- Email: [email protected]
An analysis of anatomic landmark mobility and setup deviations in radiotherapy for lung cancer
Int. J. Radiation Oncology Biol. Phys., Vol. 43, No. 4, pp. 827– 832, 1999 Copyright © 1999 Elsevier Science Inc. Printed in the USA. All rights reserved 0360-3016/99/$–see front matter
PII S0360-3016(98)00467-2
CLINICAL INVESTIGATION
Lung
AN ANALYSIS OF ANATOMIC LANDMARK MOBILITY AND SETUP DEVIATIONS IN RADIOTHERAPY FOR LUNG CANCER MICHAEL J. SAMSON, M.D.,* JOHN R. VAN SO¨ RNSEN DE KOSTE, B.SC.,* HANS C. J. DE BOER, M.SC.,† HANS TANKINK, B.SC.,* MARJOLEIN VERSTRAATE, B.SC.,* MARION ESSERS, PH.D.,† ANDRIES G. VISSER, PH.D.,† AND SURESH SENAN, M.R.C.P., F.R.C.R., PH.D.* Departments of *Radiation Oncology and †Clinical Physics, University Hospital Rotterdam–Daniel den Hoed Cancer Center/ University Hospital Rotterdam, Rotterdam, The Netherlands Purpose: To identify thoracic structures that exhibit little internal motion during irradiation and to determine setup variations in patients with lung cancer. Methods and Materials: Intrafractional images were generated with an electronic portal-imaging device from the AP fields of 10 patients, during several fractions. To determine the intrafractional mobility of thoracic structures, visible structures were contoured in every image and matched with a reference image by means of a crosscorrelation algorithm. Setup variations were determined by comparing portal images with the digitized simulator films using the stable structures as landmarks. Results: Mobility was limited in the lateral direction for the trachea, thoracic wall, paraspinal line, and aortic notch, and in the craniocaudal direction for the clavicle, aortic notch, and thoracic wall. Analysis of patient setup revealed random deviations of 2.0 mm (1 SD) in the lateral direction and 2.8 mm in the craniocaudal direction, while the systematic deviations were 2.5 and 2.0 mm (1 SD) respectively. Conclusions: We have identified thoracic structures that exhibit little internal motion in the frontal plane, and recommend that these structures be used for verifying patient setup during radiotherapy. The daily variation in the setup of lung cancer patients at our center appears to be acceptable. © 1999 Elsevier Science Inc. Portal imaging, Anatomic landmark mobility, Treatment verification, Radiotherapy.
The importance of evaluating and correcting setup errors during radiotherapy is well established. However, visual comparison of portal films with simulator films to extract setup deviations is a tedious process. Manual measurements of field edges and blocks relative to the anatomic landmarks may be subject to many inaccuracies, including interobserver variability. Electronic portal imaging devices (EPIDs) provide digital data that allow the evaluation of patient positioning during irradiation. These images can be compared with digitized simulator films or digitally reconstructed radiographs by using image-matching algorithms. These computerized techniques make the task of evaluating patient setup less laborious, fast and accurate, enabling on-line evaluation and the possibility of correcting deviations in patient positioning. Several centers, including our own, have studied the distribution of setup deviations in treatment sites such as the pelvis (1, 2), breast (3–5), and head and neck (6, 7). There is, however, a relative lack of data on patients with intrathoracic tumors. Early studies reported large discrepancies
between simulation and portal films of patients with thoracic lesions and also indicated significant treatment-totreatment variations to exist during irradiation (8, 9). However, these deviations were not extensively investigated in subsequent studies (10, 11). Structures in the thoracic region may be mobile during the respiratory and cardiac cycles, leading to uncertainty when they are used for image comparison. While (short exposure) simulator films may provide sharp images of mobile structures, they represent a random position of these structures in the respiratory or cardiac cycle. Consequently, the use of such structures for matching portal images with short exposure simulator films might introduce systematic deviations. Similarly, mobile structures in portal images may enlarge the extracted random deviations due to their blurred appearance (as a consequence of the long exposure) or their random position (as a consequence of the short exposure). Therefore, a reliable analysis of setup deviations of patients with thoracic lesions requires the use of stable landmarks. However, the available literature does not provide this information and the studies that have evaluated setup variability in the thoracic region have not always
Presented at the 39th Annual meeting of the American Society for Therapeutic Radiology and Oncology, Orlando, Fla., October 1997. Reprint requests to: M. J. Samson, Department of Radiation
Oncology, St. Elisabeth Hospital, Breedestraat (O), Curac¸ao, The Netherlands Antilles. Accepted for publication 29 October 1998.
