Practical Radiation Oncology (2014) 4, e59–e65
www.practicalradonc.org
Original Report
Uncertainties of 4-dimensional computed tomographybased tumor motion measurement for lung stereotactic body radiation therapy Fan Zhang, Chris R. Kelsey, David Yoo, Fang-Fang Yin, Jing Cai ⁎ Department of Radiation Oncology, Duke University Medical Center, Durham, North Carolina Received 31 December 2012; revised 18 February 2013; accepted 19 February 2013
Abstract Purpose: To evaluate how well tumor motion measured prior to treatment based on 4-dimensional computer tomography (4DCT) reflects actual tumor motion during beam-on throughout the course of treatment. Methods and Materials: Twenty-three patients who had lung stereotactic body radiation therapy (SBRT) treatments were retrospectively selected. All patients had 4DCT simulation for treatment planning, from which tumor motion ranges were measured (R4DCT). Tumor motion was monitored during treatment using megavoltage (MV) imaging. Tumor motion trajectories were extracted from cine MV images and were used to determine mean and maximum tumor motion range (Mean RMV, Max RMV) throughout entire course of treatment. Comparison and correlations between mean and max RMV and R4DCT were calculated. Results: On average, an insignificant difference was found between mean RMV and R4DCT (P = .67, mean [± SD] difference = −0.7 [± 1.6] mm); meanwhile a significant difference was found between Max RMV and R4DCT (P = .03, mean [± SD] difference = 1.9 [± 1.6] mm). The difference between RMV and R4DCT was found inversely proportional to R4DCT (Y = −0.4X + 0.6, r = 0.76). Max RMV was greater than R4DCT in all patients; difference between the 2 showed no correlation with R4DCT (Y = −0.02X + 1.9, r = 0.05). Correlation between Mean RMV and R4DCT and between Max RMV and R4DCT can be expressed as Y = 0.7X (r = 0.88) and Y = 0.8X (r = 0.50), respectively. The same analysis performed on tumors that moved less than 5 mm from 4DCT revealed the following correlations: Y = 1.3X (r = 0.83) and Y = 1.7X (r = 0.49). Conclusions: Tumor motion measured from 4DCT approximates the overall average tumor motion range, but consistently underestimates the overall maximum tumor motion range. These findings may lead to a potential strategy for managing uncertainties of 4DCT in the application of lung SBRT. © 2014 American Society for Radiation Oncology. Published by Elsevier Inc. All rights reserved.
Introduction Conflicts of interest: None. ⁎ Corresponding author. Department of Radiation Oncology, Duke University Medical Center, Box 3295, Durham, NC 27710. E-mail address: [email protected] (J. Cai).
Respiratory motion poses a great challenge for precise tumor localization and dose delivery in radiation therapy. 1 Motion management is critical in radiation therapy (RT) for treating moving targets, particularly in stereotactic
1879-8500/$ – see front matter © 2014 American Society for Radiation Oncology. Published by Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.prro.2013.02.009
e60
F. Zhang et al
Practical Radiation Oncology: January-February 2014
body radiation therapy (SBRT) where the radiation dose distribution is highly conformal to the target volume and dose falls off sharply outside the target volume. 2,3 In the clinical process of RT, respiratory motion measurement is primarily performed during simulation to determine tumor motion range and internal target volume (ITV), and during treatment to verify tumor motion within preset margins or to guide the treatment. Four-dimensional CT (4DCT) is the current clinical standard for imaging tumor respiratory motion, especially for tumors in the thorax. 1,4 While 4DCT has been widely adapted in the clinic, it is not without limitations. The main limitation of 4DCT lies in the fact that it produces only 1 effective breathing cycle of data although it is acquired over several breathing cycles. Because real patients’ breathing can be very irregular and the breathing pattern may change from day to day, tumor motion measurement based on 1 breathing cycle of 4DCT cannot fully characterize the tumor motion. Using tumor motion measured from 4DCT for lung SBRT treatment may thus introduce uncertainties in target volume delineation and patient positioning. Understanding and reducing these uncertainties can potentially lead to further
Table 1 Summary of patient characteristics and results of tumor motion measurements Patient Age Gender GTV Tumor R4DCT Mean Max (y) (cm 3) location (mm) RMV RMV (mm) (mm) 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23
