A deformable head and neck phantom with in-vivo dosimetry for adaptive radiotherapy quality assurance Yan Jiang Graves, Arthur-Allen Smith, David Mcilvena, Zherrina Manilay, Yuet Kong Lai, Roger Rice, Loren Mell, Xun Jia, Steve B. Jiang, and Laura Cerviño Citation: Medical Physics 42, 1490 (2015); doi: 10.1118/1.4908205 View online: http://dx.doi.org/10.1118/1.4908205 View Table of Contents: http://scitation.aip.org/content/aapm/journal/medphys/42/4?ver=pdfcov Published by the American Association of Physicists in Medicine Articles you may be interested in A GPU based high-resolution multilevel biomechanical head and neck model for validating deformable image registration Med. Phys. 42, 232 (2015); 10.1118/1.4903504 A three-dimensional head-and-neck phantom for validation of multimodality deformable image registration for adaptive radiotherapy Med. Phys. 41, 121709 (2014); 10.1118/1.4901523 A virtual phantom library for the quantification of deformable image registration uncertainties in patients with cancers of the head and neck Med. Phys. 40, 111703 (2013); 10.1118/1.4823467 The need for application-based adaptation of deformable image registration Med. Phys. 40, 011702 (2013); 10.1118/1.4769114 A method to evaluate dose errors introduced by dose mapping processes for mass conserving deformations Med. Phys. 39, 2119 (2012); 10.1118/1.3684951
A deformable head and neck phantom with in-vivo dosimetry for adaptive radiotherapy quality assurance Yan Jiang Graves Center for Advanced Radiotherapy Technologies and Department of Radiation Medicine and Applied Sciences, University of California San Diego, La Jolla, California 92037-0843 and Department of Physics, University of California San Diego, La Jolla, California 92093
Arthur-Allen Smith, David Mcilvena, Zherrina Manilay, and Yuet Kong Lai Department of Mechanical and Aerospace Engineering, University of California San Diego, La Jolla, California 92093
Roger Rice and Loren Mell Center for Advanced Radiotherapy Technologies and Department of Radiation Medicine and Applied Sciences, University of California San Diego, La Jolla, California 92037-0843
Xun Jia and Steve B. Jianga) Center for Advanced Radiotherapy Technologies and Department of Radiation Medicine and Applied Sciences, University of California San Diego, La Jolla, California 92037-0843 and Department of Radiation Oncology, University of Texas Southwestern Medical Center, Dallas, Texas 75235
Laura Cerviñoa) Center for Advanced Radiotherapy Technologies and Department of Radiation Medicine and Applied Sciences, University of California San Diego, La Jolla, California 92037-0843
(Received 22 January 2014; revised 2 September 2014; accepted for publication 27 January 2015; published 11 March 2015) Purpose: Patients’ interfractional anatomic changes can compromise the initial treatment plan quality. To overcome this issue, adaptive radiotherapy (ART) has been introduced. Deformable image registration (DIR) is an important tool for ART and several deformable phantoms have been built to evaluate the algorithms’ accuracy. However, there is a lack of deformable phantoms that can also provide dosimetric information to verify the accuracy of the whole ART process. The goal of this work is to design and construct a deformable head and neck (HN) ART quality assurance (QA) phantom with in vivo dosimetry. Methods: An axial slice of a HN patient is taken as a model for the phantom construction. Six anatomic materials are considered, with HU numbers similar to a real patient. A filled balloon inside the phantom tissue is inserted to simulate tumor. Deflation of the balloon simulates tumor shrinkage. Nonradiopaque surface markers, which do not influence DIR algorithms, provide the deformation ground truth. Fixed and movable holders are built in the phantom to hold a diode for dosimetric measurements. Results: The measured deformations at the surface marker positions can be compared with deformations calculated by a DIR algorithm to evaluate its accuracy. In this study, the authors selected a Demons algorithm as a DIR algorithm example for demonstration purposes. The average error magnitude is 2.1 mm. The point dose measurements from the in vivo diode dosimeters show a good agreement with the calculated doses from the treatment planning system with a maximum difference of 3.1% of prescription dose, when the treatment plans are delivered to the phantom with original or deformed geometry. Conclusions: In this study, the authors have presented the functionality of this deformable HN phantom for testing the accuracy of DIR algorithms and verifying the ART dosimetric accuracy. The authors’ experiments demonstrate the feasibility of this phantom serving as an end-to-end ART QA phantom. C 2015 American Association of Physicists in Medicine. [http://dx.doi.org/10.1118/1.4908205] Key words: deformable registration, adaptive radiotherapy, deformable phantom, deformable registration verification, ART verification 1. INTRODUCTION The goal of radiation therapy is to irradiate the tumor target with a prescribed dose while sparing the nearby normal tissues. For this purpose, a treatment plan is generated based 1490
Med. Phys. 42 (4), April 2015
on the patient’s anatomy captured prior to the treatment. The patient’s anatomy may, however, change during the whole treatment course. This anatomical variation can compromise the initial treatment plan quality and therefore treatment outcomes.1,2 Adaptive radiotherapy (ART) has been proposed
0094-2405/2015/42(4)/1490/8/$30.00
© 2015 Am. Assoc. Phys. Med.
1490
1491
Graves et al.: Deformable phantom with dosimetry for ART QA
to overcome this challenge by modifying or reoptimizing the treatment plan based on the patient’s new anatomy in order to maintain the plan quality.3–10 Deformable image registration (DIR) algorithm is a crucial tool for ART. It establishes the correspondence between voxels in a reference image (e.g., the treatment planning image) and those in a target image (e.g., daily acquired image). DIR has been, for instance, wildly used to propagate the contours of the tumor target and the organs at risk (OARs) from the initial planning computed tomography (CT) to the daily patient images.11–13 DIR has also been used for dose accumulation, where the delivered dose is reconstructed on the daily patient image and transferred back to the reference geometry image according to voxel correspondences derived from a DIR algorithm.11 The delivered dose accumulation helps the physician to evaluate the treatment and decide when and how a replan or plan modification is needed. In light of these applications, the accuracy of DIR algorithm has a significant impact for a safe and accurate implementation of the ART technology in clinics. Over the years, a lot of efforts have been devoted to developing accurate DIR algorithms,14–21 as well as to finding appropriate ways to evaluate their accuracy. Common evaluation methods include comparing the calculated deformation vector fields (DVFs) with that derived based on landmark points,16,17,20,22 calculating image similarity metrics,16,21,23 and inspecting contours.24 Phantom-based deformable registration validation has also received much attention. Physical deformation displacements can be directly derived and compared with the calculated ones. Several deformable phantoms have been built for this purpose. Kashani et al.25 and Serban et al.26 designed two different lung phantoms that can mimic diaphragm motion and deform the modeled lung tissue. They performed quantitative evaluation of DIR algorithms based on the embedded landmarks in the phantoms. The landmarks are usually visible in the images and hence lead to good registration locally, but registration errors elsewhere, especially in homogeneous regions, are hard to identify and evaluate. A research group at University
1491
of California, San Francisco (UCSF) designed two anthropomorphic deformable phantoms, i.e., a head and neck (HN) phantom and a pelvic phantom.27,28 Both phantoms resemble the corresponding tumor sites closely. A grid of nonradiopaque surface makers is placed in the phantoms. These markers can be captured by optical cameras but are invisible in the CT images for DIR algorithms. The actual deformation can be extracted from the optical camera images and used to verify the DIR algorithm’s output. As pointed out by Yan,29 a clinical quality assurance (QA) workflow for ART demands (1) DIR algorithm accuracy tests and (2) delivered dose verification. So far there is a lack of deformable phantoms that also provide dosimetric information for the complete ART QA purpose. The goal of this study is to design and construct an ART QA phantom that could perform both of these. We selected HN as the anatomical site, because some HN tumors respond dramatically to the radiotherapy treatment and shrink through the treatment course, greatly benefiting from ART. This ART phantom can be implemented in clinical HN ART workflow to QA the whole ART procedure and technologies. The first part of the paper focuses on the design and fabrication of the phantom, while the second part demonstrates the feasibility of using this phantom to verify the DIR algorithm’s accuracy and to serve as an end-to-end ART dosimetry QA phantom.