INTRODUCTION
827
828
I. J. Radiation Oncology
●
Biology
●
Physics
Volume 43, Number 4, 1999
indicated the landmarks that have been used for matching images. The aim of this study was to identify thoracic structures that exhibit little internal motion during irradiation. These structures can, subsequently, be used to determine daily setup deviations during a course of radiotherapy of patients with lung cancer. METHODS AND MATERIALS After simulation and treatment planning, 14 patients with stage III non-small cell lung cancer were irradiated on a linear accelerator equipped with an EPID (Siemens Beam ViewPLUS). The patients were set up in the supine position, and aligned using lasers. The arms were either positioned above the head (12 patients), by holding on to a cross bar, or along the sides of the body (3 patients), depending on the irradiation technique. No other devices were used for patient immobilization. Intrafractional motion was studied in the first 10 patients. Variations in patient setup were evaluated in 4 of the former group (i.e., those in whom at least five fractions were imaged) and in 4 additional patients. Intrafractional motion In the first 10 patients, the stability of different anatomic landmarks in the thorax was determined by analyzing the multiple images generated during administration of a single fraction of radiotherapy (12). Between 6 to 12 sequential images were generated with the EPID; once every 2 seconds (exposure 0.54 sec; processing and “dead” time 1.5 sec) for the first 6 images and, thereafter, at a rate depending on the remaining dose to be delivered. These intrafractional images were generated for the anteroposterior (AP) field of patients during 3–5 fractions of the total radiation scheme. The acquired images were imported into the visualization software package Advanced Visual Systems (AVS) (Advanced Visual Systems, Inc., Waltham, MA). Visible structures such as the trachea, carina, the upper chest wall, aortic arch, clavicle, and paraspinal line (Fig. 1) were manually contoured in each image. In order to determine the mobility of each structure separately during a particular fraction, each image was matched with the first image (the reference image) of the corresponding sequence, by means of a crosscorrelation algorithm. The method has been developed using AVS and determines the maximum of the correlation function of contours in both images. The position of the maximum is the necessary translation to match the two images. An example of the match procedure is given in Fig. 2. The accuracy of the method is approximately 1 mm at the isocenter. The match procedure was repeated for all the individual structures that were visible in the treatment portal. Translations in the craniocaudal and lateral directions and in-plane rotations were determined for each structure separately. As the reference image may represent a random position, relative movements were determined by comparing the translations and rotation for every image to the
Fig. 1. Examples of how structures, such as the clavicle, trachea, carina, thoracic wall, paraspinal line, and aortic notch were generally contoured.
calculated mean per sequence. Consequently, the distributions of the relative translations are independent of the phase of the reference image in the respiratory cycle and also independent of the interfractional movement. We as-
Fig. 2. An example of the match procedure. In this case the thoracic wall and trachea are contoured in the portal image and the reference image (top two images). Two different colors are used in each image. The visual result of the match is represented in the image at the bottom left, where the overlapping parts of the contours are visualized in a third color. The numerical results of the match are given on the right.
Landmark mobility and setup deviations
sumed that the position of patients did not change during data acquisition and also that the distribution of respirationinduced translations and rotations was independent of the radiation fraction. The relative translations and rotations of all the sequences (for each structure and each patient) could, consequently, be analyzed together. Standard deviations (SDs) of the relative movements were determined for each structure and each patient. For each patient, 3 distributions of relative movements were determined per structure. Analysis of intrafractional motion was also done for 3 different pairs of the structures that had been studied separately. Interobserver variation To determine the interobserver variation in the contouring of the thoracic structures and its effect on image comparison, 5 observers (2 radiation oncologists and 3 technologists) contoured and matched portal images with the corresponding digitized simulator image. The images of 3 or 4 fractions from 3 different patients were contoured and matched by each observer. Patient setup variation Data from 8 patients, in whom 5 or more fractions were imaged, were selected for studying patient setup. Of these, 4 patients were from the study of intrafractional motion and 4 were additional patients. As relatively large margins of 2 cm for the planning target volume (PTV) were already in use to account for both internal motion and setup variability, deviations detected in patient setup were not corrected during the course of treatment. This enabled us to study uncorrected variations in patient setup during a course of treatment. Based on the results of the intrafractional motion analysis, a combination of two structures was used for image comparison. The portal image of each fraction was contoured and matched with the digitized simulator film in order to determine translations and in-plane rotation. A field edge match was performed to determine the position of the actual field edge relative to the prescribed field edge. The difference between the anatomical contour match and the field contour match yields the actual setup deviation. The individual setup deviations were determined and were characterized by their means and SDs. Subsequently the overall, random and systemic setup deviations were determined for the group. The terms random and systematic setup variations used here are in accordance with the breakdown of setup variations by Bijhold et al. (1). The overall accuracy in patient setup is reflected by the SDs of the mean differences between the simulator films and portal images, averaged over all patients. The random variation (denoted by s after Bijhold et al.) was determined by calculating the spread (1 SD) of these differences around the corresponding mean in each patient and subsequently calculating the average of these SDs for the whole group. The systematic variation (denoted by S after Bijhold et al.) was calculated by determining the spread (1 SD) in the individual means of
● M. J. SAMSON et al.