69 79 80 70 75 77 72 83 55 75 57 57 70 47 41 63 80 65 78 72 84 59 78
M M M F F F M M M F F M F M M M M M M M M F F
7.5 9.5 30.9 20.2 16.9 21.4 2.6 9.2 4.3 10.3 7.1 2.9 3.9 9.5 5.4 2.0 2.9 7.8 4.6 6.0 2.6 8.5 6.0
MR LR UL UL ML UL MR UL UR UL UL UR UR LR UR UR ML ML LR UR MR UL MR
0.5 7.7 1.2 0.7 1.1 9.1 6.5 2.3 2.1 0.2 5.1 8.0 1.9 4.8 3.3 2.0 1.0 3.2 13.5 5.4 1.1 0.8 4.7
0.5 7.8 1.1 1.0 2.2 5.9 5.2 2.2 3.9 0.6 4.3 5.0 2.1 3.8 1.6 4.4 1.4 1.7 9.5 3.1 0.5 0.4 2.8
1.4 11.9 2.0 2.9 4.1 11.2 9.5 5.2 7.0 1.5 5.3 10.0 5.1 7.1 4.4 7.8 5.2 5.6 17.6 9.2 2.3 2.2 5.8
4DCT, 4-dimensional computed tomography; GTV, gross tumor volume; LR, lower right lung lobe; ML, middle left lung lobe; MR, middle right lung lobe; R4DCT, tumor motion range measured from 4DCT; RMV, tumor motion range measured from cine MV images; UL, upper left lung lobe; UR, upper right lung lobe.
improvement in the efficacy of lung SBRT. However, evaluation and management of these uncertainties are challenging partly due to technical limitations to image tumor motion during the treatment. Previous studies have analyzed tumor motion uncertainties based on information acquired with 4DCT, 5 magnetic resonance imaging, 6 fluoroscopy, 7 etc, before or after the treatment. Implanted markers have been used to improve tracking accuracy, but shown to have risk of pneumothorax. 8-10 Recently, a few studies have shown the feasibility of tracking tumor motion during beam-on with cine megavoltage (MV) imaging using electronic portal imaging devices (EPID). 11,12 One fundamental question regarding the uncertainty of 4DCT lies in how accurately tumor motion measured with 4DCT before the treatment reflects actual tumor motion during beam-on throughout the entire course of lung SBRT treatment. It is of clinical interest to understand whether there is a significant difference in tumor motion measurements between CT simulation and treatment, and whether the difference (if it exists) is random or systematic? It is therefore the aim of this study to answer these questions by statistically comparing tumor motion measured with 4DCT during CT simulation with those measured with cine MV during beam-on throughout the entire lung SBRT treatment. We will only measure tumor motion during beam-on because tumor motion during beam-off, such as between different beams and between different treatment fractions, does not contribute to the actually delivered dose to the tumor.
Methods and materials Patients and lung SBRT treatment Twenty-three lung cancer patients (15 male, 8 female, mean age 68.9 years) who received SBRT treatments in our institution were selected for this study. Table 1 summarizes the characteristics of the patients. All patients underwent 4DCT simulation on a 4-slice CT scanner (Lightspeed Plus 4; GE Healthcare, Waukesha, WI) with slice thickness of 2.5 mm. The real-time position management system (Varian Medical System, Palo Alto, CA) was used to monitor patient’s breathing for retrospective sorting. All patients were imaged in an immobilization device during free breathing. All treatment plans were created using a 3D conformal technique with 611 coplanar beams (no couch kick) and wedges (if needed) with an Eclipse treatment planning system (Varian Medical System). Analytical anisotropy algorithm with heterogeneity correction was used for dose calculation. A Novalis Tx machine (Varian Medical System), equipped with kilovoltage on-board imaging, cone-beam CT (CBCT), and MV EPID was used for lung SBRT treatment. Before the treatment, patients were first
Practical Radiation Oncology: January-February 2014
Uncertainties in 4DCT tumor motion assessment
e61
Figure 1 Example of continuously acquired cine megavolt (MV) images illustrating the motion of lung tumor. The template image is indicated by the rectangular box. Tumor is indicated by the free-hand contour.
positioned using lasers, then kV orthogonal on-board imaging, and CBCT by matching the tumor. The CBCT was acquired using half-rotation (199.2 degrees) with a gantry rotation speed of 6 degrees per second, resulting in 333 projections per scan. Tumor position and motion was then verified with kV fluoroscopy immediately before the treatment and with cine MV images during the treatment (frame rate = 1 frame/second). Beam-on time for each beam ranged from about 20 seconds to 80 seconds, depending upon the prescription dose and wedges that were used.