2. METHODS AND MATERIALS 2.A. Phantom design and fabrication
The geometry of this phantom is determined based on a single axial CT slice of a typical HN patient who is likely to require ART during the treatment [Fig. 1(A)]. The key organs found in this slice are mandible, cord, tumor, parotids, pharynx, skin, and soft tissue, including dense soft tissue (i.e., muscle) and light soft tissue (i.e., fat). The PTV and the parotids contours are displayed on the patient CT slice. The thickness of the phantom along the superior–inferior
F. 1. (A) CT image of modeled HN patient with PTV and parotids contours. (B) Phantom construction diagram with marked diode holder locations (1–4). The dimensions are in cm. Medical Physics, Vol. 42, No. 4, April 2015
1492
Graves et al.: Deformable phantom with dosimetry for ART QA
(SI) direction is 4.5 cm, and its axial geometry remains unchanged throughout that thickness. Therefore, this phantom is considered as a 2D phantom. The phantom consists of two parallel acrylic plates with the deformable tissue sandwiched between them. To make the phantom realistic, the deformation material is selected such that they have similar x-ray attenuation properties to that in the human body. Specifically, gypsum plaster is used for casting the bony anatomy, i.e., mandible and spine, and they are rigidly attached to the acrylic plates with small nylon bolts. Soft tissue is made of silicone rubber gel and microballoons filler. The base silicon rubber gel represents the dense soft tissue, and the microballoons filler is added into silicon rubber gel to create lighter weight casting. Parotids and light soft tissue are mixtures of base silicone gel with 10% and 20% of microballoons filler, respectively. The pharynx is represented by a nylon tube, which creates an air cavity. Urethane rubber is painted on the soft tissue surface to simulate skin. A latex balloon is inserted in the deformable tissue and glued to a catheter that goes outside the acrylic plates, shown in Fig. 4. A syringe is attached to the balloon catheter so that the balloon volume can be adjusted to simulate tumor shrinkage by modifying the amount of salt water injected. The concentration of the salt water inside the balloon is adjusted, so that it has a similar HU value to the surrounding tissue. On the two sides of deformable material surfaces adjacent to the acrylic plates, small glow-in-dark nonradiopaque marker dots are glued to form a 2D grid with approximately 1.5 cm resolution. The markers are visible to the optical camera through the clear acrylic plates, but they do not appear on CT images. Middle size nylon bolts are installed at the four corners of the acrylic plates to hold the deformable tissue in place. Four glow-in-the-dark markers are glued to the bolt tops as calibration markers to assist CT to optical images alignment and surface marker positions derivation. In order to minimize the constraining effect from the acrylic plates, an even layer of water-based lubricant is applied to the acrylic plates in order to create frictionless contact between the plates and the deformable material surfaces. To acquire the in vivo dosimetry, we selected SFD stereotactic dosimetry diode (IBA Dosimetry, Barlett, TN) as our detector due to its small volume size. The detector holders are made of acrylic and designed for the SFD diode dimension specifically. There are four diode detector holders installed in the deformable phantom, as shown in Fig. 1(B). Two of the detectors are fixed near the mandible (diode 1) and inside the cord (diode 2). The other two, which are installed on rails and can move with tissue deformation, are close to the tumor (diode 3) and inside the left parotid (diode 4). For the installation of the rails, we first estimated the tissue deformation direction from the tumor shrinkage at the locations of diode 3 and diode 4 and then installed the rail tracks along the deformation direction. The diode holders are designed to freely move on the rail track and the design diagram is shown in Fig. 2. The rail tracks and moving components are made of plastic. The diode holder is constructed such that, when the diode is placed, its tip aligns with the middle plane of the deformable phantom. Medical Physics, Vol. 42, No. 4, April 2015
1492
F. 2. Design diagram of the diode holders on rail.
2.B. Phantom deformation and data acquisition
A syringe is attached to the balloon catheter outside the acrylic plate. To represent the original large tumor volume, 160 ml salt water is initially injected into the latex balloon catheter. After charging the phosphorescent surface markers, optical camera images of a pixel size 0.33 mm are acquired in the dark for the markers on both sides of the phantom. To simulate tumor shrinkage, we deflate the balloon volume to 65 ml by withdrawing the salt water out of the balloon with the syringe. Both diode 3 and diode 4 holders move closer to the tumor along the rails with the tissue deformation. Again optical images of both sides of the phantom are acquired at the same distance as for the original geometry phantom. Since the phantom mimics tissue deformations from tumor shrinkage, the terms before and after correspond to an inflated and deflated latex balloon, respectively. Two sets of CT images are obtained for the phantom before and after the deformation using a GE CT scanner (GE Four-Slice Qxi LightSpeed CT, GE Healthcare). The CT resolution is 0.98 mm and the slice thickness is 2.5 mm. 2.C. DIR accuracy experiment
The grids of small nonradiopaque markers on both deformable phantom surfaces provide the ground truth to verify the accuracy of the DVFs calculated from DIR algorithms. Optical images of both sides of the phantom are taken before and after tissue deformation. An in-house program is developed to detect the physical marker dots in the optical images by applying a threshold value in the green channel of the RGB optical images. The origin of the position coordinate system aligns with the most anterior-right calibration marker. y and x denote the anterior and patient-left directions, respectively. The calibration markers on the nylon bolts are known to form a square with 29.7 cm sides. This information is then utilized to calculate the x and y coordinates of the initial and the final centroid positions of the surface marker dots from the optical images. Finally, the deformation motion vectors are averaged from both sides of phantom surface markers and considered as the measured deformation ground truth, which is represented as − ν⇀m . To test the accuracy of a DIR algorithm, the measured deformation is compared to the algorithm’s output. Under
1493
Graves et al.: Deformable phantom with dosimetry for ART QA
1493
the assumption that the phantom is 2D, i.e., there is no deformation along the SI direction, the middle CT slice is used in the DIR algorithms. The CT slice with the large tumor before simulated tumor shrinkage is considered as the reference image, and the CT slice with the small tumor is the moving image. The calculated DVFs transfer the moving image to the reference image. DVFs at 94 marker locations are extracted and denoted as − ν−⇀ cal. In this study, we selected a GPUbased Demons algorithm20 as our DIR algorithm example and reported the average error between the measured deformation and the calculated one as N −−⇀ −−−⇀ ∥ νi, m − νi,cal ∥ , (1) δ = l=1 N where N = 94. Note that the main focus of this study is to demonstrate the feasibility of utilizing the surface markers for the DIR algorithm accuracy test, not to investigate the DIR algorithm accuracy. Users can also choose any comparison metrics that they believe are best for the DIR evaluation. 2.D. Diode angular dependence properties and dosimetry calibration
F. 4. Prototype of the deformable HN phantom.
A conventional setup for the diode use is to align its long axis parallel to the beam axis. However, in our application, the diode is inserted into the phantom such that its long axis is perpendicular to the beam axis. Hence, before adopting this
setup, the angular dependence of this SFD diode should be evaluated. The angular dependence is defined as the response of the diode from different beam angles around the diode’s long axis under the isocentric condition, normalized to the response from the reference condition. The reference condition refers to the isocentric setup with gantry angle at zero degrees. In our experiment, an acrylic buildup cap is designed, shown in Fig. 3(A) and built to increase dose depositions to the diode for this angular dependence investigation. The reference response is obtained when 100 MU radiation is delivered with a 5×5 cm2 field size. Different diode responses with 100 MU radiation from seven coplanar equiangular angles (i.e., at 0◦, 51◦, 102◦, 153◦, 207◦, 258◦, and 309◦) are measured and are normalized to the reference response. The beam angles with respect to the orientation of the diode are illustrated in Fig. 3(B). The diode dosimeter reading needs to be calibrated before conducting dosimetric validations using the proposed phantom. Specifically, the diode calibration is performed at
F. 3. (A) Diagram of the buildup cap. Measurements are in mm. (B) Orientation of beam angles with respect to the diode.
F. 5. (A) CT slice at the middle of the phantom before tumor shrinkage. (B) CT slice at the middle of the phantom after tumor shrinkage.
Medical Physics, Vol. 42, No. 4, April 2015
1494
Graves et al.: Deformable phantom with dosimetry for ART QA
1494
T I. Average HUs for phantom and patients’ components.
Bone
Dense soft tissue
Light soft tissue
Parotids
865|920
140|100
−35| − 50
90|65
Components Phantom|patient
the isocentric position for a 10 × 10 cm2 field size at 10 cm depth in a solid water phantom with 100 MU 6 MV beam radiation delivered. The calibration factor f cal, which converts the electrometer reading to dose in water, is obtained from D0 , (2) R0 where R0 and D0 are the electrometer reading and dose to water under the calibration condition. Here, D0 is derived based on TMR ratio at depth of 10 cm for the same condition. It is also found that the diode sensitivity slightly depends on the radiation history. Therefore, the diode is frequently calibrated every time right before a dosimetric verification experiment. Once the calibration coefficient f cal is obtained, in a dose measurement, we can obtain the absolute dose value Dm as f cal =
Dm = f cal × Rm ,
(3)
where Rm is the electrometer reading of the diode response in the measurement.
F. 6. (A) Marker positions for large tumor geometry phantom before tumor shrinkage. (B) Marker positions for small tumor geometry phantom after tumor shrinkage. (C) The measured DVFs overlaid on CT slice of the large tumor phantom. (D) The DIR generated DVFs overlaid on CT slice of the large tumor phantom.
