Technical Portal

On-line

Developments

Imaging:

Computer-assisted Materials

Wing-Chee Lam, PhD Michael G. Herman, PhD Kam-Shing Lam, PhD Ding-Jen Lee, MD, PhD

from

a patient

simula-

tion film are superposed on the daily on-line portal images for visual evaluation of placement accuracy. The ability to check the alignment after administration of 3-6 cGy of radiation and the minimal operator interaction required make this system useful in radiation therapy delivery. Index

terms:

peutic peutic

radiology, radiology,

Radiology

Therapeutic radiology computer-assisted quality assurance

1991;

#{149} Thera#{149} Thera-

179:871-873

years of research (i-3) and development, equipment for ondigital imaging of radiation thera-

FTER

line

py treatment

portals

clinical use (4-6). of this equipment treatment portals fields

is now

ready

for

The major advantage is that it enables for all stationary

to be recorded

daily.

When

the

equipment is fully implemented, the technologist can review each treatment field with a small dose of radiation, check the alignment, and then deliver the full daily dose. Previously, we have reported that an on-line imager can enable detection of field displacements in a significant number of cases (3). To facilitate the measurement of portal misalignment and to inform the technologist

or clinician

needed, surement use with

of the

adjustment

a computer-aided technique

error was

developed

meafor

an on-line imager. This article details our approach and results in the development of a computerized system for comparison of the portal (treatment) and simulation (reference) images.

From the Department of Radiation Oncology, The Johns Hopkins Oncology Center, 600 N Wolfe St, Baltimore, MD 21205. Received September 21, 1990; revision requested November 13; revision received December 1 1; accepted December 26. Address reprint requests to I

W.C.L. c RSNA,

Volume

1991

179

#{149} Number

3

Error

Measurement’

and Methods image acquisition subthe SRI-i00 (Philips, Conn) (4). A block diagram of

used

Shelton, the

derived

Instrumentation

The on-line system

A microcomputer-based system was implemented in an on-line portal imaging system to determine portal misalignment in radiation therapy. Reference markers

and

system

was

is shown

in Figure

i. The

x-

ray imager consists of a fluorescent screen, a mirror, and a video camera. For convenience of patient setup, the screen-mirror box is detachable from the gantry mounting support. The image processor consists of a Data Translation model 285i high-resolution frame grabber and a model 2858 auxiliary frame processor (Marlboro, Mass). Video signals can be digitized, proceased, and displayed on a video monitor. A radiation dose of 3-6 cGy is required to generate a portal image on the video monitor. There are many ways to visualize the misalignment of the treatment portal. In one method, a few landmarks from the reference image are lined up with the corresponding landmarks in the treatment image. The difference in the boundaries shows the setup error (7). In another method, the boundaries of the reference image are aligned with those from the treatment image. The displacement of the landmarks between the two images shows the setup error. We used the second method because the computer was used to align boundaries and, therefore, very little operator interaction was required to obtain field-placement errors. To determine the extent of portaldisplacement errors, a comparison is made with the simulation film. Direct image comparison is difficult, since the reference and treatment portal images are presented on two different display media. Our approach was to video digitize the simulation film image and display it simultaneously with the on-line image on another video monitor. Due to the difference in the magnification of the two images, the extent of the error was difficult to assess. This method allows the information from the reference image to be superposed on the treatment portal image with the same magnification for quick error detection. The automated error measurement process consists of the following five steps: 1. Capture the simulation film. To implement the computer-aided error measurement procedure, two pieces of hardware were added to the configuration: a video camera and a light box (see shaded area of Fig 1). The reference portal is captured into the system from the simulation film by using the video camera.

Figure

aging

1.

Block

system

diagram

of an on-line

for radiation

Rfrnc

Figure denotes sents

Outline

2.

on-line reference

matching.

image outline.

and

im-

therapy.