829
Table 1. Frequency of structure visibility and number of matches per structure Structure
Patients
Matches
Matches/patient
Clavicle Trachea Carina Thoracic wall Paraspinal line Aortic notch
5 10 6 6 3 6
195 381 228 244 121 215
39 38 38 40 40 36
the differences between the simulator films and portal images. RESULTS Intrafractional motion Table 1 shows the frequency with which the various structures were visible in the treatment fields, the total number of matches per structure, and the average number of matches per patient. Other structures within the field were either not clearly visible (e.g., vertebra and ribs), or thought to be too mobile (e.g., heart and diaphragm). The averages of the SDs of relative translations for all patients are shown in Table 2. They are represented by SDx for relative movements in lateral directions, SDy for the craniocaudal directions, and SDr for rotations. Note that this data includes the inaccuracies of the matching procedure. The spread in lateral translations was limited (#1 mm, 1 SD) for the trachea, thoracic wall, paraspinal line, and aortic arch and similar results were found for craniocaudal translations of the clavicle and thoracic wall. The carina exhibited a somewhat greater movement (SDs .1 mm) in both directions. The spread for in-plane rotations was negligible (,0.5 deg., 1 SD) for all structures. The spread (1 SD) of the SDs is a measure for the interpatient variation in the extent of movement, and had a maximum value of 0.5 mm for the carina (data not shown). Analysis of intrafractional motion was repeated for various combinations of two structures that exhibited limited movement in perpendicular directions. This was done for three different combinations, and for two patients each Table 2. Intrafractional motion of anatomical structures
Clavicle Trachea Carina Thoracic wall Paraspinal line Aortic notch
SDx (mm)
SDy (mm)
SDr (°)
Xmax (mm)
Ymax (mm)
1.2 0.8 1.4 0.8 0.9 0.9
0.7 1.7 1.8 0.6 1.6 1.1
0.5 0.4 0.5 0.4 0.3 0.4
3.5 3.7 6.1 2.6 2.1 3.5
2.3 5.5 6.5 1.9 4.3 3.8
Abbreviations: SDx 5 SD of movements in the lateral direction; SDy 5 SD of movements in the craniocaudal direction; SDr 5 SD of rotations; Xmax 5 maximal lateral movement; Ymax 5 maximal craniocaudal movement.
830
I. J. Radiation Oncology
●
Biology
●
Physics
Volume 43, Number 4, 1999
Table 3. Intrafractional motion of structure pairs
Trachea–clavicle Thoracic wall–clavicle Thoracic wall–trachea
N
SDx (mm)
SDy (mm)
Xmax (mm)
Ymax (mm)
2 2 2
0.5 0.7 0.5
0.4 0.6 0.5
1.6 1.5 1.3
1.2 1.5 1.5
For definition of abbreviations, see Table 2.
(Table 3). Combining two structures not only increased stability in the direction of the largest movement, but also in the direction of the limited mobility (Fig. 3). The interobserver variation was defined as the spread in setup deviations determined by the 5 observers, averaged over all images. The interobserver variation in contouring the structures for matching was found to be 0.9 mm and 1.3 mm (1 SD) respectively in the lateral and craniocaudal direction. Patient setup variation Images from a total of 45 fractions were matched with their corresponding simulator films. A combination of two structures was used for image comparison. Measured translations are shown in Fig. 4. The plots show the mean translations and SDs of the distribution of measurements for each patient. For the group the overall setup deviations for translations in the lateral and craniocaudal direction were 3.1 mm and 3.2 mm (1 SD), respectively. The random deviations for translations in the lateral and craniocaudal direction were 2.0 mm and 2.8 mm (1 SD), respectively, while the corresponding systematic deviations were 2.5 and 2.0 mm (1 SD) (Table 4). Both the random and systematic deviations for in-plane rotations were negligible.
Fig. 4. Mean translations and their SDs for each patient in the cranial– caudal (A) and lateral (B) directions.
DISCUSSION At present, there is insufficient data available that makes it possible to select thoracic structures as stable landmarks. We have evaluated the mobility of different thoracic structures during radiation by generating multiple images with an EPID. Most of the normal structures that we evaluated had relative movements in the frontal plane which were of a similar magnitude as the accuracy of the image registration method used (1 mm at isocenter) and the interobserver variation in contouring (SD 5 1 mm). The structures that were relatively stable in at least one direction included the clavicle (in the craniocaudal axis), trachea, and paraspinal line (in the lateral direction), and the thoracic wall and aortic arch (in both directions). Only the carina was relatively Table 4. Setup deviations (1 SD) Fig. 3. Scatter plots representing the intrafractional mobility of the clavicle and trachea separately as compared to a combination of the clavicle and trachea. Each dot represents a measurement. The scatter plots show an increased stability in both directions for the combination as compared to both structures separately.
X (mm) Y (mm) r (°)
Overall
Random
Systematic
3.1 3.2 1.2
2.0 2.8 0.7
2.5 2.0 0.9
Landmark mobility and setup deviations
mobile in both directions. Even less variability was observed when combinations of two of these structures were used for image comparison. These results will allow the setup of patients with thoracic tumors to be accurately evaluated in the frontal plane. We are currently in the process of evaluating motion and setup deviations in the lateral projections as part of a three-dimensional setup evaluation and correction protocol. Earlier studies had reported large variations in patient setup for the thoracic region (8, 9, 11, 13). However, the definitions and methodology used in evaluating and reporting patient setup varied, which makes it difficult to compare results. Byhardt et al. (8) reported an average degree of misplacement of 1.0 cm (range: 0.5–2.0 cm) for tumors localized in the chest (lung and esophagus). Rabinowitz et al. (9) reported treatment-to-treatment variations for the thorax of approximately 4 mm and a simulation-to-treatment SD of 6 mm. A more recent study by Schewe et al. (13) reported SDs of setup deviations for the chest to be 7.1 and 8.3 mm for translations in the lateral and craniocaudal directions, respectively. These deviations are of a much larger magnitude than those found in our study. The differences probably reflect differences in the methodology used to evaluate patient setup, but could also be the result of differences in the stability of patient setup. The two main areas of uncertainty associated with portal imaging are sampling and interpretation (14). There are, for example, uncertainties associated with evaluating patient setup with port films. One study found significant discrepancies in same day field displacements detected by port filming as compared to electronic portal imaging (15). Interobserver variation in interpretation of the images and inaccuracies in measurements are other potential sources of inaccuracy when matching is not performed electronically. The use of unstable landmarks for matching may also lead to inaccuracies when studying patient setup. Errors may be introduced when mobile structures are used as landmarks to compare the relatively long exposure portal films (with resulting vague contours) to short exposure simulator films
● M. J. SAMSON et al.