2.E. ART dosimetry experiment
To serve as an end-to-end ART QA phantom, this deformable phantom is treated exactly in the same way as for a real patient, going through an initial CT scan, treatment planning, plan delivery, geometric deformation, re-CT scan, replanning, and replan delivery procedure. First, after the CT image is acquired for the phantom with a large tumor, the PTV and organ contours are drawn. We deliberately extend the tumor contour to include diode 3 inside the PTV, so that it provides the dose measurement inside the PTV target. Then, an IMRT plan with a set of seven coplanar equiangular 6 MV beams is generated on the large tumor geometry using Eclipse (Varian Medical, Inc., Palo Alto, CA) treatment planning system (TPS). Clinically relevant constraints to organs are satisfied in the plan. After the phantom is properly set up on
the treatment couch using onboard imaging (OBI) system for image guidance, the plan is delivered to the phantom four times, each time with the diode placed in a different diode holder and plastic inserts in the rest of the diode holders. The four diode readings are obtained and compared to the calculated results from the TPS. After tumor shrinkage, the PTV and organ contours are redrawn in the new CT image. With the same beam angles and plan constraints, the new geometry is used for replanning in Eclipse. Then, both the reoptimized plan and the initial plan are delivered to the new phantom geometry in order to evaluate the dosimetric gain from ART replanning. The diode readings are compared with the calculation from Eclipse to verify ART accuracy.
F. 7. (A) Difference image between the reference and the moving image. (B) Difference between the reference and the deformed image. Medical Physics, Vol. 42, No. 4, April 2015
1495
Graves et al.: Deformable phantom with dosimetry for ART QA
1495
T II. Mean and standard deviation of the average error for DIR output compared to the measured ground truth. A–P (mm) 1.0 ± 1.0
L–R (mm)
δ (mm)
1.5 ± 2.0
2.1 ± 2.2
3. RESULTS 3.A. Deformable HN phantom
The prototype of this deformable HN phantom is presented in Fig. 4. The deformable phantom consists of six materials that represent bone, shrinking tumor, air cavity, parotids, and soft tissue including light and denser tissue. Figures 5(A) and 5(B) show the CT slices at the middle of the deformable phantom before and after the tumor shrinkage, with each anatomical components easily observed. The PTV contour, as well as other organ contours, is displayed, and the diode detector locations are shown in Fig. 5. The average HU values from the modeled HN patient for bone, dense and light soft tissue, and parotids are summarized in Table I. For each phantom organ material, we are able to reproduce the similar HU values to that in the real patient and keep the reasonable relative HU difference between any two kinds.
F. 8. Graph of diode response at different angles against response at zero degree gantry angle.
the vector field is also summarized in Table II. The average error magnitude at the 94 surface markers is 2.1 mm, with a larger variance for left–right (L–R) direction than for the anterior–posterior (A–P) direction.
3.B. DIR accuracy evaluation
3.C. Diode angular dependence
Figures 6(A) and 6(B) display the detected glow-in-thedark surface marker positions before and after the simulated tumor shrinkage. The measured marker motion vector fields and the Demons generated DVFs are overlaid on the CT slice of the large tumor phantom in Figs. 6(C) and 6(D), respectively. Good visual agreement between the measured deformation and the DIR output is observed. The CT slice with the large tumor before the shrinkage is considered as the reference image, and the slice with the small tumor is the moving image. The moving image is registered to the reference image to produce the deformed image. The DIR performance is qualitatively evaluated by comparing the difference images between the reference and the moving images [Fig. 7(A)] with the difference image between the reference and the deformed ones [Fig. 7(B)]. It can be seen that the difference image before registration is relatively large in intensity. After the deformable registration, the intensity in the difference image is greatly reduced, indicating the accuracy of the registration. Moreover, the mean pixel difference drops from 2% to 0.038% of the maximum reference image intensity after deformable registration. The quantitative difference of
The response at different gantry angles is averaged from two rotations of radiation delivery and the result values are plotted in Fig. 8. The greatest asymmetry appears at 207◦ angle, at which the response is 1.36% lower than the reference value. Derivations from the reference response at other angles are within 0.5%. Because of the low angular dependence of the diode response, we will ignore this issue when using it for in vivo measurement inside of the phantom with its long axes being set perpendicular to the beam axis. 3.D. ART dosimetry evaluation
Before conducting dosimetric validations using this proposed phantom, the diode dosimeter reading is calibrated. The error expected for this diode is 0.97%. To evaluate the dosimetric effect from the tumor shrinkage, we deliver the initial optimized plan to the large tumor geometry as well as to the small tumor geometry, referred as delivery I and delivery II, respectively. After replanning, we deliver the new plan to the small tumor geometry to study the dosimetric gain from ART replanning. This replan delivery will be referred as delivery
T III. Plan calculations and diode measurements in percentage of the prescription dose before and after deformation, with the original plan and the replan irradiated. Numbers in parenthesis are percentage difference. Dose
Diode 1 D c |D m
Diode 2 D c |D m
Diode 3 D c |D m
Diode 4 D c |D m
Delivery I Delivery II Delivery III
72.6|74.9 (2.3) 71.6|73.8 (2.2) 57.7|59.7 (2.0)
42.9|46.0 (3.1) 44.0|47.1 (3.1) 36.6|37.6 (1.0)
103.0|105.1 (2.1) 106.8|108.8 (2.0) 102.1|103.8 (1.7)
35.5|35.6 (1.0) 49.5|49.5 (0.0) 29.9|32.0 (2.1)
Medical Physics, Vol. 42, No. 4, April 2015
1496
Graves et al.: Deformable phantom with dosimetry for ART QA
F. 9. DVH plots of initial plan on new small tumor geometry (solid lines) and replan on new small tumor geometry (dashed lines).
III. Table III summarizes the dosimetric results from the three radiation deliveries. Dm and Dc represent diode measurements and Eclipse plan calculations, respectively. The values are converted into a unit of percentage of the prescription. A good agreement is obtained between the diode measurements and the plan calculations, with a maximum difference of 3.1%. The point dose to left parotid (diode 4) dramatically increases when the original plan is delivered to the small tumor geometry due to the parotid shifting into the high dose area after the tumor shrinkage. This dose is greatly decreased when the replan is applied, indicating the necessity of ART in this situation. This is consistent with the observations on DVHs shown in Fig. 9. When the initial plan is applied to the new geometry, increase of doses to organs is observed. With the reoptimized plan for the new geometry, the organ doses are brought back to low levels, particularly for the left parotid. This observation also agrees with the diode measurements and plan calculations for the lower point dose to organs (diode 1, diode 2, and diode 4) in delivery III compared to delivery II.