Outlln

Shaded open

area

area repre-

2. Generate reference markers from the simulation image. The microcomputer mouse is linked through software to the graphics cursor of the image processor and is used as a locator to process the reference image. Reference line segments marking the outline of the portal, as well as a few selected radiographically distinct anatomic landmarks, are generated. In addition, the position of the and a reference

central

axis

of the

portal

point such as the field size cross wires are also digitized into the system. They are used to give a scaling factor. A central point of the port is also digitized if it does not coincide with the central axis. These reference markers are used as a standard for cornparison of all treatment images. One set of reference markers is generated for each treatment field. 3. Acquire and display the treatment portal image. The on-line portal image is acquired

This

occurs

during during

patient treatment. the administration

of the first few centigrays of radiation of each field. The image is displayed for immediate review. 4. Optimize the matching of reference boundaries to treatment field boundaries. Optimum matching of the Radiology

#{149} 871

(Fig

2). A threshold

values

for

is used

the

pixel

for edge

de-

The x21 which describes the of the boundaries of these

tection. matching

two

level

of the image

images,

is defined

as

(R1 - V.)2.

x2

To achieve a better match of the two boundaries, the x2 is minimized by applying three transformations to the line segments, that is, by means of scaling or magnification, translation, and rotation. The scaling factor, denoted by m, 3.

is given

Figures 3, 4. (3) The optimized alignment is determined from marks.

outline the 1-cm

is superposed

grid.

(4) Brain

on

port

the

on-line

simulation

image.

image

Brain

port

with

reference

by the

equation

mis-

IV.IRI/IRI

m

Ik,I

where the

and

V,are R1 and V.

vectors

the translations denoted

x

=

.pRj,

the lengths respectively.

in the x and

as x and

y, then

(

-

R

of If

y axes

are

V)/n

and y

(:

=

R’

-

:

v)/

n,

where

R.x and RY are the x and y cornponents of the vector R, V and VY are those for vector V1. and n is the total number of points marking the outline. The rotational correction is determined by applying a series of 10 rotational transformations in both

6.

5.

5, 6 (6) Reference Figures

(5) Optimized reference outline superposed on an on-line marks and outline for the intended lung field.

portal

image.

clockwise

u__w

r ..

r_

4

counterclockwise

portal

boundary

tioned

earlier,

direc-

portal line segfor a local mini-

each

transforma-

reference outline new treatment

is generated as menthe x2 is calculated. The entire process is repeated until the x2 is minimized. This optimizing algo-

:

_

and

tions to the reference ments and searching mum in the x2. After tion is applied to the in the image plane, a and

rithm works well in all cases except those in which the radial vectors intersect the boundary segments at small angles; this results in artificially large

difference

7. Figures

7, 8.

(8) Simulator

(7) Optimized

film

and

reference

reference

outline

markers

superposed

for the intended

on

an

pelvic

on-line

pelvic

portal

image.

field.

terms

(without

Clara, takes

image

is divided

ments

at approximately

By using

the

as an origin,

872

#{149} Radiology

point

into

short

line

i-cm

of the portal

a set of radial

seg-

intervals.

vectors,

center R,, is

generated to the edge of the reference boundary at the ends of the line segments. By searching along these radial vector directions in i-mm increments on the on-line image, the edge of the treatment portal can be detected and_ denoted with a set of radial vectors, V1

x2. Hence,

a math

coprocessor)

final

results

the

radial are exan IBMwith an unit

(Santa

Calif), the minimizing only a few seconds.

5. Display portal outline is carried out by means of an automated algorithm. The reference portal outline from the simulation

in the

line segments that intersect the vectors at less than a 45#{176} angle cluded from the process. With compatible personal computer Intel 80386 central processing

procedure

with

measure-

ment grid for error assessment. After the matching of boundaries has been optimized, the results are shown on two display monitors. The treatment portal image, the reference markers, and a i-cm square grid are displayed the

imaging

monitor.