Data analysis For each patient, the 10-phase 4DCT images and cine MV images of all beams from all fractions were retrospectively collected and analyzed. For cine MV images, tumor motion trajectories were extracted using an automatic motion tracking algorithm. Cine MV images acquired using EPID from exiting treatment beams provide a beams-eye view of the internal lung tumor motion in 2 dimensions. It has been shown that real-time tumor motion during beam-on can be quantified using cine MV images. 11-13 In this study, we used an automatic tumor motion tracking algorithm to determine the tumor
motion trajectories from cine MV images. Details of the tracking algorithm and its validation have been described in a previous study 14 and will only be summarized here. Figure 1 illustrates a series of cine MV images acquired during the lung SBRT treatment, where the tumor is readily seen in each frame. To extract the tumor motion trajectory, a template image (white box in Fig 1) encompassing the entire tumor was first created manually from the first frame of the cine MV images. Tumor position in the following frames was then determined by searching for the maximal cross-correlative region to match the template image. To qualitatively evaluate whether the algorithm correctly tracked the tumor motion, the actual tumor positions in the cine MV images were visually compared against the tumor contour (red contour in Fig 1) that moves with the calculated shifts in a timeresolved fashion. If apparent (visually detectable) discrepancies between the actual tumor position and the tumor contour were observed, the tracking results (tumor motion trajectories) were considered “un-usable” and were discarded. If no apparent discrepancies were found, the tracking results were considered as “usable” for further analysis. The accuracy of the “usable” tracking results has been investigated previously and was found to be within a millimeter. 10 The validation of tracking algorithm for
Figure 2 Example of the superior-inferior tumor motion trajectories extracted from cine megavoltage (MV) images of a single beam in 5 fractions.
e62
F. Zhang et al
Figure 3 Example of a tumor motion probability density distribution (PDF) derived from cine megavoltage (MV) images and the definition of Max RMV. Original PDF is shown in dark blue bars and the smoothed PDF is shown in red curve. Light blue lines are the 0.5% cutoff probability threshold that is used to calculate Max RMV.
motion measurement using different imaging modality (cine MRI and cine MV) has been demonstrated in our previous publication. 15 Both vertical and horizontal tumor motions were extracted from the cine MV images, but only the vertical motion was analyzed. Because all beams were coplanar, vertical motion in cine MV was essentially the superiorinferior (SI) motion. Figure 2 shows an example of the SI tumor motion trajectories of the same beam from 5 treatment fractions, in which intrafractional and interfractional variations of tumor motion were clearly seen. Mean displacement of each tumor motion trajectory was set to zero. This is so that the pretreatment patient positioning
Practical Radiation Oncology: January-February 2014
matched the mean tumor position between the CBCT and planning CT. An overall mean and max tumor motion range throughout the entire course of treatment, labeled as Mean RMV and Max RMV, respectively, was determined from cine MV images for each patient using all “usable” tumor motion trajectories from all treatment fractions. Mean RMV was simply calculated as the average distance between all respiratory peaks and respiratory valleys at immediate neighbors. To determine Max RMV, an overall tumor motion probability density distribution (PDF) was first generated for each patient using all motion trajectories of the patient that were derived from cine MV images of all beams from all fractions. The resultant tumor motion PDF reveals the probability of finding the tumor at a specific location relative to the mean tumor position. It should be noted that an occasional deep inspiration or cough may result in significantly large but uncharacteristic tumor displacements. The dosimetric effects of these extremely low probability incidences are expected to be negligible. To avoid such uncertainty and determine an effective Max RMV, a cutoff probability threshold of 0.5% was first applied to the tumor motion PDF, prior to the calculation of Max RMV. Figure 3 illustrates an example of the overall tumor motion PDF and the determination of the effective Max RMV. The cutoff threshold is an empirical value purely based on our clinical experience. The purpose of applying the threshold is to remove the effect of sudden movement of the tumor, such as cough, on the measurement of the tumor motion range. For simplicity and convenience, Max RMV was still used instead of effective Max RMV throughout this article. For 4DCT, tumor motion trajectories were automatically determined using the same cross-correlation algorithm as was used for cine MV images. Tumor motion range in the SI direction was calculated as the distance
Figure 4 Differences between Mean RMV and R4DCT (A) and between Max RMV and R4DCT (B) as a function of R4DCT. 4DCT, 4dimensional computed tomography; MV, megavoltage.