4. DISCUSSION AND CONCLUSIONS We have designed and constructed a deformable and dosimetric HN ART phantom. This phantom consists of six anatomic components and resembles a real HN patient geometry with similar HU numbers. Deflation of a balloon catheter inside the phantom tissue simulates tumor shrinkage. Nonradiopaque surface markers, which do not influence DIR algorithms, are added on the most superior and most inferior surface of the phantom, and provide the deformation ground truth. Fixed and movable holders are built in the phantom to hold a diode for dosimetric measurements. In vivo dosimetry, providing point dose measurements when the treatment plans are delivered to the original or deformed geometry phantom, can be used to verify the accuracy of ART technology. In this study, we have demonstrated the functionality of the proposed phantom for testing the accuracy of DIR algorithms. A Demons algorithm is used as an example, and a quantitative error analysis is performed between the measured deformation from the physical surface markers and the calculated deformation fields. Most of the surface markers lie within the homogeneous tissue region. These regions Medical Physics, Vol. 42, No. 4, April 2015
1496
constitute a challenge to the selected Demons algorithm, because this algorithm is an intensity based DIR algorithm, where the registration is more accurate in the high contrast region. The main focus of this study is not to investigate the Demons algorithm but to utilize the surface markers for the accuracy test. We also demonstrate the feasibility of this deformable phantom as an end-to-end ART QA phantom, going through all the ART treatment steps. In the test shown in the paper, the dose to parotid significantly increases when the initial plan is delivered to the small tumor geometry phantom, because the parotid shifts into the high dose area due to tumor shrinkage. In contrast, the dose to parotid dramatically decreases when the replan is applied. This indicates the necessity and benefit of ART in this situation. The point dose measurements from the in vivo diode dosimeters show a good match with the calculated doses from the treatment planning system with a maximum difference of 3.1% of prescription dose. The biggest point dose difference appears at the high dose gradient region, which is always a challenge for any dose measurements. A limitation of this work is the assumption of the 2D nature of the phantom. By being a 3D physical phantom, 3D motion is probably occurring, with different slices moving slightly differently. However, an attempt to minimize this effect has been made by different design strategies. First, the thickness of the phantom has been limited so that no big 3D distortions are expected. Also, an even layer of lubricant has been created between the phantom walls and the acrylic plates to avoid phantom-plate friction that could lead to 3D motion. Last, surface markers have been placed on both phantom wall surfaces and the average of deformations on both sides is taken as the ground truth. Measuring from the sagittal slices of the phantom CTs and taking into account the average of both phantom surfaces, the maximum difference in motion between a point on the surface of phantom and that in the middle CT slice is about 2.5 mm. This displacement error is localized in the posterior side of the balloon catheter, which is not close to the diode locations. Hence, this error has minimum effect on the ART dosimetry experiments. For the DIR evaluation test, the displacement errors from a few neighboring markers near the posterior side of the balloon catheter have minimum effect on quantifying the overall DIR algorithm accuracy using a total of 94 markers. In order to evaluate the DIR algorithm locally, the area of displacement with errors larger than 1 mm will be subtracted from the validation tests. To minimize the displacement difference between the phantom surfaces and the middle of phantom will be the main topic in future investigation. With the use of OBI system, daily CBCT images become available for ART. Several studies demonstrate the use of CBCT for dose reconstruction and ART replanning.30,31 This phantom has been verified to be tissue equivalent for MV imaging modalities, so it is not limited by image modality and can truly be an end-to-end QA phantom for any new ART technologies before implementing in clinics. This ART QA phantom is modeled based on a HN CT slice; therefore, it is designed to be used in HN ART clinical workflow, and mainly to evaluate DIR algorithms that are suitable for HN treatment
1497
Graves et al.: Deformable phantom with dosimetry for ART QA
sites. ART QA phantoms for other tumor sites will come out as future generations. They can be used in ART clinical workflow for other tumor sites, and also to test the performance of DIR algorithm over a wide range of tumor sites. ACKNOWLEDGMENT The authors would like to thank Joseph Graves for his contribution to material selection and phantom fabrication. a)Electronic
addresses: [email protected] and steve.jiang@utsouthwestern. edu 1J. A. Antolak, I. I. Rosen, C. H. Childress, G. K. Zagars, and A. Pollack, “Prostate target volume variations during a course of radiotherapy,” Int. J. Radiat. Oncol., Biol., Phys. 42, 661–672 (1998). 2L. van de Bunt, U. A. van der Heide, M. Ketelaars, G. A. P. de Kort, and I. M. Jurgenliemk-Schulz, “Conventional, conformal, and intensity-modulated radiation therapy treatment planning of external beam radiotherapy for cervical cancer: The impact of tumor regression,” Int. J.Radiat. Oncol., Biol., Phys. 64, 189–196 (2006). 3D. Yan, F. Vicini, J. Wong, and A. Martinez, “Adaptive radiation therapy,” Phys. Med. Biol. 42, 123–132 (1997). 4C. Wu, R. Jeraj, G. H. Olivera, and T. R. Mackie, “Re-optimization in adaptive radiotherapy,” Phys. Med. Biol. 47, 3181–3195 (2002). 5R. Mohan, X. D. Zhang, H. Wang, Y. X. Kang, X. C. Wang, H. Liu, K. Ang, D. Kuban, and L. Dong, “Use of deformed intensity distributions for on-line modification of image-guided IMRT to account for interfractional anatomic changes,” Int. J. Radiat. Oncol., Biol., Phys. 61, 1258–1266 (2005). 6A. de la Zerda, B. Armbruster, and L. Xing, “Formulating adaptive radiation therapy (ART) treatment planning into a closed-loop control framework,” Phys. Med. Biol. 52, 4137–4153 (2007). 7Q. J. Wu, D. Thongphiew, Z. Wang, B. Mathayomchan, V. Chankong, S. Yoo, W. R. Lee, and F.-F. Yin, “On-line re-optimization of prostate IMRT plans for adaptive radiation therapy,” Phys. Med. Biol. 53, 673–691 (2008). 8W. Lu, M. Chen, Q. Chen, K. Ruchala, and G. Olivera, “Adaptive fractionation therapy: I. Basic concept and strategy,” Phys. Med. Biol. 53, 5495–5511 (2008). 9C. Men, X. Gu, D. Choi, A. Majumdar, Z. Zheng, K. Mueller, and S. B. Jiang, “GPU-based ultrafast IMRT plan optimization,” Phys. Med. Biol. 54, 6565–6573 (2009). 10C. Men, X. Jia, and S. B. Jiang, “GPU-based ultra-fast direct aperture optimization for online adaptive radiation therapy,” Phys. Med. Biol. 55, 4309–4319 (2010). 11W. G. Lu, G. H. Olivera, Q. Chen, K. J. Ruchala, J. Haimerl, S. L. Meeks, K. M. Langen, and P. A. Kupelian, “Deformable registration of the planning image (kVCT) and the daily images (MVCT) for adaptive radiation therapy,” Phys. Med. Biol. 51, 4357–4374 (2006). 12T. Z. Zhang, Y. W. Chi, E. Meldolesi, and D. Yan, “Automatic delineation of on-line head-and-neck computed tomography images: Toward on-line adaptive radiotherapy,” Int. J. Radiat. Oncol., Biol., Phys. 68, 522–530 (2007). 13N. Hardcastle, W. A. Tome, D. M. Cannon, C. L. Brouwer, P. W. H. Wittendorp, N. Dogan, M. Guckenberger, S. Allaire, Y. Mallya, P. Kumar, M. Oechsner, A. Richter, S. Song, M. Myers, B. Polat, and K. Bzdusek, “A
Medical Physics, Vol. 42, No. 4, April 2015
1497
multi-institution evaluation of deformable image registration algorithms for automatic organ delineation in adaptive head and neck radiotherapy,” Radiat. Oncol. 7, 90 (7pp.) (2012). 14J. Thirion, “Image matching as a diffusion process: An analogy withMaxwell’s demons,” Med. Image Anal. 2, 243–260 (1998). 15B. Zitova and J. Flusser, “Image registration methods: A survey,” Image Vis. Comput. 21, 977–1000 (2003). 16W. G. Lu, M. L. Chen, G. H. Olivera, K. J. Ruchala, and T. R. Mackie, “Fast free-form deformable registration via calculus of variations,” Phys. Med. Biol. 49, 3067–3087 (2004). 17K. K. Brock, M. B. Sharpe, L. A. Dawson, S. M. Kim, and D. A. Jaffray, “Accuracy of finite element model-based multi-organ deformable image registration,” Med. Phys. 32, 1647–1659 (2005). 18D. Yang, H. Li, D. A. Low, J. O. Deasy, and I. El Naqa, “A fast inverse consistent deformable image registration method based on symmetric optical flow computation,” Phys. Med. Biol. 53, 6143–6165 (2008). 19M. Holden, “A review of geometric transformations for nonrigid body registration,” IEEE Trans. Med. Imaging 27, 111–128 (2008). 20X. Gu, H. Pan, Y. Liang, R. Castillo, D. Yang, D. Choi, E. Castillo, A. Majumdar, T. Guerrero, and S. B. Jiang, “Implementation and evaluation of various demons deformable image registration algorithms on a GPU,” Phys. Med. Biol. 55, 207–219 (2010). 21X. Zhen, H. Yan, X. Gu, L. Zhou, X. Jia, and S. Jiang, “CT to cone-beam CT deformable registration with simultaneous intensity correction,” Phys. Med. Biol. 52, 6807–6826 (2012). 22M. R. Kaus, K. K. Brock, V. Pekar, L. A. Dawson, A. M. Nichol, and D. A. Jaffray, “Assessment of a model-based deformable image registration approach for radiation therapy planning,” Int. J. Radiat. Oncol., Biol.,Phys. 68, 572–580 (2007). 23P. Castadot, J. A. Lee, A. Parraga, X. Geets, B. Macq, and V. Gregoire, “Comparison of 12 deformable registration strategies in adaptive radiation therapy for the treatment of head and neck tumors,” Radiother. Oncol. 89, 1–12 (2008). 24M. Foskey, B. Davis, L. Goyal, S. Chang, E. Chaney, N. Strehl, S. Tomei, J. Rosenman, and S. Joshi, “Large deformation three-dimensional image registration in image-guided radiation therapy,” Phys. Med. Biol. 50, 5869–5892 (2005). 25R. Kashani, M. Hub, M. L. Kessler, and J. M. Balter, “Technical note: A physical phantom for assessment of accuracy of deformable alignment algorithms,” Med. Phys. 34, 2785–2788 (2007). 26M. Serban, E. Heath, G. Stroian, D. L. Collins, and J. Seuntjens, “A deformable phantom for 4D radiotherapy verification: Design and image registration evaluation,” Med. Phys. 35, 1094–1102 (2008). 27N. Kirby, C. Chuang, and J. Pouliot, “A two-dimensional deformable phantom for quantitatively verifying deformation algorithms,” Med. Phys. 38, 4583–4586 (2011). 28N. Kirby, C. Chuang, U. Ueda, and J. Pouliot, “The need for applicationbased adaptation of deformable image registration,” Med. Phys. 40, 011702 (10pp.) (2013). 29D. Yan, “Developing quality assurance processes for image-guided adaptive radiation therapy,” Int. J. Radiat. Oncol., Biol., Phys. 71, S28–S32 (2008). 30Y. Yong, E. Schreibmann, L. Tianfang, W. Chuang, and X. Lei, “Evaluation of on-board kV cone beam CT (CBCT)-based dose calculation,” Phys. Med. Biol. 52, 685–705 (2007). 31W. Fu, Y. Yang, N. J. Yue, D. E. Heron, and M. S. Huq, “A cone beam CT-guided online plan modification technique to correct interfractional anatomic changes for prostate cancer IMRT treatment,” Phys. Med. Biol. 54, 1691–1703 (2009).
