic imaging was not available.

Volume 64, Number 4, 2006

An important consideration when a 3D surface-imaging


device is used as a setup tool is the quality of the reference
model SMR. Accuracy of the SMR is essential because it is
the defining topology against which subsequent treatment
SMs are compared. Possible sources of the SMR include the
breast surface (1) derived from the treatment-planning CT,
(2) acquired during conventional treatment simulation by
use of a second 3D surface-imaging system, or (3) acquired
at the time of the first treatment session. Each reference
surface has its strengths and weaknesses as discussed below.
The treatment-planning CT is typically acquired with the
patient breathing lightly. This situation may lead to motion
artifacts in the CT data, typically visualized as a slight
wavelike structure on the patients skin (19). We have
observed chest motion (1.9 1.1 mm), as measured by the
3D surface studies. Motion of 2 to 3 mm has been reported
by Baroni et al. (16). These motion artifacts may differ from
nongated captures with the 3D camera system, because CT
acquisition requires considerably more time (20 s) than
the less than 10 ms needed for 3D video based surface
measurement. An artifact-degraded reference surface diminishes the advantage of a gated surface model acquired at
daily setup because of the assumption of registering SML to
a rigid body.
Acquisition of the reference surface during treatment
simulation requires a second 3D camera system, which was
not available at MGH. The advantage of acquiring at simulation time is that the same imaging technology for the
SMR is used, in distinction to deriving a reference surface
by segmenting and extracting a skin model from CT data.
Unlike MV portal imaging, clips implanted in the tumor
cavity would be visible in kV imaging. This feature is an
advantage compared with SMR acquisition at the first treatment fraction.
Use of the SMT from the first treatment session as SMR
requires additional high-precision setup devices because the
3D surface-imaging system can only be used in consecutive
sessions. Our analysis of 3D surface imaging suggests that
portal films, as used at MGH, or laser and skin marks, may
not provide sufficient accuracy for satisfactory PBI patient
alignment. A disadvantage of reference surfaces other than
from the first fraction is that they require a coordinate
system change. Although CT is used as the dosimetric
reference coordinate system, other machine coordinate systems, such as that of the linac, conventional radiographic
simulator, lasers, or 3D camera system, are not likely to be
exactly congruent with the CT coordinate system.
The clinical utility of 3D surface imaging for target alignment may be limited by skin-to-tumor position correlation.
However the system installed at MGH can be used for
precise position measurement (17). The images presented in
Fig. 6b show that the arm position of this patient changes
significantly from day to day. This change is likely to
influence breast deformation at some level, primarily along
the lateral/inferior region of the breast. Such a deformation
cannot be detected by portal imaging, because the arm is not
imaged in the irradiated field, and small soft-tissue changes

3D surface imaging for PBI setup

are not visible. Quantitatively, patient repositioning on the


basis of the SMT is at times sufficiently large to be clearly
visible if the surface models are overlaid. For translations of
more than 10 mm, the need for repositioning can also be
detected by a comparison of the 2D texture images.
Analysis based on distance to the reference surface is
primarily for visualization, because in most cases, detection
of the necessary couch transformation back to the reference
is not possible from these color-wash images. Even the
reversed direction, for example, use of the recommended
transformation to appreciate the color-wash plots, is often
too complex, especially if rotations are involved. Nonetheless, a correlation clearly exists between the recommended
couch shifts and distance because the distance is used as the
cost function in the optimization process of the supplied
software. The results that include all patients can also be
interpreted in that manner. The combined distance histograms for laser and portal-film alignment are both broad
distributions, and the mean and standard deviation required
to bring the SML or the SMT back to the SMR are of the
same magnitude.
The interfraction patient-motion analyses were based on
free-breathing data. Free breathing can affect the measured
surface models and, hence, lead to incorrect predictions for
the realignment. The current system can minimize motion
artifacts by performing gated surface imaging, which results
in surface-model capture at the same respiratory phase. This
feature is still a compromise because breathing studies show
that respiration baseline can drift with time, such as when
the patient relaxes on the couch after a few minutes (20).
Gated captures were implemented toward the latter part of
this patient study but not used to allow comparison of the
data acquired previously. Through postprocessing of continuous-acquisition data, the breathing signal can be extracted. The analysis showed a peak-to-peak amplitude of
1.9 1.1 mm over the breast for the combined results of all
patients at isocenter. Baroni et al. (21) studied the influence
of breathing in the positioning of breast patients. They
reported that controlled acquisition at end of exhale decreased marker displacement only slightly in comparison to
free-breathing measurement. This finding is consistent with
small-breathing peak-to-peak amplitude and suggests that
the nongated acquisitions we performed on 9 patients are of
value.
The calculated alignment may be influenced by systematic errors if used with CT coordinates but do reflect patient
motion during treatment. In the combined data for all patients, patient position after laser alignment and portal-film
alignment appear to be of comparable accuracy. The standard deviation of each translation direction is on the order of
4 to 5 mm and 2.5 for rotations. Maximum deviations of up
to 18 mm in translation and up to 12 for the couch rotation
about the vertical axis were observed. A possible explanation for this very large observed shift is that in one treatment
fraction, the patient slid down the breast board during the
time interval of film processing. Extreme translational values of greater than 1 cm were observed in 2 of 9 patients and