Practical Radiation Oncology: January-February 2014
Uncertainties in 4DCT tumor motion assessment
e63
Figure 5 Correlations between Mean RMV and R4DCT (A) and between Max RMV and R4DCT (B). Solid lines are for all patients; dotted lines are for patients whose tumor motion ranges are less than 5 mm based on 4DCT. 4DCT, 4-dimensional computed tomography; MV, megavoltage.
between the most superior and the most inferior tumor positions, labeled as R4DCT. Tumor motion ranges measured on cine MV (Mean RMV, Max RMV) were compared with those measured on 4DCT (R4DCT) using a paired signed rank test with a significance level of 0.05. Correlations between Mean and Max RMV and R4DCT were studied for all patients and for patients whose tumors moved less than 5 mm on the 4DCT.
Results Table 1 summarizes the tumor motion measurements for all patients. On average of all patients, insignificant differences were found between Mean RMV and R4DCT (P = .67, mean [± SD] difference = −0.7 [± 1.6] mm). The difference between RMV and R4DCT was found inversely proportional to R4DCT (Y = −0.4X + 0.6, r = 0.76), as shown in Fig 4A. Significant differences were found between Max RMV and R4DCT (P = .03, mean [± SD] difference = 1.9 [± 1.6] mm). Max RMV was greater than R4DCT in all patients; the difference between the 2 showed no correlation with R4DCT (Y = −0.02X + 1.9, r = 0.05), as shown in Fig 4B. Figure 5 shows the correlation between Mean RMV and R4DCT and between Max RMV and R4DCT, which can be expressed as Y = 0.7X (r = 0.88) and Y = 0.8X (r = 0.50), respectively. These results indicate that in general Mean RMV tends to be smaller than R4DCT, while Max RMV tends to be greater than R4DCT. The same analysis performed on tumors that moved less than 5 mm on 4DCT revealed the following correlations: Y = 1.3X (r = 0.83) and Y = 1.7X (r = 0.49), implying both Mean and Max RMV tend to be greater than R4DCT when R4DCT is less than 5 mm.
Discussion In this study, we investigated the correlation between tumor motion measured from 4DCT at CT simulation and tumor motion measured from cine MV images during beam-on throughout the entire course of lung SBRT treatment. In general the tumor motion range measured from 4DCT does estimate the mean tumor motion range exhibited throughout the course of treatment. Differences between the 2 exist, but are not significant. However, it was found that the tumor motion range measured on 4DCT was consistently smaller than the maximum tumor motion range measured throughout the course of treatment across all patients. This systematic difference between the 2 is presumably due to patient breathing variations, which is expected and confirmed with our study via statistical analysis of tumor motion throughout the course of treatment. In addition, this study quantitatively revealed the relationship between the 2, as illustrated in Fig 5B. Interestingly, it was found that this systematic difference is more substantial for tumors that move small amplitudes versus those that move large amplitudes. As a result, this systematic difference is rather persistent, around 1.9 mm, and is independent of the tumor motion range measured from 4DCT. While it is well known that tumor motion exhibits large intrafractional and interfractional variations, 16 there is still lack of proper strategies to manage these variations. There have been a few studies conducted to investigate the clinical impact of these variations, especially in the application of lung SBRT, and to explore potential solutions. For example, Cai et al 6 and Simon et al 17 demonstrated in separate studies that 4DCT can lead to errors in delineation of tumor ITV and tumor motion PDF due to breathing variations. Ge et al 18 in a recent
e64
F. Zhang et al
publication proposed to minimize these uncertainties in determining ITV for lung SBRT by combining information from 3DCT and 4DCT. In this study, we quantitatively evaluated the relationships in tumor motion between 4DCT and cine MV images. The results provide additional, and more importantly, statistical information about tumor motion variations during lung SBRT, which may potentially aid the development of a simple yet strategic method to manage these uncertainties during the application of lung SBRT. Tumor respiratory motion is inherently complex. Tumor motion measurement based on only 1 effective breathing cycle extracted from 4DCT is evidently insufficient. One potential solution may include measuring statistical characteristics of tumor motion such as mean and max motion range, motion PDF, and motion variability, to provide detailed information about the tumor motion. It is believed that these statistical characteristics are more reproducible from day to day than a single measurement of tumor motion range. In fact, our previous studies have shown that tumor motion PDF maintains high reproducibility between fractions. 