A deformable head and neck phantom with in-vivo dosimetry for adaptive radiotherapy quality assurance Yan Jiang Graves Center for Advanced Radiotherapy Technologies and Department of Radiation Medicine and Applied Sciences, University of California San Diego, La Jolla, California 92037-0843 and Department of Physics, University of California San Diego, La Jolla, California 92093
Arthur-Allen Smith, David Mcilvena, Zherrina Manilay, and Yuet Kong Lai Department of Mechanical and Aerospace Engineering, University of California San Diego, La Jolla, California 92093
Roger Rice and Loren Mell Center for Advanced Radiotherapy Technologies and Department of Radiation Medicine and Applied Sciences, University of California San Diego, La Jolla, California 92037-0843
Xun Jia and Steve B. Jianga) Center for Advanced Radiotherapy Technologies and Department of Radiation Medicine and Applied Sciences, University of California San Diego, La Jolla, California 92037-0843 and Department of Radiation Oncology, University of Texas Southwestern Medical Center, Dallas, Texas 75235
Laura Cerviñoa) Center for Advanced Radiotherapy Technologies and Department of Radiation Medicine and Applied Sciences, University of California San Diego, La Jolla, California 92037-0843
(Received 22 January 2014; revised 2 September 2014; accepted for publication 27 January 2015; published 11 March 2015) Purpose: Patients’ interfractional anatomic changes can compromise the initial treatment plan quality. To overcome this issue, adaptive radiotherapy (ART) has been introduced. Deformable image registration (DIR) is an important tool for ART and several deformable phantoms have been built to evaluate the algorithms’ accuracy. However, there is a lack of deformable phantoms that can also provide dosimetric information to verify the accuracy of the whole ART process. The goal of this work is to design and construct a deformable head and neck (HN) ART quality assurance (QA) phantom with in vivo dosimetry. Methods: An axial slice of a HN patient is taken as a model for the phantom construction. Six anatomic materials are considered, with HU numbers similar to a real patient. A filled balloon inside the phantom tissue is inserted to simulate tumor. Deflation of the balloon simulates tumor shrinkage. Nonradiopaque surface markers, which do not influence DIR algorithms, provide the deformation ground truth. Fixed and movable holders are built in the phantom to hold a diode for dosimetric measurements. Results: The measured deformations at the surface marker positions can be compared with deformations calculated by a DIR algorithm to evaluate its accuracy. In this study, the authors selected a Demons algorithm as a DIR algorithm example for demonstration purposes. The average error magnitude is 2.1 mm. The point dose measurements from the in vivo diode dosimeters show a good agreement with the calculated doses from the treatment planning system with a maximum difference of 3.1% of prescription dose, when the treatment plans are delivered to the phantom with original or deformed geometry. Conclusions: In this study, the authors have presented the functionality of this deformable HN phantom for testing the accuracy of DIR algorithms and verifying the ART dosimetric accuracy. The authors’ experiments demonstrate the feasibility of this phantom serving as an end-to-end ART QA phantom. C 2015 American Association of Physicists in Medicine. [http://dx.doi.org/10.1118/1.4908205] Key words: deformable registration, adaptive radiotherapy, deformable phantom, deformable registration verification, ART verification 1. INTRODUCTION The goal of radiation therapy is to irradiate the tumor target with a prescribed dose while sparing the nearby normal tissues. For this purpose, a treatment plan is generated based 1490
Med. Phys. 42 (4), April 2015
on the patient’s anatomy captured prior to the treatment. The patient’s anatomy may, however, change during the whole treatment course. This anatomical variation can compromise the initial treatment plan quality and therefore treatment outcomes.1,2 Adaptive radiotherapy (ART) has been proposed
0094-2405/2015/42(4)/1490/8/$30.00
© 2015 Am. Assoc. Phys. Med.
1490
1491
Graves et al.: Deformable phantom with dosimetry for ART QA
to overcome this challenge by modifying or reoptimizing the treatment plan based on the patient’s new anatomy in order to maintain the plan quality.3–10 Deformable image registration (DIR) algorithm is a crucial tool for ART. It establishes the correspondence between voxels in a reference image (e.g., the treatment planning image) and those in a target image (e.g., daily acquired image). DIR has been, for instance, wildly used to propagate the contours of the tumor target and the organs at risk (OARs) from the initial planning computed tomography (CT) to the daily patient images.11–13 DIR has also been used for dose accumulation, where the delivered dose is reconstructed on the daily patient image and transferred back to the reference geometry image according to voxel correspondences derived from a DIR algorithm.11 The delivered dose accumulation helps the physician to evaluate the treatment and decide when and how a replan or plan modification is needed. In light of these applications, the accuracy of DIR algorithm has a significant impact for a safe and accurate implementation of the ART technology in clinics. Over the years, a lot of efforts have been devoted to developing accurate DIR algorithms,14–21 as well as to finding appropriate ways to evaluate their accuracy. Common evaluation methods include comparing the calculated deformation vector fields (DVFs) with that derived based on landmark points,16,17,20,22 calculating image similarity metrics,16,21,23 and inspecting contours.24 Phantom-based deformable registration validation has also received much attention. Physical deformation displacements can be directly derived and compared with the calculated ones. Several deformable phantoms have been built for this purpose. Kashani et al.25 and Serban et al.26 designed two different lung phantoms that can mimic diaphragm motion and deform the modeled lung tissue. They performed quantitative evaluation of DIR algorithms based on the embedded landmarks in the phantoms. The landmarks are usually visible in the images and hence lead to good registration locally, but registration errors elsewhere, especially in homogeneous regions, are hard to identify and evaluate. A research group at University
1491
of California, San Francisco (UCSF) designed two anthropomorphic deformable phantoms, i.e., a head and neck (HN) phantom and a pelvic phantom.27,28 Both phantoms resemble the corresponding tumor sites closely. A grid of nonradiopaque surface makers is placed in the phantoms. These markers can be captured by optical cameras but are invisible in the CT images for DIR algorithms. The actual deformation can be extracted from the optical camera images and used to verify the DIR algorithm’s output. As pointed out by Yan,29 a clinical quality assurance (QA) workflow for ART demands (1) DIR algorithm accuracy tests and (2) delivered dose verification. So far there is a lack of deformable phantoms that also provide dosimetric information for the complete ART QA purpose. The goal of this study is to design and construct an ART QA phantom that could perform both of these. We selected HN as the anatomical site, because some HN tumors respond dramatically to the radiotherapy treatment and shrink through the treatment course, greatly benefiting from ART. This ART phantom can be implemented in clinical HN ART workflow to QA the whole ART procedure and technologies. The first part of the paper focuses on the design and fabrication of the phantom, while the second part demonstrates the feasibility of using this phantom to verify the DIR algorithm’s accuracy and to serve as an end-to-end ART dosimetry QA phantom.