831
(representing a random position of the structure). The evaluation of patient setup can be accurately performed under the following conditions: 1) when images are obtained during treatment by means of an EPID, 2) when image comparison is done (semi) automatically with image-matching algorithms, and 3) when stable thoracic structures are utilized as landmarks for image comparison. The reported deviations in our study reflect the uncorrected overall, systematic and random deviations, as the patient setup was not adjusted at any time during the course of treatment. The lack of significant intratreatment motion and the small random deviations derived from the analysis of patient setup appear to indicate that rigorous immobilization of patients treated for lung cancer is not necessary in our center. Focusing on systematic deviations may be of greater importance, as these persist throughout the course of treatment if uncorrected. Systematic deviations can be reduced by using a setup correction protocol. The decision rules must be based on knowledge of the probability distributions of systematic and random deviations (1, 14, 16, 17). Adjustment of patient setup without this knowledge may lead to erroneous decisions. The data presented here will permit a protocol for off-line correction of patient setup to be devised, and will allow the PTV margins for setup deviations to be reduced in conformal radiotherapy for lung cancer.
CONCLUSIONS We have identified a number of structures that exhibit little internal motion in the frontal plane, and recommend that a combination of these structures be used as anatomic landmarks for verifying patient setup during thoracic radiotherapy. The daily variation in the setup of patients with lung cancer in our center appears to be acceptable. Under these circumstances rigid immobilization seems unnecessary and smaller margins can be applied to account for patient setup deviations.
REFERENCES 1. Bijhold J, Lebesque JV, Hart AAM, et al. Maximizing setup accuracy using portal images as applied to a conformal boost technique for prostatic cancer. Radiother Oncol 1992;24:261– 271. 2. Creutzberg CA, Althof VG, Hoog de M, et al. A quality control study of the accuracy of patient positioning in irradiation of pelvic fields. Int J Radiat Oncol Biol Phys 1996;34: 697–708. 3. Creutzberg CA, Althof VG, Huizenga H, et al. Quality assurance using portal imaging: The accuracy of patient positioning in irradiation of breast cancer. Int J Radiat Oncol Biol Phys 1993;25:529 –539. 4. Fein DA, McGee KP, Schultheiss TE, et al. Intra- and interfractional reproducibility of tangential breast fields: A prospective on-line portal imaging study. Int J Radiat Oncol Biol Phys 1996;34:733–740.
5. Lirette A, Pouliot J, Aubin M, et al. The role of electronic portal imaging in tangential breast irradiation: A prospective study. Radiother Oncol 1995;37:241–245. 6. Huizenga H, Levendag PC, De Porre PMZR, et al. Accuracy in radiation field alignment in head and neck cancer: A prospective study. Radiother Oncol 1988;11:181–187. 7. Rosenthal SA, Galvin JM, Goldwein JW, et al. Improved methods for determination of variability in patient positioning for radiation therapy using simulation and serial portal film measurement. Int J Radiat Oncol Biol Phys 1992;23:621– 625. 8. Byhardt RW, Cox JD, Hornburg A, et al. Weekly localization films and detection of field placement errors. Int J Radiat Oncol Biol Phys 1978;4:881– 887. 9. Rabinowitz I, Broomberg J, Goitein M, et al. Accuracy of radiation field alignment in clinical practice. Int J Radiat Oncol Biol Phys 1985;11:1857–1867.
832
I. J. Radiation Oncology
●
Biology
●
Physics
10. Levine EL, Burt PA, Stout R, et al. The efficacy of megavoltage imaging in the radical radiotherapy of non-small cell lung cancer. Br J Radiol 1995;68:646 – 648. 11. Rodrigus P, Van den Weyngaert D, Van den Bogaert W. The value of treatment portal films in radiotherapy for bronchial carcinoma. Radiother Oncol 1987;9:27–31. 12. Leong JC, Stracher MA. Visualization of internal motion within a treatment portal during a radiation therapy treatment. Radiother Oncol 1987;9:153–156. 13. Schewe JE, Balter JM, Lam KL, et al. Measurement of patient setup errors using port films and a computer-aided graphical alignment tool. Med Dosim 1996;21:97–104.
Volume 43, Number 4, 1999
14. Denham JW, Dally MB, Hunter K, et al. Objective decisionmaking following a portal film: The results of a pilot study. Int J Radiat Oncol Biol Phys 1993;26:869 – 876. 15. Valicenti RK, Michalski JM, Bosch WR, et al. Is weekly port filming adequate for verifying patient position in modern radiation therapy? Int J Radiat Oncol Biol Phys 1994;30:431– 438. 16. Bel A, van Herk M, Bartelink H, et al. A verification procedure to improve patient set-up accuracy using portal images. Radiother Oncol 1993;29:253–260. 17. Kutcher GJ, Mageras GS, Leibel SA. Control, correction and modeling of setup errors and organ motion. Semin Radiat Oncol 1995;5:134 –145.
