- Email: [email protected]
Ultrasound image fusion for external beam radiotherapy for prostate cancer
Int. I Radiation
Oncology
Bid
Phys.. Vol. 3s. No. 5, pp 975-984, 19% Copyright Q IQ96 Elsevier Science Inc. Prinird in the I!SA. All rights reserved Oi’~O-W16/9h $lS.oo t .(H)
PII: SO360-3016(96)00231-3
l
Physics Original Contribution ULTRASOUND
IMAGE
E. J. HOLUPKA,*
FUSION FOR EXTERNAL BEAM FOR PROSTATE CANCER
I. D. KAPLAN,*
E. C. BUFCDETTE'
AND
RAD3OTHERAPY
G. K. SVENSSON*
*Joint Center for Radiation Therapy and Department of Radiation Oncology, Harvard Medical School. Boston. MA; and ‘Domier Medical Systems, Champaign, IL Purpose: To determine whether real-time uitrasound imaging and targeting system for the treatmeut of prostate cancer was feasible. The initial phase of’ this project included a study to develop and determine (a) &Ware for the fusion of ultrasound images to standard x-rays obtained during simulation, and (b) the potential reduction in field size with real-thne imaging. Methods and Materials: During 13 patient simulations a transrectal uRrasound image was obtatned. Orthogonal x-ray films were acquired with the rectal probe in place. Both the x-ray and llitmmdimageswePe and a R&on image was created of the prostate position in relation to the probe, bktdder, and re&un. The twodimensional area of the rectum, bladder, and prostate was determined in the lateral projection. Potential conformal blocks were designed for the lateral portals in a four-field treatment teclrrdque. Results: The transrectal ultrasound probe enabled real-time prostate hnaging. The lateral Wd sixecan be reduced to 6.1 x 5.68 cm2 t 0.62 x 0.48 cm’ from the standard 8 x 8 cm2 fieid. The posterior rectal wall was physically dispked out of the lateral field. The area of the rectum included in the lateral field is 1.75 cm’ 2 0.85 cm*. Con&sion: The prostate position can be determined with certainty on a regular basis with m tdtrasonograpby. The amount of normal tissue in the high dose volume can be reduced. This approach may reduce acute and chronic morbidity and allow further dose escalation. Ultrasound, Image fusion, Prostate cancer, Conformal radiation therapy.
INTRODUCTION Adenocarcinoma of the prostate is the most commonly diagnosed noncutaneous cancer in the United States male population. It is estimated that 200,000 new cases of prostate cancer will be diagnosed in 1994. Approximately 16% of these newly diagnosed cases will be locally advanced nonmetastatic cancers or, more specifically, AJCC Stage III-T3NOMO. Treatment for patients in this stage is problematic. Several authors have reported significantly low rates of local control with conventional doses of radiation delivered with external beam radiation therapy for clinically locally advanced disease. To improve the rate of local control, more aggressive therapy with conformal techniques and dose escalation is being investigated. Perez et al. reported local recurrence rates of 38% for patients with T3 tumors treated with less 60 Gy and 12% treated with 70 Gy or greater (17). Similar results for improved local control with higher doses were reported in the Pattern of Care Study (9). Strategies to increase the dose to the prostate in excess of 70 Gy are being investigated. Syed et ~1. and Bagshaw et al. have reported im-
proved rates of local control with interstitial 19’Ir implant boosts, which increases the dose to the prostate to approximately 80 Gy (1, 25). Benk and associates from the Massachusets General Hospital report excellent rates of local control in patients with T3 and T4 disease treated with a combination of photon and proton beam therapy to a dose of 75.6 Cobalt Gray Equivalent (3). Several groups have reported on the use of conformal techniques in an attempt to escalate the dose to the prostate with external beam irradiation. Careful dose escalation studies are underway at several institutions to determine the acute and late toxicities of increased dose to an albeit reduced volume of rectum and bladder. Early studies are encouraging. Schultheiss et al. reports that in a multivariate analysis fewer Grade II toxicities were observed in patients treated with conformal therapy (22). Sandler et al. reports low rates of chronic rectal morbidity with a similar conformal approach. Follow-up may be too short to fully ascertain the late complication rates of increased dose to normal tissue in this studies. Benk et al. describes late rectal bleeding with increased dose delivered with a proton boost. The rate of rectal bleeding
Accepted for publication 29 April 1996. Reprint requests to: Dr. Edward J. Holupka. Joint Center for
Radiation Therapy, Department of Radiation Oncology. Harvard Medical School, Boston, MA 02115. 975
976
I. J. RadiationOncologyl Biologyl Physics
Volume 35, Number 5, 1996
of both dose and rectal volume irradiated. Importantly, the rate of rectal bleeding continued to increase with follow-up over 2 years (3). Many of the conformal techniques that are necessary for the safe and proper study of dose escalation requires exact delineation of the target volume. This can be accomplished if the prostate tumor volume is accurately visualized and when precise control or knowledge of those variables that determine the margin around the tumor volume is gained. Extent of microscopic disease, dose-limiting structures such as the anterior rectal wall, and prostate motion are important variables for determining the target volume. Studies have indicated significant motion of the prostate in relation to bony landmarks (2, 21, 26). This variability of the gross target volume makes conformal therapies more difficult to perform. In this study we describe the use of transrectal diagnostic ultrasonography and image fusion to accurately delineate the prostate and rectal wall during the simulation and therapy for prostate cancer. The benefit of diagnostic ultrasound imaging can be best appreciated by spatially correlating the diagnostic information of the ultrasound image to other imaging modalities that have proven to be useful for radiation therapy such as Computerized Tomography, Magnetic Resonance Imaging, and high- and low-energy radiographic x-ray images. Ultrasound medical imaging has a long history, and the present day technology is practically distortion free. For the purposes of this study, any and all distortions of diagnostic ultrasound is negligable. Image correlation methods have recently received much attention for its application in the field of radiation therapy (5, 8, 10-l 1, 14, 16). There are numerous algorithms for matching medical images in both two- and three-dimensional frameworks (5, 10, 12,27). Many of the applications include the multimodality fusion of CT, MRI, Single Photon Emmision Computerized Tomography (SPECT), and x-ray images (14,20,24). However, little work has been done concerning the fusion of diagnostic ultrasound images to other imaging modalities. The purpose of this study is to demonstrate an innovative image fusion technique that allows real time superposition of an ultrasound image of the prostate onto a digital simulator film or high energy portal film. The method is a two-dimensional, landmark-based algorithm that can be extended to correlate the ultrasound image to any imaging modality (5). We will also discuss the potential clinical benefits of using ultrasound image-guided simulation and treatment of the prostate. Our hypothesis is that ultrasound image-guided simulation enhances our
ability for accurate definition of the prostate gross target volume in relation to critical structures such as the anterior rectal wall and the bladder. We also postulate that ultrasound image-guided therapy will increase the spatial treatment accuracy by the direct visualization of the prostate during therapy and, hence, provide the necessary input data for real-time dose optimization and the potential for dose escalation. Furthermore, the use of a rectal ultrasound probe tends to immobilize the prostate in relation to the rectal wall.
‘Urovision Ultrasound System, Dornier Medical Systems, Inc., Kennesaw, GA. *Model ER-S, DomierMedical Systems,Inc., Kennesaw,GA.
‘PercepticsPixel PipelineandPixel Tools, Perceptics,Knoxville, TN.
is observed to increase as a function
METHODS
AND MATERIALS
The ultrasound diagnostic image system’ operates together with a transrectal probe’ operating at 7.5 MHz. The transrectal probe has 96 active transducer elements with a maximum display depth of 13.5 cm. The transrectal probe is approximately 12 cm long by 1.8 cm in diameter at its thickest part. After the standard simulation procedure the transrectal probe is inserted into the patient and diagnostic ultrasound images are obtained. The entire ultrasound session is videotaped and digitized. The transrectal probe is positioned such that the resulting image is in a vertical plane through the isocenter and parallel to the simulator film plane. The simulator image is a projection image that results in the divergent sum of all parallel image planes from the x-ray source to the film plane. Hence, structures that lie above and below the isocentric plane of rotation will appear in the simulator image. This is to be contrasted with the ultrasound image, which is a single planar image. If the prostate is an approximately spherical, midplane structure, this position maximizes the cross-section of the prostate on the image. However, often the prostate tumor volume wraps around the anterior rectal wall, in which case the single plane ultrasound image will not represent the maximum cross-section of the prostate. This situation maybe particularly common when the ultrasound probe pushes the prostate gland anteriorly. To quantitatively asses the three-dimensional shape of the prostate while the probe is in place, a three-dimensional CT scan was obtained with the probe in place. The prostate, rectum, and probe were then reconstructed in three dimensions to determine the position of these structures with respect to each other. The results are displayed in Fig. 1. The prostate does wrap around the rectum and probe. The amount of prostate tissue that wraps around the probe was quantified. Continuing with the ultrasound imaging of the prostate during the simulation session, the maximum width and height of the prostate volume are measured and also marked on the ultrasound. Once the session is over, the videotape of the ultrasound imaging session is digitized, 3 resulting in a dig-
Ultrasound image fusion for external beam radiotherapy
0 E. .I. HOI
L’PKA
ef al.