2. METHODS AND MATERIALS 2.A. Phantom design and fabrication
The geometry of this phantom is determined based on a single axial CT slice of a typical HN patient who is likely to require ART during the treatment [Fig. 1(A)]. The key organs found in this slice are mandible, cord, tumor, parotids, pharynx, skin, and soft tissue, including dense soft tissue (i.e., muscle) and light soft tissue (i.e., fat). The PTV and the parotids contours are displayed on the patient CT slice. The thickness of the phantom along the superior–inferior
F. 1. (A) CT image of modeled HN patient with PTV and parotids contours. (B) Phantom construction diagram with marked diode holder locations (1–4). The dimensions are in cm. Medical Physics, Vol. 42, No. 4, April 2015
1492
Graves et al.: Deformable phantom with dosimetry for ART QA
(SI) direction is 4.5 cm, and its axial geometry remains unchanged throughout that thickness. Therefore, this phantom is considered as a 2D phantom. The phantom consists of two parallel acrylic plates with the deformable tissue sandwiched between them. To make the phantom realistic, the deformation material is selected such that they have similar x-ray attenuation properties to that in the human body. Specifically, gypsum plaster is used for casting the bony anatomy, i.e., mandible and spine, and they are rigidly attached to the acrylic plates with small nylon bolts. Soft tissue is made of silicone rubber gel and microballoons filler. The base silicon rubber gel represents the dense soft tissue, and the microballoons filler is added into silicon rubber gel to create lighter weight casting. Parotids and light soft tissue are mixtures of base silicone gel with 10% and 20% of microballoons filler, respectively. The pharynx is represented by a nylon tube, which creates an air cavity. Urethane rubber is painted on the soft tissue surface to simulate skin. A latex balloon is inserted in the deformable tissue and glued to a catheter that goes outside the acrylic plates, shown in Fig. 4. A syringe is attached to the balloon catheter so that the balloon volume can be adjusted to simulate tumor shrinkage by modifying the amount of salt water injected. The concentration of the salt water inside the balloon is adjusted, so that it has a similar HU value to the surrounding tissue. On the two sides of deformable material surfaces adjacent to the acrylic plates, small glow-in-dark nonradiopaque marker dots are glued to form a 2D grid with approximately 1.5 cm resolution. The markers are visible to the optical camera through the clear acrylic plates, but they do not appear on CT images. Middle size nylon bolts are installed at the four corners of the acrylic plates to hold the deformable tissue in place. Four glow-in-the-dark markers are glued to the bolt tops as calibration markers to assist CT to optical images alignment and surface marker positions derivation. In order to minimize the constraining effect from the acrylic plates, an even layer of water-based lubricant is applied to the acrylic plates in order to create frictionless contact between the plates and the deformable material surfaces. To acquire the in vivo dosimetry, we selected SFD stereotactic dosimetry diode (IBA Dosimetry, Barlett, TN) as our detector due to its small volume size. The detector holders are made of acrylic and designed for the SFD diode dimension specifically. There are four diode detector holders installed in the deformable phantom, as shown in Fig. 1(B). Two of the detectors are fixed near the mandible (diode 1) and inside the cord (diode 2). The other two, which are installed on rails and can move with tissue deformation, are close to the tumor (diode 3) and inside the left parotid (diode 4). For the installation of the rails, we first estimated the tissue deformation direction from the tumor shrinkage at the locations of diode 3 and diode 4 and then installed the rail tracks along the deformation direction. The diode holders are designed to freely move on the rail track and the design diagram is shown in Fig. 2. The rail tracks and moving components are made of plastic. The diode holder is constructed such that, when the diode is placed, its tip aligns with the middle plane of the deformable phantom. Medical Physics, Vol. 42, No. 4, April 2015
1492
F. 2. Design diagram of the diode holders on rail.
2.B. Phantom deformation and data acquisition
A syringe is attached to the balloon catheter outside the acrylic plate. To represent the original large tumor volume, 160 ml salt water is initially injected into the latex balloon catheter. After charging the phosphorescent surface markers, optical camera images of a pixel size 0.33 mm are acquired in the dark for the markers on both sides of the phantom. To simulate tumor shrinkage, we deflate the balloon volume to 65 ml by withdrawing the salt water out of the balloon with the syringe. Both diode 3 and diode 4 holders move closer to the tumor along the rails with the tissue deformation. Again optical images of both sides of the phantom are acquired at the same distance as for the original geometry phantom. Since the phantom mimics tissue deformations from tumor shrinkage, the terms before and after correspond to an inflated and deflated latex balloon, respectively. Two sets of CT images are obtained for the phantom before and after the deformation using a GE CT scanner (GE Four-Slice Qxi LightSpeed CT, GE Healthcare). The CT resolution is 0.98 mm and the slice thickness is 2.5 mm. 2.C. DIR accuracy experiment
The grids of small nonradiopaque markers on both deformable phantom surfaces provide the ground truth to verify the accuracy of the DVFs calculated from DIR algorithms. Optical images of both sides of the phantom are taken before and after tissue deformation. An in-house program is developed to detect the physical marker dots in the optical images by applying a threshold value in the green channel of the RGB optical images. The origin of the position coordinate system aligns with the most anterior-right calibration marker. y and x denote the anterior and patient-left directions, respectively. The calibration markers on the nylon bolts are known to form a square with 29.7 cm sides. This information is then utilized to calculate the x and y coordinates of the initial and the final centroid positions of the surface marker dots from the optical images. Finally, the deformation motion vectors are averaged from both sides of phantom surface markers and considered as the measured deformation ground truth, which is represented as − ν⇀m . To test the accuracy of a DIR algorithm, the measured deformation is compared to the algorithm’s output. Under
1493
Graves et al.: Deformable phantom with dosimetry for ART QA
1493
the assumption that the phantom is 2D, i.e., there is no deformation along the SI direction, the middle CT slice is used in the DIR algorithms. The CT slice with the large tumor before simulated tumor shrinkage is considered as the reference image, and the CT slice with the small tumor is the moving image. The calculated DVFs transfer the moving image to the reference image. DVFs at 94 marker locations are extracted and denoted as − ν−⇀ cal. In this study, we selected a GPUbased Demons algorithm20 as our DIR algorithm example and reported the average error between the measured deformation and the calculated one as N −−⇀ −−−⇀ ∥ νi, m − νi,cal ∥ , (1) δ = l=1 N where N = 94. Note that the main focus of this study is to demonstrate the feasibility of utilizing the surface markers for the DIR algorithm accuracy test, not to investigate the DIR algorithm accuracy. Users can also choose any comparison metrics that they believe are best for the DIR evaluation. 2.D. Diode angular dependence properties and dosimetry calibration
F. 4. Prototype of the deformable HN phantom.
A conventional setup for the diode use is to align its long axis parallel to the beam axis. However, in our application, the diode is inserted into the phantom such that its long axis is perpendicular to the beam axis. Hence, before adopting this
setup, the angular dependence of this SFD diode should be evaluated. The angular dependence is defined as the response of the diode from different beam angles around the diode’s long axis under the isocentric condition, normalized to the response from the reference condition. The reference condition refers to the isocentric setup with gantry angle at zero degrees. In our experiment, an acrylic buildup cap is designed, shown in Fig. 3(A) and built to increase dose depositions to the diode for this angular dependence investigation. The reference response is obtained when 100 MU radiation is delivered with a 5×5 cm2 field size. Different diode responses with 100 MU radiation from seven coplanar equiangular angles (i.e., at 0◦, 51◦, 102◦, 153◦, 207◦, 258◦, and 309◦) are measured and are normalized to the reference response. The beam angles with respect to the orientation of the diode are illustrated in Fig. 3(B). The diode dosimeter reading needs to be calibrated before conducting dosimetric validations using the proposed phantom. Specifically, the diode calibration is performed at
F. 3. (A) Diagram of the buildup cap. Measurements are in mm. (B) Orientation of beam angles with respect to the diode.
F. 5. (A) CT slice at the middle of the phantom before tumor shrinkage. (B) CT slice at the middle of the phantom after tumor shrinkage.
Medical Physics, Vol. 42, No. 4, April 2015
1494
Graves et al.: Deformable phantom with dosimetry for ART QA
1494
T I. Average HUs for phantom and patients’ components.
Bone
Dense soft tissue
Light soft tissue
Parotids
865|920
140|100
−35| − 50
90|65
Components Phantom|patient
the isocentric position for a 10 × 10 cm2 field size at 10 cm depth in a solid water phantom with 100 MU 6 MV beam radiation delivered. The calibration factor f cal, which converts the electrometer reading to dose in water, is obtained from D0 , (2) R0 where R0 and D0 are the electrometer reading and dose to water under the calibration condition. Here, D0 is derived based on TMR ratio at depth of 10 cm for the same condition. It is also found that the diode sensitivity slightly depends on the radiation history. Therefore, the diode is frequently calibrated every time right before a dosimetric verification experiment. Once the calibration coefficient f cal is obtained, in a dose measurement, we can obtain the absolute dose value Dm as f cal =
Dm = f cal × Rm ,
(3)
where Rm is the electrometer reading of the diode response in the measurement.
F. 6. (A) Marker positions for large tumor geometry phantom before tumor shrinkage. (B) Marker positions for small tumor geometry phantom after tumor shrinkage. (C) The measured DVFs overlaid on CT slice of the large tumor phantom. (D) The DIR generated DVFs overlaid on CT slice of the large tumor phantom.
2.E. ART dosimetry experiment
To serve as an end-to-end ART QA phantom, this deformable phantom is treated exactly in the same way as for a real patient, going through an initial CT scan, treatment planning, plan delivery, geometric deformation, re-CT scan, replanning, and replan delivery procedure. First, after the CT image is acquired for the phantom with a large tumor, the PTV and organ contours are drawn. We deliberately extend the tumor contour to include diode 3 inside the PTV, so that it provides the dose measurement inside the PTV target. Then, an IMRT plan with a set of seven coplanar equiangular 6 MV beams is generated on the large tumor geometry using Eclipse (Varian Medical, Inc., Palo Alto, CA) treatment planning system (TPS). Clinically relevant constraints to organs are satisfied in the plan. After the phantom is properly set up on
the treatment couch using onboard imaging (OBI) system for image guidance, the plan is delivered to the phantom four times, each time with the diode placed in a different diode holder and plastic inserts in the rest of the diode holders. The four diode readings are obtained and compared to the calculated results from the TPS. After tumor shrinkage, the PTV and organ contours are redrawn in the new CT image. With the same beam angles and plan constraints, the new geometry is used for replanning in Eclipse. Then, both the reoptimized plan and the initial plan are delivered to the new phantom geometry in order to evaluate the dosimetric gain from ART replanning. The diode readings are compared with the calculation from Eclipse to verify ART accuracy.