The

on

reference

June

1991

markers can be toggled on and off with a single keystroke for easy observation of low-contrast anatomic structures in the portal image. The second monitor is a VGA graphics monitor (IBM, Armonk, NY) that shows the original simuiation image. The VGA display with 640 X 480 pixels with 64 levels of gray is adequate

for

tion image additional

displaying

the

simula-

and enables comparison anatomic landmarks.

of

Results The application of this technique is demonstrated in three sample cases. The first case is a straightforward treatment port for a brain tumor. The optimized matching of the boundaries is displayed in Figure 3. Simultaneously, the reference or simulation portal and reference markers are displayed on the VGA monitor as shown in Figure 4. By observing the markers and the bony landmarks of the sinuses, and with the help of the i-cm square grid, it is easy to see that the patient has to be moved anteriorly 0.5 cm and inferiorly 0.5 cm. The second case shows a lung cancer treatment portal in which part of the treatment boundary is concave. An inferior block creates several boundary line segments that make small angles with the radial vector from the center of the portal. These segments are excluded from the automatic search and optimizing procedure as described in the previous section, since they generate poor matching portal edge points. The automatic alignment algorithm performs well to match the boundaries, though

there

ment trated

was

of the inferior in Figure 5. sition of the trachea, the corresponding the patient should mately 1 cm to the

Volume

179

a 0.5-cm

misplace-

block. This is illusBy observing the pothe clavicle, and reference markers, be moved approxileft. Figure 6 shows

#{149} Number

3

the simultaneous simulator

film,

VGA

display

including

of the

reference

marks. The third example illustrates a case of mismatched treatment boundaries for a pelvic port. The final optimized match is shown in Figure 7. Note that a rotational transformation has been applied to optimize boundary matching. The superior and inferior boundaries of the treatment port, which are determined with treatment machine collimators, are expected to be parallel to those of the simulation port. The optimized match indicates that these boundaries are rotated relative to each other to have maximum boundary matching defined by the lateral focused

blocks.

The

boundary

mismatch

suggests that the shadow blocks were not mounted properly. For this lowcontrast image, the position of the bony landmarks can be more easily observed by toggling the reference markers on and off. There is more than a 1cm mismatch at the widest part of the pelvic brim. This suggests that there may also be a magnification problem for this treatment portal. Either the source-to-skin distance is improperly set or the field size and the blocks are set too close to the isocenter. Investigation reveals that the two lateral blocks were mounted too close together and one was slightly rotated. To correct the setup, the blocks have to be remounted. The reference simulator image is shown in Figure 8.

so that corrective action can be taken if necessary. Acceptance by the community demands that the operation be made simple. The image comparison a!gorithm in which the portal boundaries are used as a reference requires the digitization of the simulation film before the patient’s first treatment. After this information is stored in the computer, all subsequent portal image evaluations involve only the visual observation of the reference markers relative to the landmarks and boundaries in the on-line image, with very little need for computer operation by the technologist or oncologist. This automatic image alignment method can be easily implemented in any on-line imaging system. U References 1.

2.

3.

Phys 1980; 6:935-939. Leong J. Use of digital fluoroscopy as an on-line verification device in radiation therapy. Phys Med Biol 1986; 31:985-992. Lam

WC,

Partowmah

MD, Lam KS. field placement 4.

diotherapy. Visser AG, type

5.

M, Lee

On-line errors

DJ, Wharam

measurement of in external beam ra-

Br J Radiol 1987; 60:361-367. Huizenga H, Althof VGM,

Swanenburg

BN.

Performance of a protoradiotherapy imaging Int J Radiat Oncol Biol Phys 1990;

fluoroscopic

system. 18:43-50. Munro P, Rawlinson JA, Fenster ital fluoroscopic imaging device

A. A digfor radioOncol Biol

localization. Int J Radiat 1990; 18:641-649. Wong JW, Binns WR, Cheng AY, et al. Online radiotherapy imaging with an array of fiber-optic image reducers. Int J Radiat Ontherapy

Phys

6.

Discussion On-line portal imaging is the next advance in quality control in radiation therapy. This article deals with the detection of treatment portal displacement, which is a primary concern in the delivery of radiation therapy. Patients will benefit from this new infovation if the system can provide immediate and easy observation of the extent and the characteristics of setup errors

Baily NA, Horn RA, Kamp TD. Fluoroscopic visualization of megavoltage therapeutic x-ray beams. mt J Radiat Oncol Biol

col Biol Phys 7.

Meertens

H,

1990; 18:1477-1484. Bijhold

od for the measurement errors in digital portal Biol 1990; 35:299-323.

J,

J. A methof field placement images. Phys Med

Strackee

Radiology

#{149} 873