C. BERT et al.

1273

from 5 to 12 mm in the translation and up to 5.8 in the


rotation about the vertical axis in the remaining 7 patients.
After application of the recommended couch transformation, the standard deviations were less than 1 mm for each
translation axis and less than 1 for the couch rotation about
the vertical axis. Thus, a good surface fit can be achieved,
with relatively little deformation of the breast surface. The
2 remaining rotational degrees of freedom were of the same
order as for laser or portal-film setup. This agreement is
expected because we did not correct for all 6 degrees of
freedom, because the couch cannot be adjusted in all 6
degrees of freedom. A full 6 degrees of freedom correction
yields an ideal result for rigid objects if a large ROI is used,
because the transformation brings the surface to the minimum of the cost function of the registration algorithm.
Virtual alignment as described is not necessarily equivalent to physical alignment of the patient for image-guided
therapy, because the patient may involuntarily move between SML acquisition and couch correction implementation, a process that currently takes approximately 2 minutes
(17). Because of new PC technology and an interface to the
patient couch, this time will decrease in upcoming versions
of the system. However, the current process still provides a
good estimate of the achievable positioning accuracy.
Baroni et al. (21) performed a similar analysis on marker
positions acquired at each fraction in comparison to a reference data set from the first fraction. They observed maximal deviations of up to 10 mm for many markers in all
principal anatomic axes and 3D marker displacements of up
to 18 mm. Their results are consistent with ours presented in
Table 1 and Fig. 7, with the exception that the maximum
deviation in one anatomic direction was up to 18 mm
(portal film, minimum). After applying the optimized rigidbody transformation (5 of 6 degrees of freedom) that could
be used for realignment of the patient, Baroni et al. (21)
observed a distinct decrease in deviations from the markers
reference position, with mean displacement values of 2 mm
and maximum 3D displacement of about 5 mm. Our data
agree with their observations.
At our institution, the focus to date has been on use of 3D
surface imaging for the setup of PBI patients. For pronepatient positioning, as it is used at some facilities for breast
irradiation (22), 3D imaging is possible but incompatible
with patients in the supine position, because it requires
different positioning of the imaging pods. We plan to explore the potential use of surface imaging for purposes other
than PBI in future studies. Again, correlation between internal anatomy and skin has to be proved before clinical use.
Three-dimensional surface imaging would seem especially
suitable for alignment of extremities, or for head-and-neck
patients (6), even though dedicated fixation devices will be
required that do not obstruct the optical imaging.
CONCLUSION
Surface imaging by close-range photogrammetry was
used to study the setup for PBI. Setup accuracy by portal

1274

I. J. Radiation Oncology

Biology Physics

imaging was found to be as accurate as laser setup, on the


basis of breast-surface congruence as a metric. Simulated
setup by surface imaging after initial laser setup yielded
better surface congruence. Although breast-surface congruence can lead to more accurate target localization, a

Volume 64, Number 4, 2006

rigorous correlation of breast surface and CTV position


must be quantitatively established. This correlation can
be measured by simultaneously gathering surface data
and radiographs of implanted radiopaque clips that surround the CTV.