PII S0360-3016(98)00467-2
CLINICAL INVESTIGATION
Lung
AN ANALYSIS OF ANATOMIC LANDMARK MOBILITY AND SETUP DEVIATIONS IN RADIOTHERAPY FOR LUNG CANCER MICHAEL J. SAMSON, M.D.,* JOHN R. VAN SO¨ RNSEN DE KOSTE, B.SC.,* HANS C. J. DE BOER, M.SC.,† HANS TANKINK, B.SC.,* MARJOLEIN VERSTRAATE, B.SC.,* MARION ESSERS, PH.D.,† ANDRIES G. VISSER, PH.D.,† AND SURESH SENAN, M.R.C.P., F.R.C.R., PH.D.* Departments of *Radiation Oncology and †Clinical Physics, University Hospital Rotterdam–Daniel den Hoed Cancer Center/ University Hospital Rotterdam, Rotterdam, The Netherlands Purpose: To identify thoracic structures that exhibit little internal motion during irradiation and to determine setup variations in patients with lung cancer. Methods and Materials: Intrafractional images were generated with an electronic portal-imaging device from the AP fields of 10 patients, during several fractions. To determine the intrafractional mobility of thoracic structures, visible structures were contoured in every image and matched with a reference image by means of a crosscorrelation algorithm. Setup variations were determined by comparing portal images with the digitized simulator films using the stable structures as landmarks. Results: Mobility was limited in the lateral direction for the trachea, thoracic wall, paraspinal line, and aortic notch, and in the craniocaudal direction for the clavicle, aortic notch, and thoracic wall. Analysis of patient setup revealed random deviations of 2.0 mm (1 SD) in the lateral direction and 2.8 mm in the craniocaudal direction, while the systematic deviations were 2.5 and 2.0 mm (1 SD) respectively. Conclusions: We have identified thoracic structures that exhibit little internal motion in the frontal plane, and recommend that these structures be used for verifying patient setup during radiotherapy. The daily variation in the setup of lung cancer patients at our center appears to be acceptable. © 1999 Elsevier Science Inc. Portal imaging, Anatomic landmark mobility, Treatment verification, Radiotherapy.
The importance of evaluating and correcting setup errors during radiotherapy is well established. However, visual comparison of portal films with simulator films to extract setup deviations is a tedious process. Manual measurements of field edges and blocks relative to the anatomic landmarks may be subject to many inaccuracies, including interobserver variability. Electronic portal imaging devices (EPIDs) provide digital data that allow the evaluation of patient positioning during irradiation. These images can be compared with digitized simulator films or digitally reconstructed radiographs by using image-matching algorithms. These computerized techniques make the task of evaluating patient setup less laborious, fast and accurate, enabling on-line evaluation and the possibility of correcting deviations in patient positioning. Several centers, including our own, have studied the distribution of setup deviations in treatment sites such as the pelvis (1, 2), breast (3–5), and head and neck (6, 7). There is, however, a relative lack of data on patients with intrathoracic tumors. Early studies reported large discrepancies
between simulation and portal films of patients with thoracic lesions and also indicated significant treatment-totreatment variations to exist during irradiation (8, 9). However, these deviations were not extensively investigated in subsequent studies (10, 11). Structures in the thoracic region may be mobile during the respiratory and cardiac cycles, leading to uncertainty when they are used for image comparison. While (short exposure) simulator films may provide sharp images of mobile structures, they represent a random position of these structures in the respiratory or cardiac cycle. Consequently, the use of such structures for matching portal images with short exposure simulator films might introduce systematic deviations. Similarly, mobile structures in portal images may enlarge the extracted random deviations due to their blurred appearance (as a consequence of the long exposure) or their random position (as a consequence of the short exposure). Therefore, a reliable analysis of setup deviations of patients with thoracic lesions requires the use of stable landmarks. However, the available literature does not provide this information and the studies that have evaluated setup variability in the thoracic region have not always
Presented at the 39th Annual meeting of the American Society for Therapeutic Radiology and Oncology, Orlando, Fla., October 1997. Reprint requests to: M. J. Samson, Department of Radiation
Oncology, St. Elisabeth Hospital, Breedestraat (O), Curac¸ao, The Netherlands Antilles. Accepted for publication 29 October 1998.