977
portal film. The probe is oriented so that the ultrasound image plane is parallel to the film plane. Therefore, the ultrasound image of the prostate, which is geom&ically related to the ultrasound probe, can be quantitatively and accurately fused with the simulator or treatment film images. Fusion of the diagnostic ultrasound image with the simulator x-ray image is achieved by using a two-point matching technique that requires the unambiguous identification of two identical point fiducids on each image. The transrectal probe’ uses a sector phased array sweeping method yielding a fan-shaped image. The arc at the top of the ultrasound image directly corresponds to the spatial position of the transducer array inside the probe head. Figure 2 displays a typical transrectal ultrasound image of the prostate. The endpoints of the transducer arc are labeled “1” and “2” in the image. Figures 3 and 4 displays the lateral simulator film and the corresponding lateral portal film with the transrectal probe in place. The transducer array inside the probe head is clearly visible in both imFig. 1. Volume reconstruction from CT scans of the prostate, rectum, and transrectal probe. The transparent nature of the volumes allow the determination of the degree overlap of the prostare volume with the transrectal probe.
ital image of the prostate in the vertical isocentric plane. The lateral radiographic film containing the rectal probe is also digitized.4 The projected outline of the probe can be clearly seen on the simulator film and the high-energy
Fig. 2. A typical ultrasound image of the prostate from the transrectal probe. The lines centered on the crosshairs indicate the dimensions of the prostate determined during the time of the simulation. The two fiducial points used for the fusion are the ends of the transducer array and are labeled as “ 1” and “2.”
Fig. 3. Lateral simulator film displaying the location of the ultrasound transrectal probe. Note how the transducer array is clearly visible in the image. The two fiducial points used for the fusion are the ends of the transducer array and are labeled as “1” and “2.”
----____ ‘TruVel
Scanning
system, TruVel,
Hendron,
VA.
I. J. Radiation Oncology 0 Biology 0 Physics
Volume 35, Number 5, 1996
the ultrasound image to the simulator image, is given by the solution to the linear system of equations given by,
The solution to such linear systems is well known and can be found in any standard text on numerical analysis (28). Once the solution is known, every point in the ultrasound image can be mapped to the simulator image using Eq. (1). Figure 5 displays the result of transforming the ultrasound image to the simulator image. This will be referred to as the fusion image. Figure 2 displays the ultrasound image data along with other useful diagnostic information. ’ However, only the ultrasound image data is transformed so that the details of the underlying radiographic image are not obscured. The accuracy of the rotational and translational component of the transformation is determined by the user’s ability to localize the two point fiducials on each of the ultrasound and radiographic images. The fiducials corre-
Fig. 4. Lateral high-energyportal film displayingthe locationof the ultrasoundtransrectalprobe. Note how the transducerarray is clearly visible in the image.
ages. The ends of the transducer array that correspond to the points on the ultrasound image are likewise labeled “1” and “2.” By zooming in on the images, the points that correspond to the fiducial marks can be determined to within a single pixel. Once the fiducials have been identified, the spatial transformation that maps the ultrasound image to the simulator image is determined. The transformation is assumed to consist of a two-dimensional rotation, translation, and a global isotropic scaling. This leads to a four-parameter transformation,
= A ‘;;$ f$)(;::) + (;;) = (;::), (Eq.1) v rere t, and tYare the translations, 13the rotation angle for
a otation about the center of the image, and A is the global scaling. (x”‘, y”“) and (x6’, y”‘) represent the coordinate pair of any point in the ultrasound and simulator images, respectively. If the two fiducial points on the ultrasound image are labeled as (XT’, y?) and (xz’, y;‘), and the two fiducials points on the simulator film as, (x?, yS’> and (xi’, y;‘), then the matrix M, and translation f which maps
Fig. 5. Fusion image displaying the spatial mapping trasound image data to the simulator image.
of the ul-
Ultrasound image fusion for external
spending to the ultrasound image can be determined automatically because for a given setting of the field of view (FOV), of the ultrasound imaging hardware, the position of the endpoints of the transducer array always appear at the same point, or pixel, in the digitized image. The two point &luciaIs for the ultrasound image can be determined a priori.’ The automatic determination of the fiducials on the simulator image presents more of a problem because the probe does not always appear exactly in the same position on the radiographic image as in the ultrasound image. However, t&e image can be zoomed and the endpoints of the transducer array can be determined to within a pixel. Therefore, the accuracy of the spatial position of the fusion is determined by the pixel size of 0.265 mm for the simulaior image. The ability to afxurately determine the extent of the gross tumor volume and resultant target volume is dependent on the accuracy of the fusion. The accuracy in the position of the hvo fiducial points in the radiographic image is determined a priori by the pixel size of the digital images, as previously discussed. However, the accuracy of the rotational and scaling transformations must be determined. After the ultrasound image is transformed to the simulator image, volumes of interest can be manually contoured on the fusion image. Figure 6 displays the fusion image where the prostate, bladder, and rectal volumes are contoured and color washed over the underlying image. The color-washed representation of the volumes allow for the complete delineation of the volumes while still allowing the structu~ of the underlying image, whether it be the ultrasound, treatment, or simulator image, to show through. In add&ion, the proposed radiation treatment field isocenter, w&h and height can be specified. Figure 7 displays the fusion image with the contoured volumes of interest and proposed radiation field superimposed. In addition to the field outline, blocks, which appear as transparent red regions, have also been drawn. The blocks have been chosen to minimize rectal wall and bladder involvement in the field while ensuring that the tumor volume with a 0.5 to 0.75 cm margin receives 100% of the dose. Once the field and blocks have been drawn, the percent area of pro&&e, bladder, and rectum, which is inside the radiation portal, is automatically calculated and reported on the top of the fusion image. The area of the resultant radiation portal is also calculated and displayed. This quantitative information is available at the time of the simuIatioB or treatment and may prove to be clinically vahble when assessing efficacy and morbidity. To f$ther illustrate the potential of this ultrasound fusion technique, a fulI treatment plan using a four-field box was generated using the simulator-ultrasound fusion image as the starting point. Axial and sagital contours were
sScanitronix
Medicai
AB, Uppsala, Sweden.
beam
radiotherapy l
E. J. HOL~JPKA
er al.
079
obtained for one patient. The planned isooen&-z was obtained from the fusion image and local.&& with respect to both contours. Using dose distributions were fusion image. Figure 8 dispIays prostate, rectum, and bladder with the resultant isodose l&es. Because the to visualize the presence of the h-an investigated. The transrectal probe was placed in a water tank and the dose was measmed a distance of 5.0 cm away from surements were made in the beam. The measurements were repeated wiahow thetransrectal probe and compared. There was no able difference in the dose whether the probe was present or not. The imaging tool described above a.lIows for the t&ion of the ultrasound image to the sequent contouring, field speci and calculation of the degree of vohune of interest involvement iu the field. 6 The uItzasou& iW64OX 480 pixels and the simulator images are 81)o X 1200 pixels. The fusion process takes approximately 5.0 s, and the calculation of the percent of the volume involved in the field takes approximately 1 to 10 s per average volume of interest. RESULTS The amount the prostate wraps around the rectum can influence the posterior extent of the target volm. Figure 1 indicates an overlap of approximatey 15% of the prostate anterior-posterior extent. This value can then be used to extend the field margin in the anterior-posterior direction to ensure adequate coverage of the pro&ate volume. The spatial accuracy of the global scaling transformation can be determined by measuring the length of the drawn axes that measure maximum width and height of the prostate on the fusion image. The maximum width and height of the prostate, indicated by the crossed lines traversing the prostate volume seen on Fig. 2, is determined and recorded during the acquisition of the ultrasound image. For the particular study in Fig. 2 the maximum width is 4.7 cm and the maximum height is 4.0 cm. These same dimensions can also be determined by inspection of the fusion image. The pixel size of the fusion image is 0.275 mm per pixel; therefore, the width and height as determined by the fusion image can de&mined. For this particular study the transformed maximum width and height are 4.9 cm and 4.2 cm, respectively, lead@ to a discrepancy of 2 mm between the ultrasound and fusion image.
%tardent
Vistm 800, Stardent Computer
Inc., Concord. MA.
I. J. Radiation Oncology 0 Biology 0 Physics
Fig. 6. Fusion imagedisplaying the color-washedvolumesof interest.The bladder(green),rectum(blue),andprostate(yellow) are color washedover the imageso as not to obscurethe underlying imagedata.
Volume 35, Number 5, 1996
Fig. 7. Fusion image displaying the color-washedvolumesof interest and the designof the blocks and radiation portal. In addition, the amount of VOI involvement in the radiation is displayedon the top of the image.