REFERENCES
1. Vicini F, Remouchamps V, Wallace M, et al. Ongoing
clinical experience utilizing 3D conformal external beam radiotherapy to deliver partial-breast irradiation in patients with earlystage breast cancer treated with breast-conserving therapy. Int J
Radiat Oncol Biol Phys 2003;57:12471253.
2. Formenti SC, Truong MT, Goldberg JD, et al. Prone accelerated partial breast irradiation after breast-conserving surgery:
Preliminary clinical results and dose-volume histogram analysis. J Radiat Oncol Biol Phys 2004;60:493504.
3. Taghian A, Kozak K, Doppke K, et al. Initial dosimetric
experience using simple three-dimensional-conformal external-beam accelerated partial breast irradiation. Int J Radiat
Oncol Biol Phys 2006; in press.
4. Baroni G, Ferrigno G, Orecchia R, et al. Real-time threedimensional motion analysis for patient positioning verification. Radiother Oncol 2000;54:2127.
5. Moore C, Lilley F, Sauret V, et al. Opto-electronic sensing of
body surface topology changes during radiotherapy for rectal
cancer. Int J Radiat Oncol Biol Phys 2003;56:248 258.
6. Rogus RD, Stern RL, Kubo HD. Accuracy of a photogrammetry-based patient positioning and monitoring system for
radiation therapy. Med Phys 1999;26:721728.
7. Soete G, van de Steene J, Verellen D, et al. Initial clinical
experience with infrared-reflecting skin markers in the positioning of patients treated by conformal radiotherapy for
prostate cancer. Int J Radiat Oncol Biol Phys 2002;52:694
698.
8. Riboldi M, Baroni G, Orecchia R, et al. Enhanced surface
registration techniques for patient positioning control in
breast cancer radiotherapy. Technol Cancer Res T 2004;3:
5158.
9. Casebow MP, Harris RL, Jarvis RJ. The use of an optical
outlining system for weekly on-set verification of patients with
carcinoma of the breast. Radiother Oncol 1998;46:99 103.
10. Johnson LS, Milliken BD, Hadley SW, et al. Initial clinical
experience with a video-based patient positioning system. Int
J Radiat Oncol Biol Phys 1999;45:205213.

11. Millikin BD, Rubin SJ, Hamilton RJ, et al. Performance of a


video-image-subtraction-based patient positioning system. Int
J Radiat Oncol Biol Phys 1997;38:855 866.
12. MacKay RI, Graham PA, Logue JP, et al. Patient positioning
using detailed three-dimensional surface data for patients undergoing conformal radiation therapy for carcinoma of the
prostate: A feasibility study. Int J Radiat Oncol Biol Phys
2001;49:225230.
13. Siebert JP, Marshall SJ. Human body 3D imaging by speckle
texture projection photogrammetry. Sens Rev 2000;20:218 226.
14. Sjdahl M, Synnergren P. Measurement of shape by using
projected random patterns and temporal digital speckle photography. Appl Opt 1999;38:19901997.
15. Smith N, Meir I, Hale G, et al. Real-time 3D surface imaging
for patient positioning in radiotherapy. Int J Radiat Oncol Biol
Phys 2003;57(Suppl. 2):S187.
16. Baroni G, Ferrigno G, Orecchia R, et al. Real-time opto-electronic verification of patient position in breast cancer radiotherapy. Comput Aided Surg 2000;5:296 306.
17. Bert C, Metheany KG, Doppke K, et al. A phantom evaluation
of a stereo-vision surface imaging system for radiotherapy
patient setup. Med Phys 2005;32:27532762.
18. The Visualization Toolkit, Version 4.0. http://www.vtk.org.
19. Chen GTY, Kung JH, Beaudette KP. Artifacts in computed
tomography scanning of moving objects. Semin Radiat Oncol
2004;14:19 26.
20. Mageras GS, Yorke E. Deep inspiration breath hold and
respiratory gating strategies for reducing organ motion in
radiation treatment. Semin Radiat Oncol 2004;14:6575.
21. Baroni G, Garibaldi C, Scabini M, et al. Dosimetric effects
within target and organs at risk of interfractional patient mispositioning in left breast cancer radiotherapy. Int J Radiat
Oncol Biol Phys 2004;59:861 871.
22. Grann A, McCormick B, Chabner ES, et al. Prone breast
radiotherapy in early-stage breast cancer: A preliminary analysis. Int J Radiat Oncol Biol Phys 2000;47:319 325.