INTRODUCTION
827
828
I. J. Radiation Oncology
●
Biology
●
Physics
Volume 43, Number 4, 1999
indicated the landmarks that have been used for matching images. The aim of this study was to identify thoracic structures that exhibit little internal motion during irradiation. These structures can, subsequently, be used to determine daily setup deviations during a course of radiotherapy of patients with lung cancer. METHODS AND MATERIALS After simulation and treatment planning, 14 patients with stage III non-small cell lung cancer were irradiated on a linear accelerator equipped with an EPID (Siemens Beam ViewPLUS). The patients were set up in the supine position, and aligned using lasers. The arms were either positioned above the head (12 patients), by holding on to a cross bar, or along the sides of the body (3 patients), depending on the irradiation technique. No other devices were used for patient immobilization. Intrafractional motion was studied in the first 10 patients. Variations in patient setup were evaluated in 4 of the former group (i.e., those in whom at least five fractions were imaged) and in 4 additional patients. Intrafractional motion In the first 10 patients, the stability of different anatomic landmarks in the thorax was determined by analyzing the multiple images generated during administration of a single fraction of radiotherapy (12). Between 6 to 12 sequential images were generated with the EPID; once every 2 seconds (exposure 0.54 sec; processing and “dead” time 1.5 sec) for the first 6 images and, thereafter, at a rate depending on the remaining dose to be delivered. These intrafractional images were generated for the anteroposterior (AP) field of patients during 3–5 fractions of the total radiation scheme. The acquired images were imported into the visualization software package Advanced Visual Systems (AVS) (Advanced Visual Systems, Inc., Waltham, MA). Visible structures such as the trachea, carina, the upper chest wall, aortic arch, clavicle, and paraspinal line (Fig. 1) were manually contoured in each image. In order to determine the mobility of each structure separately during a particular fraction, each image was matched with the first image (the reference image) of the corresponding sequence, by means of a crosscorrelation algorithm. The method has been developed using AVS and determines the maximum of the correlation function of contours in both images. The position of the maximum is the necessary translation to match the two images. An example of the match procedure is given in Fig. 2. The accuracy of the method is approximately 1 mm at the isocenter. The match procedure was repeated for all the individual structures that were visible in the treatment portal. Translations in the craniocaudal and lateral directions and in-plane rotations were determined for each structure separately. As the reference image may represent a random position, relative movements were determined by comparing the translations and rotation for every image to the
Fig. 1. Examples of how structures, such as the clavicle, trachea, carina, thoracic wall, paraspinal line, and aortic notch were generally contoured.
calculated mean per sequence. Consequently, the distributions of the relative translations are independent of the phase of the reference image in the respiratory cycle and also independent of the interfractional movement. We as-
Fig. 2. An example of the match procedure. In this case the thoracic wall and trachea are contoured in the portal image and the reference image (top two images). Two different colors are used in each image. The visual result of the match is represented in the image at the bottom left, where the overlapping parts of the contours are visualized in a third color. The numerical results of the match are given on the right.
Landmark mobility and setup deviations
sumed that the position of patients did not change during data acquisition and also that the distribution of respirationinduced translations and rotations was independent of the radiation fraction. The relative translations and rotations of all the sequences (for each structure and each patient) could, consequently, be analyzed together. Standard deviations (SDs) of the relative movements were determined for each structure and each patient. For each patient, 3 distributions of relative movements were determined per structure. Analysis of intrafractional motion was also done for 3 different pairs of the structures that had been studied separately. Interobserver variation To determine the interobserver variation in the contouring of the thoracic structures and its effect on image comparison, 5 observers (2 radiation oncologists and 3 technologists) contoured and matched portal images with the corresponding digitized simulator image. The images of 3 or 4 fractions from 3 different patients were contoured and matched by each observer. Patient setup variation Data from 8 patients, in whom 5 or more fractions were imaged, were selected for studying patient setup. Of these, 4 patients were from the study of intrafractional motion and 4 were additional patients. As relatively large margins of 2 cm for the planning target volume (PTV) were already in use to account for both internal motion and setup variability, deviations detected in patient setup were not corrected during the course of treatment. This enabled us to study uncorrected variations in patient setup during a course of treatment. Based on the results of the intrafractional motion analysis, a combination of two structures was used for image comparison. The portal image of each fraction was contoured and matched with the digitized simulator film in order to determine translations and in-plane rotation. A field edge match was performed to determine the position of the actual field edge relative to the prescribed field edge. The difference between the anatomical contour match and the field contour match yields the actual setup deviation. The individual setup deviations were determined and were characterized by their means and SDs. Subsequently the overall, random and systemic setup deviations were determined for the group. The terms random and systematic setup variations used here are in accordance with the breakdown of setup variations by Bijhold et al. (1). The overall accuracy in patient setup is reflected by the SDs of the mean differences between the simulator films and portal images, averaged over all patients. The random variation (denoted by s after Bijhold et al.) was determined by calculating the spread (1 SD) of these differences around the corresponding mean in each patient and subsequently calculating the average of these SDs for the whole group. The systematic variation (denoted by S after Bijhold et al.) was calculated by determining the spread (1 SD) in the individual means of
● M. J. SAMSON et al.