The maximum width and height of the prostate for 11 of the 13 patients was determined using both the ultrasound
image and the fusion image. The average and standard deviations for all 13 patients were computed. The average difference is determined to be 2.0 + 2.6 mm standard deviation. Because the error resulting from the pixel size is small (5 0.275 mm) as compared to this difference, we have chosen to interpret this as the average error in the global scaling transformation. Note that because the rectum completely collapses around the probe, the position of the anterior rectal wall can be determined within the accuracy of the simulator or portal film, or 0.275 mm. An additional error in the transformed image results if the ultrasound image plane is noncoplanar with the radiographic film plane. This occurs if the planes are rotated by an angle of B and 4 with respect to each other, where these angles are defined in Fig. 9. The degree of coplanar coincidence of the ultrasound image with respect to the radiographic image is completely determined by the distance between the two point fiducials and the distance of the perpendicular bisector of this line to the arc of the transducer array as measured on the fusion image. If the true distance between the two point fiducials and the perpendicular bisector on the ultrasound image are given by dy$and d$ respectively, and the distance between the two
Fig. 8. Fusionimagedisplayingthe.color washedvolumesof interestandthe isodoselinesfor a pmposed four-fieldbox technique.
Ultrasound image fusion for external beam radiotherapy 0 E. J. HOLUPKA
4
=sizeofloIlgaxis
dUS P, =sireofsholtaxis
cos(e) = 2
dUT pb
COS(#=s 47
Fig. 9. Geometric par;uneters used in the definition of the degree of coplanar coincidence be.hvem the ultrasound and simulator images.
point fiducials and the perpendicular bisector measured on the simulator image are given by d% and d$, respectively, then the degree of planar coincidence is given by,
et al.
981
ment volumes. This comparison is obviously not exact because a total of four fields contribute the true treatment volume; however, it gives a reasonable estimate of the contribution of the lateral fields to the treatment volume. The results indicate that for the 13 patients studied the average area of the actual lateral fields delivered was 64.0 2 2.5 cm’, which is compared to 36.0 + 1.2 cm* for the ultrasound guided fields. This represents a significant reduction in the ultrasound guided treatment volume. For the standard treatment, the amount of rectum treated is difficult to assess because of the uncer&inties in its position during all of the treatment fractions. Because the ultrasound probe serves to fixate both the prostate and rectum, the amount of rectum that would be involved in the radiation field can be determined from the ultrasound fusion image. In columns 4 through 6, the area of the prostate, bladder, and rectum that are directly involved in the field are reported. On average 1.75 cm’ of rectal tissue and 2.35 cm* of the bladder are involved in the lateral radiation field when guided by the fusion technique. The method of ultrasound localization can be used to give a measure of the “goodness” of a treatment by considering the quantitative amount of healty tissue involvement. DISCUSSION
(Eq. 3)
As shown in Fig. 9, the rotation is defined about the center of the line connecting the two point fiducials and, hence, no effect for the divergence of the simulator x-ray field, or equivalently, the distance from the x-ray source to the film plane, is taken into account. This approximation is teasmable bwause the size of the transducer array is small co* to the distance of the source to the film plane. The error in the translational and rotational transformation used in the fusion process does not effect the calculation of the degree of coplanar coincidence because these particular transformations are defined to be in the radiographic plane and do not effect the absolute distance between the fiducial points, for instance, they am orthogonal s. However, the error in the global scaling does effect the absolute distance of the fiducials. The error in the scaling is small compared to the size of the transducer array, and the effect of ignoring this transformation error is small. The degree of coplanar coincidence was mewured for all patients studied. Note that the ultrasound image plane was coincident to the film plane with in an average error or 8 = 9.3” and 4 = 3.4”. A total of 13 patients were studied. Table 1 shows the resnIts for all the patients studied. Column 2 indicates the area of the lateral field portals actually delivered for each patient. Column 3 Micates the area of the lateral field portals as deeed by the fusion image. The difference between these two quantities gives an indication of the difference between the actual and ultrasound guided treat-
At the Joint Center for Radiation Therapy, a four-field box technique is used that is comprised of two opposing laterals, an anterior-posterior and a posterior-anterior field. In the lateral fields, the posterior rectum and anterior bladder are blocked. The Iirst course of therapy uses a 10 x lOcm*toll x 11cm2fieldfollowedbya8X8cm2 to 9 x 9 cm* shaped cone-down field. The simulator image has a high contrast for bony stuchues and to a certain extent, some soft tissue structures. However, the prostate can not be seen in a simulator image and its position must be approximated from neighboring structures, in particular, the bladder, rectum, and neighboring bony structures. Other information, such as size, shape, and approximate relative position can be inferred from a prior CT scan. However, many uncertainties including the gross tumor volume, clinical tumor volume, patient setup uncertainty, and motion of the prostate gland contribute to the definition of a necessarily large planning target volume. The uncertainty in defining the gross target volume (the prostate), the subclinical tumor extension leading to the clinical target vohnne, and finally technical uncertainties such as field edges and organ motion have lead to an approximate margin of 1 S cm around the perceived gross tagret volume. Historically, this margin was thought to safely encompass the gross tumor volume. However, recent studies of patient motion using postal have indiimaging and of organ motion during cated that more attention is needed to determine the margins of uncertainty included in the target volume. A conventional field size and blocking arrangement may,
I. J. Radiation Oncology 0 Biology 0 Physics
982
Table 1. Volume Study 1 2 3 4 2 7 8 9 10 11 12 13 Average Standard deviation
Actual field portal (cm21
involvement
Volume 35, Number 5, 19%
for ultrasound
Proposed field portal (cm*)
guided radiation
portals
Rectum in field (cm2)
Bladder in field (cm2)
Prostate in field (cm21
8.0 8.0 8.0 8.0
x x x x
8.0 8.0 8.0 8.0
6.0 6.5 6.0 6.0
x x x x
6.0 6.0 6.0 6.0
1.72 1.21 0.91 1.86
3.46 0.00 0.00 2.66
16.22 11.61 18.38 12.29
8.0 8.0 8.0 8.0 8.0 8.0 8.0 8.0 8.0 0.0
x x x x x x x x x x
8.0 8.0 8.0 8.0 8.0 8.0 8.0 8.0 8.0 0.0
6.0 6.0 5.0 5.7 7.0 7.5 5.7 5.7 6.08 0.62
x x x x x x x x x x
6.0 5.5 5.7 5.0 6.0 6.0 4.7 5.0 5.68 0.48
2.14 1.90 3.63 1.64 1.74 2.17 2.68 0.90 0.28 1.75 0.85
2.31 1.60 4.08 1.51 2.51 4.87 3.99 0.50 3.00 2.35 1.57
17.39 8.40 15.67 11.94 12.57 19.69 20.20 10.52 14.08 14.53 3.70
indeed, result in a partial miss of the prostate tumor volume (2). Because patient setup accuracy can be improved with techniques such as online portal imaging, this problem is largely a result of inherent prostate motion as well as the spatial inaccuracy of the delineated gross tumor volume. The motion of the prostate occurs primarily in the anterior-posterior direction. Lateral motion is small. For this reason, it is particularly useful to monitor the vertical isocentric plane with the ultrasound imaging system because this image will clearly identify any posterior-anterior motion. We have shown that with the aid of diagnostic ultrasound fusion, the gross tumor volume can be localized with an accuracy of approximately 2 mm. When monitoring the ultrasound images, while the transrectal probe was in place, the prostate gland was completely immobile relative to the probe. The physical presence of the probe acts to fixate the prostate gland and eliminate any target motion. The probe was reinserted a number of times to investigate the reproducibility of the position of the prostate on a day to day basis. No perceptible difference in the position of the prostate gland was noticed. The probe also acts as a rectal obturator and physically displaces the posterior rectal wall out of the lateral radiation fields. The treatment volume can be significantly reduced with this degree of target volume localization and fixation. In the design of the treatment plan, leading to the dose distribution appearing in Fig. 8, a margin of 0.5 cm was used to surround the gross tumor volume. A protocol in our institution is being developed that recommends a 1.0 cm margin if there is clinical suspicion of extracapsular disease and a margin of 0.5 cm if not. Figure 8 indicates that the target volume is completely covered by the 80% isodose line and the dose drops from 80 to 10% in the region between the posterior prostate and the anterior rectal wall. The actual prototype design for the technique will involve a sterotactic couch mount for a biplanar ultrasound
probe for use in both the simulation and treatment phases of the therapy. The biplanar probe will be capable of imagiug the coronal and saggital planes, which are to be fused to the anterior-posterior and lateral simulator images. Therefore, the technique described in this work will be applied to the anterior-posterior fields as well as the lateral fields. The couch mount for the probe will be designed so that the ultrasound imaging planes are exactly coincident to the x-ray planes, eliminating the need for a planar coincidence check. Once the two-dimensional contours of the target volume and volumes of interest have been drawn on the ultrasound and x-ray images, they can be imported to our planning system for the development of a full three-dimensional treatment plan. A protocol is under development that specifies that the ultrasound guided therapy will be used only during the cone-down treatment of 10 fractions. The cone-down treatment is to be treated before the first-course treatment. This will reduced any radiation induced sensitivity to the rectum due to the first-course treatment. One of the benefits of such a highly localized treatment is the possibility of investigating the effects of dose escalation on treatment outcome. Distant metastases are more frequently observed in patients who develop local prostatic recurrence after radiotherapy (7, 18). Several investigators have reported that both local control and disease-specific survival is inferior for patients who present with locally advanced prostate T3 and T4 cancer (13, 17). More aggressive local therapy is undertaken in an attempt to improve local control for Stage T3 and T4 disease. Specifically, dose escalation studies have been and continue to be performed. Doses to the prostate in excess of 70 Gy may be required to control bulky, local disease. To date, more aggressive conformal therapy techniques have used images obtained during the simulation to design static treatment portals to aid in the process of dose com-
Ultrasound
image fusion
for external
putation. The prostate location, in relation to the pelvic bones is inconsistent. The prostate position in the pelvis varies daily, mostly in the anterior-posterior plane, depending on rectal and bladder content (2). Therefore, target volumes are increased to ensure inclusion of the prostate volume in the high dose region. This also increases the treatment volume containing normal tissues in the high dose region. CONCLUSIONS Note that the transrectal ultrasound fusion technique has the ability to remedy many of the uncertainties in the treatment of prostate cancer. The technique has the ability to localize the prostate to within 2 mm and to reproducibly fix the prostate gland relative to the probe eliminating target motion. In addition, the probe fixates and localizes the rectum, and most importantly, the anterior rectal wall, to within approximately 0.3 mm. We are currently implementing the technique of ultrasound image fusion in a full three-dimensional, conformal, stereotatic frame work for real time moni-
beam radiotherapy
0 E. J. HOLUPKA
ef al.