F. 7. (A) Difference image between the reference and the moving image. (B) Difference between the reference and the deformed image. Medical Physics, Vol. 42, No. 4, April 2015
1495
Graves et al.: Deformable phantom with dosimetry for ART QA
1495
T II. Mean and standard deviation of the average error for DIR output compared to the measured ground truth. A–P (mm) 1.0 ± 1.0
L–R (mm)
δ (mm)
1.5 ± 2.0
2.1 ± 2.2
3. RESULTS 3.A. Deformable HN phantom
The prototype of this deformable HN phantom is presented in Fig. 4. The deformable phantom consists of six materials that represent bone, shrinking tumor, air cavity, parotids, and soft tissue including light and denser tissue. Figures 5(A) and 5(B) show the CT slices at the middle of the deformable phantom before and after the tumor shrinkage, with each anatomical components easily observed. The PTV contour, as well as other organ contours, is displayed, and the diode detector locations are shown in Fig. 5. The average HU values from the modeled HN patient for bone, dense and light soft tissue, and parotids are summarized in Table I. For each phantom organ material, we are able to reproduce the similar HU values to that in the real patient and keep the reasonable relative HU difference between any two kinds.
F. 8. Graph of diode response at different angles against response at zero degree gantry angle.
the vector field is also summarized in Table II. The average error magnitude at the 94 surface markers is 2.1 mm, with a larger variance for left–right (L–R) direction than for the anterior–posterior (A–P) direction.
3.B. DIR accuracy evaluation
3.C. Diode angular dependence
Figures 6(A) and 6(B) display the detected glow-in-thedark surface marker positions before and after the simulated tumor shrinkage. The measured marker motion vector fields and the Demons generated DVFs are overlaid on the CT slice of the large tumor phantom in Figs. 6(C) and 6(D), respectively. Good visual agreement between the measured deformation and the DIR output is observed. The CT slice with the large tumor before the shrinkage is considered as the reference image, and the slice with the small tumor is the moving image. The moving image is registered to the reference image to produce the deformed image. The DIR performance is qualitatively evaluated by comparing the difference images between the reference and the moving images [Fig. 7(A)] with the difference image between the reference and the deformed ones [Fig. 7(B)]. It can be seen that the difference image before registration is relatively large in intensity. After the deformable registration, the intensity in the difference image is greatly reduced, indicating the accuracy of the registration. Moreover, the mean pixel difference drops from 2% to 0.038% of the maximum reference image intensity after deformable registration. The quantitative difference of
The response at different gantry angles is averaged from two rotations of radiation delivery and the result values are plotted in Fig. 8. The greatest asymmetry appears at 207◦ angle, at which the response is 1.36% lower than the reference value. Derivations from the reference response at other angles are within 0.5%. Because of the low angular dependence of the diode response, we will ignore this issue when using it for in vivo measurement inside of the phantom with its long axes being set perpendicular to the beam axis. 3.D. ART dosimetry evaluation
Before conducting dosimetric validations using this proposed phantom, the diode dosimeter reading is calibrated. The error expected for this diode is 0.97%. To evaluate the dosimetric effect from the tumor shrinkage, we deliver the initial optimized plan to the large tumor geometry as well as to the small tumor geometry, referred as delivery I and delivery II, respectively. After replanning, we deliver the new plan to the small tumor geometry to study the dosimetric gain from ART replanning. This replan delivery will be referred as delivery
T III. Plan calculations and diode measurements in percentage of the prescription dose before and after deformation, with the original plan and the replan irradiated. Numbers in parenthesis are percentage difference. Dose
Diode 1 D c |D m
Diode 2 D c |D m
Diode 3 D c |D m
Diode 4 D c |D m
Delivery I Delivery II Delivery III
72.6|74.9 (2.3) 71.6|73.8 (2.2) 57.7|59.7 (2.0)
42.9|46.0 (3.1) 44.0|47.1 (3.1) 36.6|37.6 (1.0)
103.0|105.1 (2.1) 106.8|108.8 (2.0) 102.1|103.8 (1.7)
35.5|35.6 (1.0) 49.5|49.5 (0.0) 29.9|32.0 (2.1)
Medical Physics, Vol. 42, No. 4, April 2015
1496
Graves et al.: Deformable phantom with dosimetry for ART QA
F. 9. DVH plots of initial plan on new small tumor geometry (solid lines) and replan on new small tumor geometry (dashed lines).
III. Table III summarizes the dosimetric results from the three radiation deliveries. Dm and Dc represent diode measurements and Eclipse plan calculations, respectively. The values are converted into a unit of percentage of the prescription. A good agreement is obtained between the diode measurements and the plan calculations, with a maximum difference of 3.1%. The point dose to left parotid (diode 4) dramatically increases when the original plan is delivered to the small tumor geometry due to the parotid shifting into the high dose area after the tumor shrinkage. This dose is greatly decreased when the replan is applied, indicating the necessity of ART in this situation. This is consistent with the observations on DVHs shown in Fig. 9. When the initial plan is applied to the new geometry, increase of doses to organs is observed. With the reoptimized plan for the new geometry, the organ doses are brought back to low levels, particularly for the left parotid. This observation also agrees with the diode measurements and plan calculations for the lower point dose to organs (diode 1, diode 2, and diode 4) in delivery III compared to delivery II.
4. DISCUSSION AND CONCLUSIONS We have designed and constructed a deformable and dosimetric HN ART phantom. This phantom consists of six anatomic components and resembles a real HN patient geometry with similar HU numbers. Deflation of a balloon catheter inside the phantom tissue simulates tumor shrinkage. Nonradiopaque surface markers, which do not influence DIR algorithms, are added on the most superior and most inferior surface of the phantom, and provide the deformation ground truth. Fixed and movable holders are built in the phantom to hold a diode for dosimetric measurements. In vivo dosimetry, providing point dose measurements when the treatment plans are delivered to the original or deformed geometry phantom, can be used to verify the accuracy of ART technology. In this study, we have demonstrated the functionality of the proposed phantom for testing the accuracy of DIR algorithms. A Demons algorithm is used as an example, and a quantitative error analysis is performed between the measured deformation from the physical surface markers and the calculated deformation fields. Most of the surface markers lie within the homogeneous tissue region. These regions Medical Physics, Vol. 42, No. 4, April 2015
1496
constitute a challenge to the selected Demons algorithm, because this algorithm is an intensity based DIR algorithm, where the registration is more accurate in the high contrast region. The main focus of this study is not to investigate the Demons algorithm but to utilize the surface markers for the accuracy test. We also demonstrate the feasibility of this deformable phantom as an end-to-end ART QA phantom, going through all the ART treatment steps. In the test shown in the paper, the dose to parotid significantly increases when the initial plan is delivered to the small tumor geometry phantom, because the parotid shifts into the high dose area due to tumor shrinkage. In contrast, the dose to parotid dramatically decreases when the replan is applied. This indicates the necessity and benefit of ART in this situation. The point dose measurements from the in vivo diode dosimeters show a good match with the calculated doses from the treatment planning system with a maximum difference of 3.1% of prescription dose. The biggest point dose difference appears at the high dose gradient region, which is always a challenge for any dose measurements. A limitation of this work is the assumption of the 2D nature of the phantom. By being a 3D physical phantom, 3D motion is probably occurring, with different slices moving slightly differently. However, an attempt to minimize this effect has been made by different design strategies. First, the thickness of the phantom has been limited so that