19 Tumor motion in only the SI direction was investigated in this study. Tumor motion in the anterior-posterior (AP) and medial-lateral (ML) directions can be nontrivial for some patients, especially in the AP direction. However, imaging AP and right-left tumor motion using cine MV requires beam angle to be 90 degrees/270 degrees and 0 degrees/180 degrees, respectively. Due to various clinical considerations, it is not guaranteed that these beams can be used in treatment planning. As a result, we cannot obtain enough “usable” cine MV images to extract the AP and ML tumor motion trajectories. However, the differences between R4DCT and RMV in the SI direction as observed in this study were presumably due to patient’s intrafractional and interfractional breathing variations; its effects can be similarly applied to tumor motion in the AP and ML directions. Therefore, we expect a similar trend, though not necessarily of the same magnitude, of correlations and differences to exist between R4DCT and RMV in the AP and ML directions to that observed in the SI direction. The cine MV images were measured at a frequency of 1 frame/second in this study. The temporal resolution is therefore relatively low for accurately measuring tumor respiratory motion. It was set clinically this way to prevent the system from slowing down or crashing due to image overloading. This suboptimal frequency may affect the accuracy of the motion trajectories and the tumor motion PDF. However, its impact is expected to be small because of the following: (1) this low frequency can be largely compensated by using a large number of data points. In this study the PDF was typically calculated using 12 tumor motion trajectories (3 “usable” beams per fraction and 4 total fractions), which correspond to approximately a total of 12 minutes of image acquisition time and a total of 720 data points; and (2) our previous study 12 showed that the
Practical Radiation Oncology: January-February 2014
accuracy of the PDF exponentially increases as the image acquisition rate increases, but reaches a plateau when the frame rate is greater than 2 frames/second. Therefore, the estimated error in the PDF calculation at a frame rate of 1 frame/second is approximately 1.5%. For future studies, especially to study tumor motion variability based on cine MV, a higher frame rate (≥ 2 frames/second) is desirable. This is clinically feasible because most systems can set the frame rate up to 8 frames/second.
Conclusions We systematically compared and quantitatively correlated tumor motion measured from 4DCT prior to treatment with tumor motion measured from cine MV images during the course of lung SBRT treatment. We found that tumor motion measured from 4DCT approximates the overall average tumor motion range, but consistently underestimates the overall maximum tumor motion range exhibited during treatment. Findings of our study may lead to a potential strategy for managing uncertainties of 4DCT in the application of lung SBRT.
Acknowledgment The authors thank Irina Vergalasova for editing the manuscript and Jing Hu, BS, CMD, for analyzing part of the cine MV data.
References 1. Mageras GS, Pevsner A, Yorke ED, et al. Measurement of lung tumor motion using respiration-correlated CT. Int J Radiat Oncol Biol Phys. 2004;60:933-941. 2. Nagata Y, Negoro Y, Aoki T, et al. Clinical outcomes of 3D conformal hypofractionated single high-dose radiotherapy for one or two lung tumors using a stereotactic body frame. Int J Radiat Oncol Biol Phys. 2002;52:1041-1046. 3. Onishi H, Araki T, Shirato H, et al. Stereotactic hypofractionated high-dose irradiation for stage I nonsmall cell lung carcinoma: clinical outcomes in 245 subjects in a Japanese multiinstitutional study. Cancer. 2004;101:1623-1631. 4. Keall PJ, Mageras GS, Balter JM, et al. The management of respiratory motion in radiation oncology report of AAPM Task Group 76. Med Phys. 2006;33:3874-3900. 5. Redmond KJ, Song DY, Fox JL, Zhou J, Rosenzweig CN, Ford E. Respiratory motion changes of lung tumor over the course of radiation therapy based on respiration-correlated four dimension computed tomography scans. Int J Radiat Oncol Biol Phys. 2009;75: 1605-1612. 6. Cai J, Read PW, Sheng K. The effect of respiratory motion variability and tumor size on the accuracy of average intensity projection from four-dimensional computed tomography: an investigation based on dynamic MRI. Med Phys. 2008;35:4974-4981. 7. Kupelian PA, Forbes A, Willoughby TR, et al. Implantation and stability of metallic fiducials within pulmonary lesions. Int J Radiat Oncol Biol Phys. 2007;69:777-785.