4x3
toring and patient repositioning during the actual treatment. In this framework the position of the prostate gland relative to the treatment portal can be monitored in real time during the treatment by perfarming a realtime fusion of the ultrasound image to the portal imaging system. Figure 4 displays a typical high-energy portal image of a lateral field with the transrectal probe in the patient. The probe and transducer array can be seen in the image, and the ultrasound image can be correlated to the portal image using the same technique used for the simulator image. To facilitate such a high precision delivery of dose to the prostate, a sterotactic couch mount for the ultrasound probe is being developed. This mount will localize the ultrasound probe, and the ultrasound image, to the reference frame of the linear accelerator much in the way that stereotatic radiosurgury of the brain is achieved. Because the images are fused, the couch mount will also serve to localize the anatomy, whether it be derived from the ultrasound or simulator images to the reference frame of the linear accelerator. The couch mount will also serve to drastically improve the patient setup accuracy.
REFERENCES 1. Bagshaw, M. A.; Cox, R. S.; Ray, G. R. Status of radiation treatments of prostate cancer at Stanford University. NC1 Monogr. 7:47#; 1988.
2. Beard, C. J.; Bussiere, M. R.; Plunkett, M. E.; Coleman, C. N.; Kijewski, P. K. Analysis of prostate and seminal vesicle motion: Impact for treatment planning. lnt. J. R&at. Oncol. Biol. Phys. 34:451-X58; 19%. 3. Benk, V. A.; Adams, J. A.; Shiplet, W. U. Rectal bleeding
4. 5. 6.
7.
8. 9. 10.
following combined X-ray and proton high dose irradiation for patients with stages T3-T4 prostate carcinoma. Int. J. Radiat. Oncol. Biol. Phys. 26551-557; 1993. Boring, C. C.; Squires, T. S.; Tong, T.; Montgomery, S. Cancer Statistics. Cancer J. Clin. 44:7-26; 1993. Ende, G.; Treuer, H.; Boeseck, R. Optimization and evahtation of Landmarked Based Image Correlation. Phys. Med. Biol. 37:261-271; 1992. Forman, J. 0.; Mesina, C. F.; Hu, T.; Devi, S. B.; Ben-Jasef, E.; Pelizzari, C.; Vijayakumar, ; Chen, G. T. Evaluation of changes in location and shape of the prostate and rectum during a seven week course of conformal radiotherapy (abstr.). lnt. J. Radiat. Oncol. Biol. Phys. 27:222; 1993. Fuks, Z.; Leibel, S. A.; Wallner, K. E.; Begg, C. B.; Fair, W. R.; Anderson, L. L.; Hilaris, B. S.; Whitmore, W. F. The effect of the local control on metastatic dissemination in carcinoma of the prostate: Long term results in patients treated lnt. J. Radiat. Oncol. Biol. Phys. with ‘25l implantation. 21:537-547; 1992. Gilhiuijs, K. G. A.; Touw, A.; van He& M.; Vijlbrief, R. E. Optimization of automatic portal image analysis. Med. Phys 22:1089-1099; 1995. Hanks, G. E.; Martz, K. L.; Diamond, J. J. The effect of dose on local control of prostate cancer (abstr.). lnt. J. Radiat. Oncol. Biol. Phys. 13:1299-1305; 1988. Henry, C. J.; Collins, D. L.; Peters, T. M. Multimodality image integration for stereotatic surgical planning. Med. Phys. 18:167-177; 1991.
Il.
Holupka, E. J.; Kooy, H. M. A geometrical algorithm for medical image correlation. Med. Pbys. 19:433-438,1992. 12. Jiang, H.; Holton, K.; Robb, R. Image registration of multimodabty 3D medical images by chamfer matching. Proc. SPIE, San Jose, CA, ; February 9-14,1992. 13. Kaplan, I. D.; Cox, R. S.; Bagshaw, M. A. Radiothempy for
prostatic cancer: Issues of patient selection and the impact of local control. Urology 433634-639; 1994. 14. Kessle.r,M. L.; Pitluck, S.; &tti, P.; Castro, J. R. Integration of multimodality imalring data far atlning. lnt. J. Radiat Oncol. Biol. 15. Leibel, S. A.; Hanks, G. E.; Kramer, S. Patterns ofcare outcome studies: Results of the national practice in adenocar16.
17.
18.
19.
20.
cinema of the prostate. lnt. J. Radiat. Gncol. Biol. Phys. 10401~; 1984. Pelizzari, C. A.; Chen, G. T. Y. Registration of multiple diagnostic image scans using surface fitting. In: In: The use of cotnputfm in radiation therapy. Amsterdam: EJsevierMorth Holland; 1987:437-440. Perez, C. A.; Lee, H. K.; Georgian, A.; Logsdon, M. D.; Lai, P. P.; Locket& M. A. Techrdcal and tumor related factors affecting outcome of definitive irradiation for locahzed carcinoma of the prostate. lnt. J. Radiat. Oncol. Biol. Phys. 26581-591; 1993. Prestidge, P. R.; Kaplan, I. D.; Cox, R. S.; Bagshaw, M. A. Predictors of survival after a positive post&a&at& prostate biopsy. Int. J. Radiat. OncoJ. Biol. Phys. 28:17-22; 1994. Sandler, H.; McLaughlin, P. W.; Haken, R.; Addison, H.; Forman, J.; Lichter, A. 3D Conformal for the treatment of prostate cancer: Low risk of chronic rectal morbidity observed in a large series of patients @b&r.). lnt. J. Radiat. 0x01. Biol. Phys. 27:135; 1993. Schad,L.;Boeseck,R.;SchlegaL, W.;~G.;Sturm,V.; Stmub, L.; Lorenz, W. Three dimenainnai coriehitton of CT, MR,andPETstudiesinradiotbempy -t planning of brain tumors. J. Comp. Assist. Tomogr. 1 I :948-w, 1987.
984
I. J. Radiation
OncoIogy
l
Bioiogy
l
Physics
21. Schild, S. E.; Casale, H. E.; Bellefontaine, L. P. Movement of the prostate due to rectal and bladder distension: Implications for radiotherapy. Med. Dosim. 18:13-15; 1993. 22. Schultheiss, T. E.; Hanks, G. E.; Hunt, M. A.; Epstein, B.; Peter, R. Factors influencing incidence of acute grade II morbidity in conformal and standard radiation treatment of prostate cancer: Univariate and multivariate analysis (abstr.). Jnt. J. Radiat. Gncol. Biol. Phys. 27:134-135; 1993. 23. Steele, G. D.; Winchester, D. P.; Men& H. R.; Murphy, G. P. National cancer data base: Annual review of patient care. American College of Surgeons; 1993:37-51. 24. Swayne, L. C. Computer assisted fusion of single photon emmision tomographic and computed tomographic images: Evaluation in complicated inflammatory disease. Invest. Radial. 27:78-83; 1992.
Volume 35, Number 5,
19%
25. Syed, A. M.; Puthwola, A. A.; Austin, P. Role of tempory Jridium-198 interstial imphmt in the treatment of carcinoma of the prostate (abstr.). Jnt. J. Radiat. Gncol. Biol. Phys. 191:190; 1991. 26. Ten Haken, R. K.; Forman, J. D.; Heimburger, D. K.; Gerhadsson, A.; McShau, D. L.; Perez-Tamayo, C.; Schoeppel, S. L.; Lichter, A. S. Treatment planning issues related to prostate movement in response to differential tilling of the rectum and bladder. Jnt. J. Radiat. Gncol. Biol. Phys. 2011317-1324; 1991. 27. van den Elsen, P. A.; Pol, E. J.; Viergever, M. A. Medical image matching: A review with classification. IEEE Eng. Med. Biol. 122639; 1993. 28. Vettering, W. T.; Press, W. H.; Flannery, B. P.; Teukolsky, W. T. In: The art of scientific computing. Cambridge, MA: Cambridge University Press; 1988.