829
Table 1. Frequency of structure visibility and number of matches per structure Structure
Patients
Matches
Matches/patient
Clavicle Trachea Carina Thoracic wall Paraspinal line Aortic notch
5 10 6 6 3 6
195 381 228 244 121 215
39 38 38 40 40 36
the differences between the simulator films and portal images. RESULTS Intrafractional motion Table 1 shows the frequency with which the various structures were visible in the treatment fields, the total number of matches per structure, and the average number of matches per patient. Other structures within the field were either not clearly visible (e.g., vertebra and ribs), or thought to be too mobile (e.g., heart and diaphragm). The averages of the SDs of relative translations for all patients are shown in Table 2. They are represented by SDx for relative movements in lateral directions, SDy for the craniocaudal directions, and SDr for rotations. Note that this data includes the inaccuracies of the matching procedure. The spread in lateral translations was limited (#1 mm, 1 SD) for the trachea, thoracic wall, paraspinal line, and aortic arch and similar results were found for craniocaudal translations of the clavicle and thoracic wall. The carina exhibited a somewhat greater movement (SDs .1 mm) in both directions. The spread for in-plane rotations was negligible (,0.5 deg., 1 SD) for all structures. The spread (1 SD) of the SDs is a measure for the interpatient variation in the extent of movement, and had a maximum value of 0.5 mm for the carina (data not shown). Analysis of intrafractional motion was repeated for various combinations of two structures that exhibited limited movement in perpendicular directions. This was done for three different combinations, and for two patients each Table 2. Intrafractional motion of anatomical structures
Clavicle Trachea Carina Thoracic wall Paraspinal line Aortic notch
SDx (mm)
SDy (mm)
SDr (°)
Xmax (mm)
Ymax (mm)
1.2 0.8 1.4 0.8 0.9 0.9
0.7 1.7 1.8 0.6 1.6 1.1
0.5 0.4 0.5 0.4 0.3 0.4
3.5 3.7 6.1 2.6 2.1 3.5
2.3 5.5 6.5 1.9 4.3 3.8
Abbreviations: SDx 5 SD of movements in the lateral direction; SDy 5 SD of movements in the craniocaudal direction; SDr 5 SD of rotations; Xmax 5 maximal lateral movement; Ymax 5 maximal craniocaudal movement.
830
I. J. Radiation Oncology
●
Biology
●
Physics
Volume 43, Number 4, 1999
Table 3. Intrafractional motion of structure pairs
Trachea–clavicle Thoracic wall–clavicle Thoracic wall–trachea
N
SDx (mm)
SDy (mm)
Xmax (mm)
Ymax (mm)
2 2 2
0.5 0.7 0.5
0.4 0.6 0.5
1.6 1.5 1.3
1.2 1.5 1.5
For definition of abbreviations, see Table 2.
(Table 3). Combining two structures not only increased stability in the direction of the largest movement, but also in the direction of the limited mobility (Fig. 3). The interobserver variation was defined as the spread in setup deviations determined by the 5 observers, averaged over all images. The interobserver variation in contouring the structures for matching was found to be 0.9 mm and 1.3 mm (1 SD) respectively in the lateral and craniocaudal direction. Patient setup variation Images from a total of 45 fractions were matched with their corresponding simulator films. A combination of two structures was used for image comparison. Measured translations are shown in Fig. 4. The plots show the mean translations and SDs of the distribution of measurements for each patient. For the group the overall setup deviations for translations in the lateral and craniocaudal direction were 3.1 mm and 3.2 mm (1 SD), respectively. The random deviations for translations in the lateral and craniocaudal direction were 2.0 mm and 2.8 mm (1 SD), respectively, while the corresponding systematic deviations were 2.5 and 2.0 mm (1 SD) (Table 4). Both the random and systematic deviations for in-plane rotations were negligible.
Fig. 4. Mean translations and their SDs for each patient in the cranial– caudal (A) and lateral (B) directions.
DISCUSSION At present, there is insufficient data available that makes it possible to select thoracic structures as stable landmarks. We have evaluated the mobility of different thoracic structures during radiation by generating multiple images with an EPID. Most of the normal structures that we evaluated had relative movements in the frontal plane which were of a similar magnitude as the accuracy of the image registration method used (1 mm at isocenter) and the interobserver variation in contouring (SD 5 1 mm). The structures that were relatively stable in at least one direction included the clavicle (in the craniocaudal axis), trachea, and paraspinal line (in the lateral direction), and the thoracic wall and aortic arch (in both directions). Only the carina was relatively Table 4. Setup deviations (1 SD) Fig. 3. Scatter plots representing the intrafractional mobility of the clavicle and trachea separately as compared to a combination of the clavicle and trachea. Each dot represents a measurement. The scatter plots show an increased stability in both directions for the combination as compared to both structures separately.
X (mm) Y (mm) r (°)
Overall
Random
Systematic
3.1 3.2 1.2
2.0 2.8 0.7
2.5 2.0 0.9
Landmark mobility and setup deviations
mobile in both directions. Even less variability was observed when combinations of two of these structures were used for image comparison. These results will allow the setup of patients with thoracic tumors to be accurately evaluated in the frontal plane. We are currently in the process of evaluating motion and setup deviations in the lateral projections as part of a three-dimensional setup evaluation and correction protocol. Earlier studies had reported large variations in patient setup for the thoracic region (8, 9, 11, 13). However, the definitions and methodology used in evaluating and reporting patient setup varied, which makes it difficult to compare results. Byhardt et al. (8) reported an average degree of misplacement of 1.0 cm (range: 0.5–2.0 cm) for tumors localized in the chest (lung and esophagus). Rabinowitz et al. (9) reported treatment-to-treatment variations for the thorax of approximately 4 mm and a simulation-to-treatment SD of 6 mm. A more recent study by Schewe et al. (13) reported SDs of setup deviations for the chest to be 7.1 and 8.3 mm for translations in the lateral and craniocaudal directions, respectively. These deviations are of a much larger magnitude than those found in our study. The differences probably reflect differences in the methodology used to evaluate patient setup, but could also be the result of differences in the stability of patient setup. The two main areas of uncertainty associated with portal imaging are sampling and interpretation (14). There are, for example, uncertainties associated with evaluating patient setup with port films. One study found significant discrepancies in same day field displacements detected by port filming as compared to electronic portal imaging (15). Interobserver variation in interpretation of the images and inaccuracies in measurements are other potential sources of inaccuracy when matching is not performed electronically. The use of unstable landmarks for matching may also lead to inaccuracies when studying patient setup. Errors may be introduced when mobile structures are used as landmarks to compare the relatively long exposure portal films (with resulting vague contours) to short exposure simulator films
● M. J. SAMSON et al.