Practical Radiation Oncology: January-February 2014 8. Shirato H, Harada T, Harabayashi T, et al. Feasibility of insertion/implantation of 2.0-mm-diameter gold internal fiducial markers for precise setup and real-time tumor tracking in radiotherapy. Int J Radiat Oncol Biol Phys. 2003;56:240-247. 9. Roman NO, Shepherd W, Mukhopadhyay N, et al. Interfractional positional variability of fiducial markers and primary tumors in locally advanced non-small-cell lung cancer during audiovisual biofeedback radiotherapy. Int J Radiat Oncol Biol Phys. 2012;83: 1566-1572. 10. Imura M, Yamazaki K, Shirato H, et al. Insertion and fixation of fiducial markers for setup and tracking of lung tumors in radiotherapy. Int J Radiat Oncol Biol Phys. 2005;63:1442-1447. 11. Meyer J, Richter A, Baier K, Wilbert J, Guckenberger M, Flentje M. Tracking moving objects with megavoltage portal imaging: a feasibility study. Med Phys. 2006;33:1275-1280. 12. Tai A, Christensen JD, Gore E, Khamene A, Boettger T, Li XA. Gated treatment delivery verification with on-line megavoltage fluoroscopy. Int J Radiat Oncol Biol Phys. 2010;76:1592-1598. 13. Richter A, Wilbert J, Baier K, Flentje M, Guckenberger M. Feasibility study for markerless tracking of lung tumors in stereotactic body radiotherapy. Int J Radiat Oncol Biol Phys. 2010;78:618-627.
Uncertainties in 4DCT tumor motion assessment
e65
14. Zhang F, Hu J, Kelsey CR, et al. Reproducibility of tumor motion probability distribution function in stereotactic body radiation therapy of lung cancer. Int J Radiat Oncol Biol Phys. 2012;84: 861-866. 15. Chang Z, Cai J, Yin FF. Evaluation of motion measurement using cine MRI for image guided stereotactic body radiotherapy on a new phantom platform. J Radiosurgery and SBRT. 2011:109-115. 16. Benedict SH, Yenice KM, Followill D, et al. Stereotactic body radiation therapy: the report of AAPM Task Group 101. Med Phys. 2010;37:4078-4101. 17. Simon L, Giraud P, Servois V, Rosenwald JC. Initial evaluation of a four-dimensional computed tomography system using a programmable motor. J Appl Clin Med Phys. 2006;7:50-65. 18. Ge H, Cai J, Kelsey C, et al. Quantification and minimization of uncertainties of internal target volume (ITV) for stereotactic-body radiation therapy (SBRT) of lung cancer. Int J Radiat Oncol Biol Phys. 2013;85:438-443. 19. Cai J, Read PW, Altes TA, Molloy JA, Brookeman JR, Sheng K. Evaluation of the reproducibility of lung motion probability distribution function (PDF) using dynamic MRI. Phys Med Biol. 2007;52:365-373.