Oncology
Bid
Phys.. Vol. 3s. No. 5, pp 975-984, 19% Copyright Q IQ96 Elsevier Science Inc. Prinird in the I!SA. All rights reserved Oi’~O-W16/9h $lS.oo t .(H)
PII: SO360-3016(96)00231-3
l
Physics Original Contribution ULTRASOUND
IMAGE
E. J. HOLUPKA,*
FUSION FOR EXTERNAL BEAM FOR PROSTATE CANCER
I. D. KAPLAN,*
E. C. BUFCDETTE'
AND
RAD3OTHERAPY
G. K. SVENSSON*
*Joint Center for Radiation Therapy and Department of Radiation Oncology, Harvard Medical School. Boston. MA; and ‘Domier Medical Systems, Champaign, IL Purpose: To determine whether real-time uitrasound imaging and targeting system for the treatmeut of prostate cancer was feasible. The initial phase of’ this project included a study to develop and determine (a) &Ware for the fusion of ultrasound images to standard x-rays obtained during simulation, and (b) the potential reduction in field size with real-thne imaging. Methods and Materials: During 13 patient simulations a transrectal uRrasound image was obtatned. Orthogonal x-ray films were acquired with the rectal probe in place. Both the x-ray and llitmmdimageswePe and a R&on image was created of the prostate position in relation to the probe, bktdder, and re&un. The twodimensional area of the rectum, bladder, and prostate was determined in the lateral projection. Potential conformal blocks were designed for the lateral portals in a four-field treatment teclrrdque. Results: The transrectal ultrasound probe enabled real-time prostate hnaging. The lateral Wd sixecan be reduced to 6.1 x 5.68 cm2 t 0.62 x 0.48 cm’ from the standard 8 x 8 cm2 fieid. The posterior rectal wall was physically dispked out of the lateral field. The area of the rectum included in the lateral field is 1.75 cm’ 2 0.85 cm*. Con&sion: The prostate position can be determined with certainty on a regular basis with m tdtrasonograpby. The amount of normal tissue in the high dose volume can be reduced. This approach may reduce acute and chronic morbidity and allow further dose escalation. Ultrasound, Image fusion, Prostate cancer, Conformal radiation therapy.
INTRODUCTION Adenocarcinoma of the prostate is the most commonly diagnosed noncutaneous cancer in the United States male population. It is estimated that 200,000 new cases of prostate cancer will be diagnosed in 1994. Approximately 16% of these newly diagnosed cases will be locally advanced nonmetastatic cancers or, more specifically, AJCC Stage III-T3NOMO. Treatment for patients in this stage is problematic. Several authors have reported significantly low rates of local control with conventional doses of radiation delivered with external beam radiation therapy for clinically locally advanced disease. To improve the rate of local control, more aggressive therapy with conformal techniques and dose escalation is being investigated. Perez et al. reported local recurrence rates of 38% for patients with T3 tumors treated with less 60 Gy and 12% treated with 70 Gy or greater (17). Similar results for improved local control with higher doses were reported in the Pattern of Care Study (9). Strategies to increase the dose to the prostate in excess of 70 Gy are being investigated. Syed et ~1. and Bagshaw et al. have reported im-
proved rates of local control with interstitial 19’Ir implant boosts, which increases the dose to the prostate to approximately 80 Gy (1, 25). Benk and associates from the Massachusets General Hospital report excellent rates of local control in patients with T3 and T4 disease treated with a combination of photon and proton beam therapy to a dose of 75.6 Cobalt Gray Equivalent (3). Several groups have reported on the use of conformal techniques in an attempt to escalate the dose to the prostate with external beam irradiation. Careful dose escalation studies are underway at several institutions to determine the acute and late toxicities of increased dose to an albeit reduced volume of rectum and bladder. Early studies are encouraging. Schultheiss et al. reports that in a multivariate analysis fewer Grade II toxicities were observed in patients treated with conformal therapy (22). Sandler et al. reports low rates of chronic rectal morbidity with a similar conformal approach. Follow-up may be too short to fully ascertain the late complication rates of increased dose to normal tissue in this studies. Benk et al. describes late rectal bleeding with increased dose delivered with a proton boost. The rate of rectal bleeding
Accepted for publication 29 April 1996. Reprint requests to: Dr. Edward J. Holupka. Joint Center for
Radiation Therapy, Department of Radiation Oncology. Harvard Medical School, Boston, MA 02115. 975
976
I. J. RadiationOncologyl Biologyl Physics
Volume 35, Number 5, 1996
of both dose and rectal volume irradiated. Importantly, the rate of rectal bleeding continued to increase with follow-up over 2 years (3). Many of the conformal techniques that are necessary for the safe and proper study of dose escalation requires exact delineation of the target volume. This can be accomplished if the prostate tumor volume is accurately visualized and when precise control or knowledge of those variables that determine the margin around the tumor volume is gained. Extent of microscopic disease, dose-limiting structures such as the anterior rectal wall, and prostate motion are important variables for determining the target volume. Studies have indicated significant motion of the prostate in relation to bony landmarks (2, 21, 26). This variability of the gross target volume makes conformal therapies more difficult to perform. In this study we describe the use of transrectal diagnostic ultrasonography and image fusion to accurately delineate the prostate and rectal wall during the simulation and therapy for prostate cancer. The benefit of diagnostic ultrasound imaging can be best appreciated by spatially correlating the diagnostic information of the ultrasound image to other imaging modalities that have proven to be useful for radiation therapy such as Computerized Tomography, Magnetic Resonance Imaging, and high- and low-energy radiographic x-ray images. Ultrasound medical imaging has a long history, and the present day technology is practically distortion free. For the purposes of this study, any and all distortions of diagnostic ultrasound is negligable. Image correlation methods have recently received much attention for its application in the field of radiation therapy (5, 8, 10-l 1, 14, 16). There are numerous algorithms for matching medical images in both two- and three-dimensional frameworks (5, 10, 12,27). Many of the applications include the multimodality fusion of CT, MRI, Single Photon Emmision Computerized Tomography (SPECT), and x-ray images (14,20,24). However, little work has been done concerning the fusion of diagnostic ultrasound images to other imaging modalities. The purpose of this study is to demonstrate an innovative image fusion technique that allows real time superposition of an ultrasound image of the prostate onto a digital simulator film or high energy portal film. The method is a two-dimensional, landmark-based algorithm that can be extended to correlate the ultrasound image to any imaging modality (5). We will also discuss the potential clinical benefits of using ultrasound image-guided simulation and treatment of the prostate. Our hypothesis is that ultrasound image-guided simulation enhances our
ability for accurate definition of the prostate gross target volume in relation to critical structures such as the anterior rectal wall and the bladder. We also postulate that ultrasound image-guided therapy will increase the spatial treatment accuracy by the direct visualization of the prostate during therapy and, hence, provide the necessary input data for real-time dose optimization and the potential for dose escalation. Furthermore, the use of a rectal ultrasound probe tends to immobilize the prostate in relation to the rectal wall.
‘Urovision Ultrasound System, Dornier Medical Systems, Inc., Kennesaw, GA. *Model ER-S, DomierMedical Systems,Inc., Kennesaw,GA.
‘PercepticsPixel PipelineandPixel Tools, Perceptics,Knoxville, TN.
is observed to increase as a function
METHODS
AND MATERIALS
The ultrasound diagnostic image system’ operates together with a transrectal probe’ operating at 7.5 MHz. The transrectal probe has 96 active transducer elements with a maximum display depth of 13.5 cm. The transrectal probe is approximately 12 cm long by 1.8 cm in diameter at its thickest part. After the standard simulation procedure the transrectal probe is inserted into the patient and diagnostic ultrasound images are obtained. The entire ultrasound session is videotaped and digitized. The transrectal probe is positioned such that the resulting image is in a vertical plane through the isocenter and parallel to the simulator film plane. The simulator image is a projection image that results in the divergent sum of all parallel image planes from the x-ray source to the film plane. Hence, structures that lie above and below the isocentric plane of rotation will appear in the simulator image. This is to be contrasted with the ultrasound image, which is a single planar image. If the prostate is an approximately spherical, midplane structure, this position maximizes the cross-section of the prostate on the image. However, often the prostate tumor volume wraps around the anterior rectal wall, in which case the single plane ultrasound image will not represent the maximum cross-section of the prostate. This situation maybe particularly common when the ultrasound probe pushes the prostate gland anteriorly. To quantitatively asses the three-dimensional shape of the prostate while the probe is in place, a three-dimensional CT scan was obtained with the probe in place. The prostate, rectum, and probe were then reconstructed in three dimensions to determine the position of these structures with respect to each other. The results are displayed in Fig. 1. The prostate does wrap around the rectum and probe. The amount of prostate tissue that wraps around the probe was quantified. Continuing with the ultrasound imaging of the prostate during the simulation session, the maximum width and height of the prostate volume are measured and also marked on the ultrasound. Once the session is over, the videotape of the ultrasound imaging session is digitized, 3 resulting in a dig-
Ultrasound image fusion for external beam radiotherapy
0 E. .I. HOI
L’PKA
ef al.