831
(representing a random position of the structure). The evaluation of patient setup can be accurately performed under the following conditions: 1) when images are obtained during treatment by means of an EPID, 2) when image comparison is done (semi) automatically with image-matching algorithms, and 3) when stable thoracic structures are utilized as landmarks for image comparison. The reported deviations in our study reflect the uncorrected overall, systematic and random deviations, as the patient setup was not adjusted at any time during the course of treatment. The lack of significant intratreatment motion and the small random deviations derived from the analysis of patient setup appear to indicate that rigorous immobilization of patients treated for lung cancer is not necessary in our center. Focusing on systematic deviations may be of greater importance, as these persist throughout the course of treatment if uncorrected. Systematic deviations can be reduced by using a setup correction protocol. The decision rules must be based on knowledge of the probability distributions of systematic and random deviations (1, 14, 16, 17). Adjustment of patient setup without this knowledge may lead to erroneous decisions. The data presented here will permit a protocol for off-line correction of patient setup to be devised, and will allow the PTV margins for setup deviations to be reduced in conformal radiotherapy for lung cancer.
CONCLUSIONS We have identified a number of structures that exhibit little internal motion in the frontal plane, and recommend that a combination of these structures be used as anatomic landmarks for verifying patient setup during thoracic radiotherapy. The daily variation in the setup of patients with lung cancer in our center appears to be acceptable. Under these circumstances rigid immobilization seems unnecessary and smaller margins can be applied to account for patient setup deviations.
REFERENCES 1. Bijhold J, Lebesque JV, Hart AAM, et al. Maximizing setup accuracy using portal images as applied to a conformal boost technique for prostatic cancer. Radiother Oncol 1992;24:261– 271. 2. Creutzberg CA, Althof VG, Hoog de M, et al. A quality control study of the accuracy of patient positioning in irradiation of pelvic fields. Int J Radiat Oncol Biol Phys 1996;34: 697–708. 3. Creutzberg CA, Althof VG, Huizenga H, et al. Quality assurance using portal imaging: The accuracy of patient positioning in irradiation of breast cancer. Int J Radiat Oncol Biol Phys 1993;25:529 –539. 4. Fein DA, McGee KP, Schultheiss TE, et al. Intra- and interfractional reproducibility of tangential breast fields: A prospective on-line portal imaging study. Int J Radiat Oncol Biol Phys 1996;34:733–740.
5. Lirette A, Pouliot J, Aubin M, et al. The role of electronic portal imaging in tangential breast irradiation: A prospective study. Radiother Oncol 1995;37:241–245. 6. Huizenga H, Levendag PC, De Porre PMZR, et al. Accuracy in radiation field alignment in head and neck cancer: A prospective study. Radiother Oncol 1988;11:181–187. 7. Rosenthal SA, Galvin JM, Goldwein JW, et al. Improved methods for determination of variability in patient positioning for radiation therapy using simulation and serial portal film measurement. Int J Radiat Oncol Biol Phys 1992;23:621– 625. 8. Byhardt RW, Cox JD, Hornburg A, et al. Weekly localization films and detection of field placement errors. Int J Radiat Oncol Biol Phys 1978;4:881– 887. 9. Rabinowitz I, Broomberg J, Goitein M, et al. Accuracy of radiation field alignment in clinical practice. Int J Radiat Oncol Biol Phys 1985;11:1857–1867.
832
I. J. Radiation Oncology
●
Biology
●
Physics
10. Levine EL, Burt PA, Stout R, et al. The efficacy of megavoltage imaging in the radical radiotherapy of non-small cell lung cancer. Br J Radiol 1995;68:646 – 648. 11. Rodrigus P, Van den Weyngaert D, Van den Bogaert W. The value of treatment portal films in radiotherapy for bronchial carcinoma. Radiother Oncol 1987;9:27–31. 12. Leong JC, Stracher MA. Visualization of internal motion within a treatment portal during a radiation therapy treatment. Radiother Oncol 1987;9:153–156. 13. Schewe JE, Balter JM, Lam KL, et al. Measurement of patient setup errors using port films and a computer-aided graphical alignment tool. Med Dosim 1996;21:97–104.
Volume 43, Number 4, 1999
14. Denham JW, Dally MB, Hunter K, et al. Objective decisionmaking following a portal film: The results of a pilot study. Int J Radiat Oncol Biol Phys 1993;26:869 – 876. 15. Valicenti RK, Michalski JM, Bosch WR, et al. Is weekly port filming adequate for verifying patient position in modern radiation therapy? Int J Radiat Oncol Biol Phys 1994;30:431– 438. 16. Bel A, van Herk M, Bartelink H, et al. A verification procedure to improve patient set-up accuracy using portal images. Radiother Oncol 1993;29:253–260. 17. Kutcher GJ, Mageras GS, Leibel SA. Control, correction and modeling of setup errors and organ motion. Semin Radiat Oncol 1995;5:134 –145.