no big 3D distortions are expected. Also, an even layer of lubricant has been created between the phantom walls and the acrylic plates to avoid phantom-plate friction that could lead to 3D motion. Last, surface markers have been placed on both phantom wall surfaces and the average of deformations on both sides is taken as the ground truth. Measuring from the sagittal slices of the phantom CTs and taking into account the average of both phantom surfaces, the maximum difference in motion between a point on the surface of phantom and that in the middle CT slice is about 2.5 mm. This displacement error is localized in the posterior side of the balloon catheter, which is not close to the diode locations. Hence, this error has minimum effect on the ART dosimetry experiments. For the DIR evaluation test, the displacement errors from a few neighboring markers near the posterior side of the balloon catheter have minimum effect on quantifying the overall DIR algorithm accuracy using a total of 94 markers. In order to evaluate the DIR algorithm locally, the area of displacement with errors larger than 1 mm will be subtracted from the validation tests. To minimize the displacement difference between the phantom surfaces and the middle of phantom will be the main topic in future investigation. With the use of OBI system, daily CBCT images become available for ART. Several studies demonstrate the use of CBCT for dose reconstruction and ART replanning.30,31 This phantom has been verified to be tissue equivalent for MV imaging modalities, so it is not limited by image modality and can truly be an end-to-end QA phantom for any new ART technologies before implementing in clinics. This ART QA phantom is modeled based on a HN CT slice; therefore, it is designed to be used in HN ART clinical workflow, and mainly to evaluate DIR algorithms that are suitable for HN treatment
1497
Graves et al.: Deformable phantom with dosimetry for ART QA
sites. ART QA phantoms for other tumor sites will come out as future generations. They can be used in ART clinical workflow for other tumor sites, and also to test the performance of DIR algorithm over a wide range of tumor sites. ACKNOWLEDGMENT The authors would like to thank Joseph Graves for his contribution to material selection and phantom fabrication. a)Electronic
addresses: [email protected] and steve.jiang@utsouthwestern. edu 1J. A. Antolak, I. I. Rosen, C. H. Childress, G. K. Zagars, and A. Pollack, “Prostate target volume variations during a course of radiotherapy,” Int. J. Radiat. Oncol., Biol., Phys. 42, 661–672 (1998). 2L. van de Bunt, U. A. van der Heide, M. Ketelaars, G. A. P. de Kort, and I. M. Jurgenliemk-Schulz, “Conventional, conformal, and intensity-modulated radiation therapy treatment planning of external beam radiotherapy for cervical cancer: The impact of tumor regression,” Int. J.Radiat. Oncol., Biol., Phys. 64, 189–196 (2006). 3D. Yan, F. Vicini, J. Wong, and A. Martinez, “Adaptive radiation therapy,” Phys. Med. Biol. 42, 123–132 (1997). 4C. Wu, R. Jeraj, G. H. Olivera, and T. R. Mackie, “Re-optimization in adaptive radiotherapy,” Phys. Med. Biol. 47, 3181–3195 (2002). 5R. Mohan, X. D. Zhang, H. Wang, Y. X. Kang, X. C. Wang, H. Liu, K. Ang, D. Kuban, and L. Dong, “Use of deformed intensity distributions for on-line modification of image-guided IMRT to account for interfractional anatomic changes,” Int. J. Radiat. Oncol., Biol., Phys. 61, 1258–1266 (2005). 6A. de la Zerda, B. Armbruster, and L. Xing, “Formulating adaptive radiation therapy (ART) treatment planning into a closed-loop control framework,” Phys. Med. Biol. 52, 4137–4153 (2007). 7Q. J. Wu, D. Thongphiew, Z. Wang, B. Mathayomchan, V. Chankong, S. Yoo, W. R. Lee, and F.-F. Yin, “On-line re-optimization of prostate IMRT plans for adaptive radiation therapy,” Phys. Med. Biol. 53, 673–691 (2008). 8W. Lu, M. Chen, Q. Chen, K. Ruchala, and G. Olivera, “Adaptive fractionation therapy: I. Basic concept and strategy,” Phys. Med. Biol. 53, 5495–5511 (2008). 9C. Men, X. Gu, D. Choi, A. Majumdar, Z. Zheng, K. Mueller, and S. B. Jiang, “GPU-based ultrafast IMRT plan optimization,” Phys. Med. Biol. 54, 6565–6573 (2009). 10C. Men, X. Jia, and S. B. Jiang, “GPU-based ultra-fast direct aperture optimization for online adaptive radiation therapy,” Phys. Med. Biol. 55, 4309–4319 (2010). 11W. G. Lu, G. H. Olivera, Q. Chen, K. J. Ruchala, J. Haimerl, S. L. Meeks, K. M. Langen, and P. A. Kupelian, “Deformable registration of the planning image (kVCT) and the daily images (MVCT) for adaptive radiation therapy,” Phys. Med. Biol. 51, 4357–4374 (2006). 12T. Z. Zhang, Y. W. Chi, E. Meldolesi, and D. Yan, “Automatic delineation of on-line head-and-neck computed tomography images: Toward on-line adaptive radiotherapy,” Int. J. Radiat. Oncol., Biol., Phys. 68, 522–530 (2007). 13N. Hardcastle, W. A. Tome, D. M. Cannon, C. L. Brouwer, P. W. H. Wittendorp, N. Dogan, M. Guckenberger, S. Allaire, Y. Mallya, P. Kumar, M. Oechsner, A. Richter, S. Song, M. Myers, B. Polat, and K. Bzdusek, “A
Medical Physics, Vol. 42, No. 4, April 2015
1497
multi-institution evaluation of deformable image registration algorithms for automatic organ delineation in adaptive head and neck radiotherapy,” Radiat. Oncol. 7, 90 (7pp.) (2012). 14J. Thirion, “Image matching as a diffusion process: An analogy withMaxwell’s demons,” Med. Image Anal. 2, 243–260 (1998). 15B. Zitova and J. Flusser, “Image registration methods: A survey,” Image Vis. Comput. 21, 977–1000 (2003). 16W. G. Lu, M. L. Chen, G. H. Olivera, K. J. Ruchala, and T. R. Mackie, “Fast free-form deformable registration via calculus of variations,” Phys. Med. Biol. 49, 3067–3087 (2004). 17K. K. Brock, M. B. Sharpe, L. A. Dawson, S. M. Kim, and D. A. Jaffray, “Accuracy of finite element model-based multi-organ deformable image registration,” Med. Phys. 32, 1647–1659 (2005). 18D. Yang, H. Li, D. A. Low, J. O. Deasy, and I. El Naqa, “A fast inverse consistent deformable image registration method based on symmetric optical flow computation,” Phys. Med. Biol. 53, 6143–6165 (2008). 19M. Holden, “A review of geometric transformations for nonrigid body registration,” IEEE Trans. Med. Imaging 27, 111–128 (2008). 20X. Gu, H. Pan, Y. Liang, R. Castillo, D. Yang, D. Choi, E. Castillo, A. Majumdar, T. Guerrero, and S. B. Jiang, “Implementation and evaluation of various demons deformable image registration algorithms on a GPU,” Phys. Med. Biol. 55, 207–219 (2010). 21X. Zhen, H. Yan, X. Gu, L. Zhou, X. Jia, and S. Jiang, “CT to cone-beam CT deformable registration with simultaneous intensity correction,” Phys. Med. Biol. 52, 6807–6826 (2012). 22M. R. Kaus, K. K. Brock, V. Pekar, L. A. Dawson, A. M. Nichol, and D. A. Jaffray, “Assessment of a model-based deformable image registration approach for radiation therapy planning,” Int. J. Radiat. Oncol., Biol.,Phys. 68, 572–580 (2007). 23P. Castadot, J. A. Lee, A. Parraga, X. Geets, B. Macq, and V. Gregoire, “Comparison of 12 deformable registration strategies in adaptive radiation therapy for the treatment of head and neck tumors,” Radiother. Oncol. 89, 1–12 (2008). 24M. Foskey, B. Davis, L. Goyal, S. Chang, E. Chaney, N. Strehl, S. Tomei, J. Rosenman, and S. Joshi, “Large deformation three-dimensional image registration in image-guided radiation therapy,” Phys. Med. Biol. 50, 5869–5892 (2005). 25R. Kashani, M. Hub, M. L. Kessler, and J. M. Balter, “Technical note: A physical phantom for assessment of accuracy of deformable alignment algorithms,” Med. Phys. 34, 2785–2788 (2007). 26M. Serban, E. Heath, G. Stroian, D. L. Collins, and J. Seuntjens, “A deformable phantom for 4D radiotherapy verification: Design and image registration evaluation,” Med. Phys. 35, 1094–1102 (2008). 27N. Kirby, C. Chuang, and J. Pouliot, “A two-dimensional deformable phantom for quantitatively verifying deformation algorithms,” Med. Phys. 38, 4583–4586 (2011). 28N. Kirby, C. Chuang, U. Ueda, and J. Pouliot, “The need for applicationbased adaptation of deformable image registration,” Med. Phys. 40, 011702 (10pp.) (2013). 29D. Yan, “Developing quality assurance processes for image-guided adaptive radiation therapy,” Int. J. Radiat. Oncol., Biol., Phys. 71, S28–S32 (2008). 30Y. Yong, E. Schreibmann, L. Tianfang, W. Chuang, and X. Lei, “Evaluation of on-board kV cone beam CT (CBCT)-based dose calculation,” Phys. Med. Biol. 52, 685–705 (2007). 31W. Fu, Y. Yang, N. J. Yue, D. E. Heron, and M. S. Huq, “A cone beam CT-guided online plan modification technique to correct interfractional anatomic changes for prostate cancer IMRT treatment,” Phys. Med. Biol. 54, 1691–1703 (2009).