977
portal film. The probe is oriented so that the ultrasound image plane is parallel to the film plane. Therefore, the ultrasound image of the prostate, which is geom&ically related to the ultrasound probe, can be quantitatively and accurately fused with the simulator or treatment film images. Fusion of the diagnostic ultrasound image with the simulator x-ray image is achieved by using a two-point matching technique that requires the unambiguous identification of two identical point fiducids on each image. The transrectal probe’ uses a sector phased array sweeping method yielding a fan-shaped image. The arc at the top of the ultrasound image directly corresponds to the spatial position of the transducer array inside the probe head. Figure 2 displays a typical transrectal ultrasound image of the prostate. The endpoints of the transducer arc are labeled “1” and “2” in the image. Figures 3 and 4 displays the lateral simulator film and the corresponding lateral portal film with the transrectal probe in place. The transducer array inside the probe head is clearly visible in both imFig. 1. Volume reconstruction from CT scans of the prostate, rectum, and transrectal probe. The transparent nature of the volumes allow the determination of the degree overlap of the prostare volume with the transrectal probe.
ital image of the prostate in the vertical isocentric plane. The lateral radiographic film containing the rectal probe is also digitized.4 The projected outline of the probe can be clearly seen on the simulator film and the high-energy
Fig. 2. A typical ultrasound image of the prostate from the transrectal probe. The lines centered on the crosshairs indicate the dimensions of the prostate determined during the time of the simulation. The two fiducial points used for the fusion are the ends of the transducer array and are labeled as “ 1” and “2.”
Fig. 3. Lateral simulator film displaying the location of the ultrasound transrectal probe. Note how the transducer array is clearly visible in the image. The two fiducial points used for the fusion are the ends of the transducer array and are labeled as “1” and “2.”
----____ ‘TruVel
Scanning
system, TruVel,
Hendron,
VA.
I. J. Radiation Oncology 0 Biology 0 Physics
Volume 35, Number 5, 1996
the ultrasound image to the simulator image, is given by the solution to the linear system of equations given by,
The solution to such linear systems is well known and can be found in any standard text on numerical analysis (28). Once the solution is known, every point in the ultrasound image can be mapped to the simulator image using Eq. (1). Figure 5 displays the result of transforming the ultrasound image to the simulator image. This will be referred to as the fusion image. Figure 2 displays the ultrasound image data along with other useful diagnostic information. ’ However, only the ultrasound image data is transformed so that the details of the underlying radiographic image are not obscured. The accuracy of the rotational and translational component of the transformation is determined by the user’s ability to localize the two point fiducials on each of the ultrasound and radiographic images. The fiducials corre-
Fig. 4. Lateral high-energyportal film displayingthe locationof the ultrasoundtransrectalprobe. Note how the transducerarray is clearly visible in the image.
ages. The ends of the transducer array that correspond to the points on the ultrasound image are likewise labeled “1” and “2.” By zooming in on the images, the points that correspond to the fiducial marks can be determined to within a single pixel. Once the fiducials have been identified, the spatial transformation that maps the ultrasound image to the simulator image is determined. The transformation is assumed to consist of a two-dimensional rotation, translation, and a global isotropic scaling. This leads to a four-parameter transformation,
= A ‘;;$ f$)(;::) + (;;) = (;::), (Eq.1) v rere t, and tYare the translations, 13the rotation angle for
a otation about the center of the image, and A is the global scaling. (x”‘, y”“) and (x6’, y”‘) represent the coordinate pair of any point in the ultrasound and simulator images, respectively. If the two fiducial points on the ultrasound image are labeled as (XT’, y?) and (xz’, y;‘), and the two fiducials points on the simulator film as, (x?, yS’> and (xi’, y;‘), then the matrix M, and translation f which maps
Fig. 5. Fusion image displaying the spatial mapping trasound image data to the simulator image.
of the ul-
Ultrasound image fusion for external
spending to the ultrasound image can be determined automatically because for a given setting of the field of view (FOV), of the ultrasound imaging hardware, the position of the endpoints of the transducer array always appear at the same point, or pixel, in the digitized image. The two point &luciaIs for the ultrasound image can be determined a priori.’ The automatic determination of the fiducials on the simulator image presents more of a problem because the probe does not always appear exactly in the same position on the radiographic image as in the ultrasound image. However, t&e image can be zoomed and the endpoints of the transducer array can be determined to within a pixel. Therefore, the accuracy of the spatial position of the fusion is determined by the pixel size of 0.265 mm for the simulaior image. The ability to afxurately determine the extent of the gross tumor volume and resultant target volume is dependent on the accuracy of the fusion. The accuracy in the position of the hvo fiducial points in the radiographic image is determined a priori by the pixel size of the digital images, as previously discussed. However, the accuracy of the rotational and scaling transformations must be determined. After the ultrasound image is transformed to the simulator image, volumes of interest can be manually contoured on the fusion image. Figure 6 displays the fusion image where the prostate, bladder, and rectal volumes are contoured and color washed over the underlying image. The color-washed representation of the volumes allow for the complete delineation of the volumes while still allowing the structu~ of the underlying image, whether it be the ultrasound, treatment, or simulator image, to show through. In add&ion, the proposed radiation treatment field isocenter, w&h and height can be specified. Figure 7 displays the fusion image with the contoured volumes of interest and proposed radiation field superimposed. In addition to the field outline, blocks, which appear as transparent red regions, have also been drawn. The blocks have been chosen to minimize rectal wall and bladder involvement in the field while ensuring that the tumor volume with a 0.5 to 0.75 cm margin receives 100% of the dose. Once the field and blocks have been drawn, the percent area of pro&&e, bladder, and rectum, which is inside the radiation portal, is automatically calculated and reported on the top of the fusion image. The area of the resultant radiation portal is also calculated and displayed. This quantitative information is available at the time of the simuIatioB or treatment and may prove to be clinically vahble when assessing efficacy and morbidity. To f$ther illustrate the potential of this ultrasound fusion technique, a fulI treatment plan using a four-field box was generated using the simulator-ultrasound fusion image as the starting point. Axial and sagital contours were
sScanitronix
Medicai
AB, Uppsala, Sweden.
beam
radiotherapy l
E. J. HOL~JPKA
er al.
079
obtained for one patient. The planned isooen&-z was obtained from the fusion image and local.&& with respect to both contours. Using dose distributions were fusion image. Figure 8 dispIays prostate, rectum, and bladder with the resultant isodose l&es. Because the to visualize the presence of the h-an investigated. The transrectal probe was placed in a water tank and the dose was measmed a distance of 5.0 cm away from surements were made in the beam. The measurements were repeated wiahow thetransrectal probe and compared. There was no able difference in the dose whether the probe was present or not. The imaging tool described above a.lIows for the t&ion of the ultrasound image to the sequent contouring, field speci and calculation of the degree of vohune of interest involvement iu the field. 6 The uItzasou& iW64OX 480 pixels and the simulator images are 81)o X 1200 pixels. The fusion process takes approximately 5.0 s, and the calculation of the percent of the volume involved in the field takes approximately 1 to 10 s per average volume of interest. RESULTS The amount the prostate wraps around the rectum can influence the posterior extent of the target volm. Figure 1 indicates an overlap of approximatey 15% of the prostate anterior-posterior extent. This value can then be used to extend the field margin in the anterior-posterior direction to ensure adequate coverage of the pro&ate volume. The spatial accuracy of the global scaling transformation can be determined by measuring the length of the drawn axes that measure maximum width and height of the prostate on the fusion image. The maximum width and height of the prostate, indicated by the crossed lines traversing the prostate volume seen on Fig. 2, is determined and recorded during the acquisition of the ultrasound image. For the particular study in Fig. 2 the maximum width is 4.7 cm and the maximum height is 4.0 cm. These same dimensions can also be determined by inspection of the fusion image. The pixel size of the fusion image is 0.275 mm per pixel; therefore, the width and height as determined by the fusion image can de&mined. For this particular study the transformed maximum width and height are 4.9 cm and 4.2 cm, respectively, lead@ to a discrepancy of 2 mm between the ultrasound and fusion image.
%tardent
Vistm 800, Stardent Computer
Inc., Concord. MA.
I. J. Radiation Oncology 0 Biology 0 Physics
Fig. 6. Fusion imagedisplaying the color-washedvolumesof interest.The bladder(green),rectum(blue),andprostate(yellow) are color washedover the imageso as not to obscurethe underlying imagedata.
Volume 35, Number 5, 1996
Fig. 7. Fusion image displaying the color-washedvolumesof interest and the designof the blocks and radiation portal. In addition, the amount of VOI involvement in the radiation is displayedon the top of the image.