www.practicalradonc.org
Original Report
Uncertainties of 4-dimensional computed tomographybased tumor motion measurement for lung stereotactic body radiation therapy Fan Zhang, Chris R. Kelsey, David Yoo, Fang-Fang Yin, Jing Cai ⁎ Department of Radiation Oncology, Duke University Medical Center, Durham, North Carolina Received 31 December 2012; revised 18 February 2013; accepted 19 February 2013
Abstract Purpose: To evaluate how well tumor motion measured prior to treatment based on 4-dimensional computer tomography (4DCT) reflects actual tumor motion during beam-on throughout the course of treatment. Methods and Materials: Twenty-three patients who had lung stereotactic body radiation therapy (SBRT) treatments were retrospectively selected. All patients had 4DCT simulation for treatment planning, from which tumor motion ranges were measured (R4DCT). Tumor motion was monitored during treatment using megavoltage (MV) imaging. Tumor motion trajectories were extracted from cine MV images and were used to determine mean and maximum tumor motion range (Mean RMV, Max RMV) throughout entire course of treatment. Comparison and correlations between mean and max RMV and R4DCT were calculated. Results: On average, an insignificant difference was found between mean RMV and R4DCT (P = .67, mean [± SD] difference = −0.7 [± 1.6] mm); meanwhile a significant difference was found between Max RMV and R4DCT (P = .03, mean [± SD] difference = 1.9 [± 1.6] mm). The difference between RMV and R4DCT was found inversely proportional to R4DCT (Y = −0.4X + 0.6, r = 0.76). Max RMV was greater than R4DCT in all patients; difference between the 2 showed no correlation with R4DCT (Y = −0.02X + 1.9, r = 0.05). Correlation between Mean RMV and R4DCT and between Max RMV and R4DCT can be expressed as Y = 0.7X (r = 0.88) and Y = 0.8X (r = 0.50), respectively. The same analysis performed on tumors that moved less than 5 mm from 4DCT revealed the following correlations: Y = 1.3X (r = 0.83) and Y = 1.7X (r = 0.49). Conclusions: Tumor motion measured from 4DCT approximates the overall average tumor motion range, but consistently underestimates the overall maximum tumor motion range. These findings may lead to a potential strategy for managing uncertainties of 4DCT in the application of lung SBRT. © 2014 American Society for Radiation Oncology. Published by Elsevier Inc. All rights reserved.
Introduction Conflicts of interest: None. ⁎ Corresponding author. Department of Radiation Oncology, Duke University Medical Center, Box 3295, Durham, NC 27710. E-mail address: [email protected] (J. Cai).
Respiratory motion poses a great challenge for precise tumor localization and dose delivery in radiation therapy. 1 Motion management is critical in radiation therapy (RT) for treating moving targets, particularly in stereotactic
1879-8500/$ – see front matter © 2014 American Society for Radiation Oncology. Published by Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.prro.2013.02.009
e60
F. Zhang et al
Practical Radiation Oncology: January-February 2014
body radiation therapy (SBRT) where the radiation dose distribution is highly conformal to the target volume and dose falls off sharply outside the target volume. 2,3 In the clinical process of RT, respiratory motion measurement is primarily performed during simulation to determine tumor motion range and internal target volume (ITV), and during treatment to verify tumor motion within preset margins or to guide the treatment. Four-dimensional CT (4DCT) is the current clinical standard for imaging tumor respiratory motion, especially for tumors in the thorax. 1,4 While 4DCT has been widely adapted in the clinic, it is not without limitations. The main limitation of 4DCT lies in the fact that it produces only 1 effective breathing cycle of data although it is acquired over several breathing cycles. Because real patients’ breathing can be very irregular and the breathing pattern may change from day to day, tumor motion measurement based on 1 breathing cycle of 4DCT cannot fully characterize the tumor motion. Using tumor motion measured from 4DCT for lung SBRT treatment may thus introduce uncertainties in target volume delineation and patient positioning. Understanding and reducing these uncertainties can potentially lead to further
Table 1 Summary of patient characteristics and results of tumor motion measurements Patient Age Gender GTV Tumor R4DCT Mean Max (y) (cm 3) location (mm) RMV RMV (mm) (mm) 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23
69 79 80 70 75 77 72 83 55 75 57 57 70 47 41 63 80 65 78 72 84 59 78
M M M F F F M M M F F M F M M M M M M M M F F
7.5 9.5 30.9 20.2 16.9 21.4 2.6 9.2 4.3 10.3 7.1 2.9 3.9 9.5 5.4 2.0 2.9 7.8 4.6 6.0 2.6 8.5 6.0
MR LR UL UL ML UL MR UL UR UL UL UR UR LR UR UR ML ML LR UR MR UL MR
0.5 7.7 1.2 0.7 1.1 9.1 6.5 2.3 2.1 0.2 5.1 8.0 1.9 4.8 3.3 2.0 1.0 3.2 13.5 5.4 1.1 0.8 4.7
0.5 7.8 1.1 1.0 2.2 5.9 5.2 2.2 3.9 0.6 4.3 5.0 2.1 3.8 1.6 4.4 1.4 1.7 9.5 3.1 0.5 0.4 2.8
1.4 11.9 2.0 2.9 4.1 11.2 9.5 5.2 7.0 1.5 5.3 10.0 5.1 7.1 4.4 7.8 5.2 5.6 17.6 9.2 2.3 2.2 5.8
4DCT, 4-dimensional computed tomography; GTV, gross tumor volume; LR, lower right lung lobe; ML, middle left lung lobe; MR, middle right lung lobe; R4DCT, tumor motion range measured from 4DCT; RMV, tumor motion range measured from cine MV images; UL, upper left lung lobe; UR, upper right lung lobe.