The maximum width and height of the prostate for 11 of the 13 patients was determined using both the ultrasound
image and the fusion image. The average and standard deviations for all 13 patients were computed. The average difference is determined to be 2.0 + 2.6 mm standard deviation. Because the error resulting from the pixel size is small (5 0.275 mm) as compared to this difference, we have chosen to interpret this as the average error in the global scaling transformation. Note that because the rectum completely collapses around the probe, the position of the anterior rectal wall can be determined within the accuracy of the simulator or portal film, or 0.275 mm. An additional error in the transformed image results if the ultrasound image plane is noncoplanar with the radiographic film plane. This occurs if the planes are rotated by an angle of B and 4 with respect to each other, where these angles are defined in Fig. 9. The degree of coplanar coincidence of the ultrasound image with respect to the radiographic image is completely determined by the distance between the two point fiducials and the distance of the perpendicular bisector of this line to the arc of the transducer array as measured on the fusion image. If the true distance between the two point fiducials and the perpendicular bisector on the ultrasound image are given by dy$and d$ respectively, and the distance between the two
Fig. 8. Fusionimagedisplayingthe.color washedvolumesof interestandthe isodoselinesfor a pmposed four-fieldbox technique.
Ultrasound image fusion for external beam radiotherapy 0 E. J. HOLUPKA
4
=sizeofloIlgaxis
dUS P, =sireofsholtaxis
cos(e) = 2
dUT pb
COS(#=s 47
Fig. 9. Geometric par;uneters used in the definition of the degree of coplanar coincidence be.hvem the ultrasound and simulator images.
point fiducials and the perpendicular bisector measured on the simulator image are given by d% and d$, respectively, then the degree of planar coincidence is given by,
et al.
981
ment volumes. This comparison is obviously not exact because a total of four fields contribute the true treatment volume; however, it gives a reasonable estimate of the contribution of the lateral fields to the treatment volume. The results indicate that for the 13 patients studied the average area of the actual lateral fields delivered was 64.0 2 2.5 cm’, which is compared to 36.0 + 1.2 cm* for the ultrasound guided fields. This represents a significant reduction in the ultrasound guided treatment volume. For the standard treatment, the amount of rectum treated is difficult to assess because of the uncer&inties in its position during all of the treatment fractions. Because the ultrasound probe serves to fixate both the prostate and rectum, the amount of rectum that would be involved in the radiation field can be determined from the ultrasound fusion image. In columns 4 through 6, the area of the prostate, bladder, and rectum that are directly involved in the field are reported. On average 1.75 cm’ of rectal tissue and 2.35 cm* of the bladder are involved in the lateral radiation field when guided by the fusion technique. The method of ultrasound localization can be used to give a measure of the “goodness” of a treatment by considering the quantitative amount of healty tissue involvement. DISCUSSION
(Eq. 3)
As shown in Fig. 9, the rotation is defined about the center of the line connecting the two point fiducials and, hence, no effect for the divergence of the simulator x-ray field, or equivalently, the distance from the x-ray source to the film plane, is taken into account. This approximation is teasmable bwause the size of the transducer array is small co* to the distance of the source to the film plane. The error in the translational and rotational transformation used in the fusion process does not effect the calculation of the degree of coplanar coincidence because these particular transformations are defined to be in the radiographic plane and do not effect the absolute distance between the fiducial points, for instance, they am orthogonal s. However, the error in the global scaling does effect the absolute distance of the fiducials. The error in the scaling is small compared to the size of the transducer array, and the effect of ignoring this transformation error is small. The degree of coplanar coincidence was mewured for all patients studied. Note that the ultrasound image plane was coincident to the film plane with in an average error or 8 = 9.3” and 4 = 3.4”. A total of 13 patients were studied. Table 1 shows the resnIts for all the patients studied. Column 2 indicates the area of the lateral field portals actually delivered for each patient. Column 3 Micates the area of the lateral field portals as deeed by the fusion image. The difference between these two quantities gives an indication of the difference between the actual and ultrasound guided treat-
At the Joint Center for Radiation Therapy, a four-field box technique is used that is comprised of two opposing laterals, an anterior-posterior and a posterior-anterior field. In the lateral fields, the posterior rectum and anterior bladder are blocked. The Iirst course of therapy uses a 10 x lOcm*toll x 11cm2fieldfollowedbya8X8cm2 to 9 x 9 cm* shaped cone-down field. The simulator image has a high contrast for bony stuchues and to a certain extent, some soft tissue structures. However, the prostate can not be seen in a simulator image and its position must be approximated from neighboring structures, in particular, the bladder, rectum, and neighboring bony structures. Other information, such as size, shape, and approximate relative position can be inferred from a prior CT scan. However, many uncertainties including the gross tumor volume, clinical tumor volume, patient setup uncertainty, and motion of the prostate gland contribute to the definition of a necessarily large planning target volume. The uncertainty in defining the gross target volume (the prostate), the subclinical tumor extension leading to the clinical target vohnne, and finally technical uncertainties such as field edges and organ motion have lead to an approximate margin of 1 S cm around the perceived gross tagret volume. Historically, this margin was thought to safely encompass the gross tumor volume. However, recent studies of patient motion using postal have indiimaging and of organ motion during cated that more attention is needed to determine the margins of uncertainty included in the target volume. A conventional field size and blocking arrangement may,
I. J. Radiation Oncology 0 Biology 0 Physics
982
Table 1. Volume Study 1 2 3 4 2 7 8 9 10 11 12 13 Average Standard deviation
Actual field portal (cm21
involvement
Volume 35, Number 5, 19%
for ultrasound
Proposed field portal (cm*)
guided radiation
portals
Rectum in field (cm2)
Bladder in field (cm2)
Prostate in field (cm21
8.0 8.0 8.0 8.0
x x x x
8.0 8.0 8.0 8.0
6.0 6.5 6.0 6.0
x x x x
6.0 6.0 6.0 6.0
1.72 1.21 0.91 1.86
3.46 0.00 0.00 2.66
16.22 11.61 18.38 12.29
8.0 8.0 8.0 8.0 8.0 8.0 8.0 8.0 8.0 0.0
x x x x x x x x x x
8.0 8.0 8.0 8.0 8.0 8.0 8.0 8.0 8.0 0.0
6.0 6.0 5.0 5.7 7.0 7.5 5.7 5.7 6.08 0.62
x x x x x x x x x x
6.0 5.5 5.7 5.0 6.0 6.0 4.7 5.0 5.68 0.48
2.14 1.90 3.63 1.64 1.74 2.17 2.68 0.90 0.28 1.75 0.85
2.31 1.60 4.08 1.51 2.51 4.87 3.99 0.50 3.00 2.35 1.57
17.39 8.40 15.67 11.94 12.57 19.69 20.20 10.52 14.08 14.53 3.70
indeed, result in a partial miss of the prostate tumor volume (2). Because patient setup accuracy can be improved with techniques such as online portal imaging, this problem is largely a result of inherent prostate motion as well as the spatial inaccuracy of the delineated gross tumor volume. The motion of the prostate occurs primarily in the anterior-posterior direction. Lateral motion is small. For this reason, it is particularly useful to monitor the vertical isocentric plane with the ultrasound imaging system because this image will clearly identify any posterior-anterior motion. We have shown that with the aid of diagnostic ultrasound fusion, the gross tumor volume can be localized with an accuracy of approximately 2 mm. When monitoring the ultrasound images, while the transrectal probe was in place, the prostate gland was completely immobile relative to the probe. The physical presence of the probe acts to fixate the prostate gland and eliminate any target motion. The probe was reinserted a number of times to investigate the reproducibility of the position of the prostate on a day to day basis. No perceptible difference in the position of the prostate gland was noticed. The probe also acts as a rectal obturator and physically displaces the posterior rectal wall out of the lateral radiation fields. The treatment volume can be significantly reduced with this degree of target volume localization and fixation. In the design of the treatment plan, leading to the dose distribution appearing in Fig. 8, a margin of 0.5 cm was used to surround the gross tumor volume. A protocol in our institution is being developed that recommends a 1.0 cm margin if there is clinical suspicion of extracapsular disease and a margin of 0.5 cm if not. Figure 8 indicates that the target volume is completely covered by the 80% isodose line and the dose drops from 80 to 10% in the region between the posterior prostate and the anterior rectal wall. The actual prototype design for the technique will involve a sterotactic couch mount for a biplanar ultrasound
probe for use in both the simulation and treatment phases of the therapy. The biplanar probe will be capable of imagiug the coronal and saggital planes, which are to be fused to the anterior-posterior and lateral simulator images. Therefore, the technique described in this work will be applied to the anterior-posterior fields as well as the lateral fields. The couch mount for the probe will be designed so that the ultrasound imaging planes are exactly coincident to the x-ray planes, eliminating the need for a planar coincidence check. Once the two-dimensional contours of the target volume and volumes of interest have been drawn on the ultrasound and x-ray images, they can be imported to our planning system for the development of a full three-dimensional treatment plan. A protocol is under development that specifies that the ultrasound guided therapy will be used only during the cone-down treatment of 10 fractions. The cone-down treatment is to be treated before the first-course treatment. This will reduced any radiation induced sensitivity to the rectum due to the first-course treatment. One of the benefits of such a highly localized treatment is the possibility of investigating the effects of dose escalation on treatment outcome. Distant metastases are more frequently observed in patients who develop local prostatic recurrence after radiotherapy (7, 18). Several investigators have reported that both local control and disease-specific survival is inferior for patients who present with locally advanced prostate T3 and T4 cancer (13, 17). More aggressive local therapy is undertaken in an attempt to improve local control for Stage T3 and T4 disease. Specifically, dose escalation studies have been and continue to be performed. Doses to the prostate in excess of 70 Gy may be required to control bulky, local disease. To date, more aggressive conformal therapy techniques have used images obtained during the simulation to design static treatment portals to aid in the process of dose com-
Ultrasound
image fusion
for external
putation. The prostate location, in relation to the pelvic bones is inconsistent. The prostate position in the pelvis varies daily, mostly in the anterior-posterior plane, depending on rectal and bladder content (2). Therefore, target volumes are increased to ensure inclusion of the prostate volume in the high dose region. This also increases the treatment volume containing normal tissues in the high dose region. CONCLUSIONS Note that the transrectal ultrasound fusion technique has the ability to remedy many of the uncertainties in the treatment of prostate cancer. The technique has the ability to localize the prostate to within 2 mm and to reproducibly fix the prostate gland relative to the probe eliminating target motion. In addition, the probe fixates and localizes the rectum, and most importantly, the anterior rectal wall, to within approximately 0.3 mm. We are currently implementing the technique of ultrasound image fusion in a full three-dimensional, conformal, stereotatic frame work for real time moni-
beam radiotherapy
0 E. J. HOLUPKA
ef al.
4x3
toring and patient repositioning during the actual treatment. In this framework the position of the prostate gland relative to the treatment portal can be monitored in real time during the treatment by perfarming a realtime fusion of the ultrasound image to the portal imaging system. Figure 4 displays a typical high-energy portal image of a lateral field with the transrectal probe in the patient. The probe and transducer array can be seen in the image, and the ultrasound image can be correlated to the portal image using the same technique used for the simulator image. To facilitate such a high precision delivery of dose to the prostate, a sterotactic couch mount for the ultrasound probe is being developed. This mount will localize the ultrasound probe, and the ultrasound image, to the reference frame of the linear accelerator much in the way that stereotatic radiosurgury of the brain is achieved. Because the images are fused, the couch mount will also serve to localize the anatomy, whether it be derived from the ultrasound or simulator images to the reference frame of the linear accelerator. The couch mount will also serve to drastically improve the patient setup accuracy.
REFERENCES 1. Bagshaw, M. A.; Cox, R. S.; Ray, G. R. Status of radiation treatments of prostate cancer at Stanford University. NC1 Monogr. 7:47#; 1988.
2. Beard, C. J.; Bussiere, M. R.; Plunkett, M. E.; Coleman, C. N.; Kijewski, P. K. Analysis of prostate and seminal vesicle motion: Impact for treatment planning. lnt. J. R&at. Oncol. Biol. Phys. 34:451-X58; 19%. 3. Benk, V. A.; Adams, J. A.; Shiplet, W. U. Rectal bleeding
4. 5. 6.
7.
8. 9. 10.
following combined X-ray and proton high dose irradiation for patients with stages T3-T4 prostate carcinoma. Int. J. Radiat. Oncol. Biol. Phys. 26551-557; 1993. Boring, C. C.; Squires, T. S.; Tong, T.; Montgomery, S. Cancer Statistics. Cancer J. Clin. 44:7-26; 1993. Ende, G.; Treuer, H.; Boeseck, R. Optimization and evahtation of Landmarked Based Image Correlation. Phys. Med. Biol. 37:261-271; 1992. Forman, J. 0.; Mesina, C. F.; Hu, T.; Devi, S. B.; Ben-Jasef, E.; Pelizzari, C.; Vijayakumar, ; Chen, G. T. Evaluation of changes in location and shape of the prostate and rectum during a seven week course of conformal radiotherapy (abstr.). lnt. J. Radiat. Oncol. Biol. Phys. 27:222; 1993. Fuks, Z.; Leibel, S. A.; Wallner, K. E.; Begg, C. B.; Fair, W. R.; Anderson, L. L.; Hilaris, B. S.; Whitmore, W. F. The effect of the local control on metastatic dissemination in carcinoma of the prostate: Long term results in patients treated lnt. J. Radiat. Oncol. Biol. Phys. with ‘25l implantation. 21:537-547; 1992. Gilhiuijs, K. G. A.; Touw, A.; van He& M.; Vijlbrief, R. E. Optimization of automatic portal image analysis. Med. Phys 22:1089-1099; 1995. Hanks, G. E.; Martz, K. L.; Diamond, J. J. The effect of dose on local control of prostate cancer (abstr.). lnt. J. Radiat. Oncol. Biol. Phys. 13:1299-1305; 1988. Henry, C. J.; Collins, D. L.; Peters, T. M. Multimodality image integration for stereotatic surgical planning. Med. Phys. 18:167-177; 1991.
Il.
Holupka, E. J.; Kooy, H. M. A geometrical algorithm for medical image correlation. Med. Pbys. 19:433-438,1992. 12. Jiang, H.; Holton, K.; Robb, R. Image registration of multimodabty 3D medical images by chamfer matching. Proc. SPIE, San Jose, CA, ; February 9-14,1992. 13. Kaplan, I. D.; Cox, R. S.; Bagshaw, M. A. Radiothempy for
prostatic cancer: Issues of patient selection and the impact of local control. Urology 433634-639; 1994. 14. Kessle.r,M. L.; Pitluck, S.; &tti, P.; Castro, J. R. Integration of multimodality imalring data far atlning. lnt. J. Radiat Oncol. Biol. 15. Leibel, S. A.; Hanks, G. E.; Kramer, S. Patterns ofcare outcome studies: Results of the national practice in adenocar16.
17.
18.
19.
20.
cinema of the prostate. lnt. J. Radiat. Gncol. Biol. Phys. 10401~; 1984. Pelizzari, C. A.; Chen, G. T. Y. Registration of multiple diagnostic image scans using surface fitting. In: In: The use of cotnputfm in radiation therapy. Amsterdam: EJsevierMorth Holland; 1987:437-440. Perez, C. A.; Lee, H. K.; Georgian, A.; Logsdon, M. D.; Lai, P. P.; Locket& M. A. Techrdcal and tumor related factors affecting outcome of definitive irradiation for locahzed carcinoma of the prostate. lnt. J. Radiat. Oncol. Biol. Phys. 26581-591; 1993. Prestidge, P. R.; Kaplan, I. D.; Cox, R. S.; Bagshaw, M. A. Predictors of survival after a positive post&a&at& prostate biopsy. Int. J. Radiat. OncoJ. Biol. Phys. 28:17-22; 1994. Sandler, H.; McLaughlin, P. W.; Haken, R.; Addison, H.; Forman, J.; Lichter, A. 3D Conformal for the treatment of prostate cancer: Low risk of chronic rectal morbidity observed in a large series of patients @b&r.). lnt. J. Radiat. 0x01. Biol. Phys. 27:135; 1993. Schad,L.;Boeseck,R.;SchlegaL, W.;~G.;Sturm,V.; Stmub, L.; Lorenz, W. Three dimenainnai coriehitton of CT, MR,andPETstudiesinradiotbempy -t planning of brain tumors. J. Comp. Assist. Tomogr. 1 I :948-w, 1987.
984
I. J. Radiation
OncoIogy
l
Bioiogy
l
Physics
21. Schild, S. E.; Casale, H. E.; Bellefontaine, L. P. Movement of the prostate due to rectal and bladder distension: Implications for radiotherapy. Med. Dosim. 18:13-15; 1993. 22. Schultheiss, T. E.; Hanks, G. E.; Hunt, M. A.; Epstein, B.; Peter, R. Factors influencing incidence of acute grade II morbidity in conformal and standard radiation treatment of prostate cancer: Univariate and multivariate analysis (abstr.). Jnt. J. Radiat. Gncol. Biol. Phys. 27:134-135; 1993. 23. Steele, G. D.; Winchester, D. P.; Men& H. R.; Murphy, G. P. National cancer data base: Annual review of patient care. American College of Surgeons; 1993:37-51. 24. Swayne, L. C. Computer assisted fusion of single photon emmision tomographic and computed tomographic images: Evaluation in complicated inflammatory disease. Invest. Radial. 27:78-83; 1992.
Volume 35, Number 5,
19%
25. Syed, A. M.; Puthwola, A. A.; Austin, P. Role of tempory Jridium-198 interstial imphmt in the treatment of carcinoma of the prostate (abstr.). Jnt. J. Radiat. Gncol. Biol. Phys. 191:190; 1991. 26. Ten Haken, R. K.; Forman, J. D.; Heimburger, D. K.; Gerhadsson, A.; McShau, D. L.; Perez-Tamayo, C.; Schoeppel, S. L.; Lichter, A. S. Treatment planning issues related to prostate movement in response to differential tilling of the rectum and bladder. Jnt. J. Radiat. Gncol. Biol. Phys. 2011317-1324; 1991. 27. van den Elsen, P. A.; Pol, E. J.; Viergever, M. A. Medical image matching: A review with classification. IEEE Eng. Med. Biol. 122639; 1993. 28. Vettering, W. T.; Press, W. H.; Flannery, B. P.; Teukolsky, W. T. In: The art of scientific computing. Cambridge, MA: Cambridge University Press; 1988.