improvement in the efficacy of lung SBRT. However, evaluation and management of these uncertainties are challenging partly due to technical limitations to image tumor motion during the treatment. Previous studies have analyzed tumor motion uncertainties based on information acquired with 4DCT, 5 magnetic resonance imaging, 6 fluoroscopy, 7 etc, before or after the treatment. Implanted markers have been used to improve tracking accuracy, but shown to have risk of pneumothorax. 8-10 Recently, a few studies have shown the feasibility of tracking tumor motion during beam-on with cine megavoltage (MV) imaging using electronic portal imaging devices (EPID). 11,12 One fundamental question regarding the uncertainty of 4DCT lies in how accurately tumor motion measured with 4DCT before the treatment reflects actual tumor motion during beam-on throughout the entire course of lung SBRT treatment. It is of clinical interest to understand whether there is a significant difference in tumor motion measurements between CT simulation and treatment, and whether the difference (if it exists) is random or systematic? It is therefore the aim of this study to answer these questions by statistically comparing tumor motion measured with 4DCT during CT simulation with those measured with cine MV during beam-on throughout the entire lung SBRT treatment. We will only measure tumor motion during beam-on because tumor motion during beam-off, such as between different beams and between different treatment fractions, does not contribute to the actually delivered dose to the tumor.
Methods and materials Patients and lung SBRT treatment Twenty-three lung cancer patients (15 male, 8 female, mean age 68.9 years) who received SBRT treatments in our institution were selected for this study. Table 1 summarizes the characteristics of the patients. All patients underwent 4DCT simulation on a 4-slice CT scanner (Lightspeed Plus 4; GE Healthcare, Waukesha, WI) with slice thickness of 2.5 mm. The real-time position management system (Varian Medical System, Palo Alto, CA) was used to monitor patient’s breathing for retrospective sorting. All patients were imaged in an immobilization device during free breathing. All treatment plans were created using a 3D conformal technique with 611 coplanar beams (no couch kick) and wedges (if needed) with an Eclipse treatment planning system (Varian Medical System). Analytical anisotropy algorithm with heterogeneity correction was used for dose calculation. A Novalis Tx machine (Varian Medical System), equipped with kilovoltage on-board imaging, cone-beam CT (CBCT), and MV EPID was used for lung SBRT treatment. Before the treatment, patients were first
Practical Radiation Oncology: January-February 2014
Uncertainties in 4DCT tumor motion assessment
e61
Figure 1 Example of continuously acquired cine megavolt (MV) images illustrating the motion of lung tumor. The template image is indicated by the rectangular box. Tumor is indicated by the free-hand contour.
positioned using lasers, then kV orthogonal on-board imaging, and CBCT by matching the tumor. The CBCT was acquired using half-rotation (199.2 degrees) with a gantry rotation speed of 6 degrees per second, resulting in 333 projections per scan. Tumor position and motion was then verified with kV fluoroscopy immediately before the treatment and with cine MV images during the treatment (frame rate = 1 frame/second). Beam-on time for each beam ranged from about 20 seconds to 80 seconds, depending upon the prescription dose and wedges that were used.
Data analysis For each patient, the 10-phase 4DCT images and cine MV images of all beams from all fractions were retrospectively collected and analyzed. For cine MV images, tumor motion trajectories were extracted using an automatic motion tracking algorithm. Cine MV images acquired using EPID from exiting treatment beams provide a beams-eye view of the internal lung tumor motion in 2 dimensions. It has been shown that real-time tumor motion during beam-on can be quantified using cine MV images. 11-13 In this study, we used an automatic tumor motion tracking algorithm to determine the tumor
motion trajectories from cine MV images. Details of the tracking algorithm and its validation have been described in a previous study 14 and will only be summarized here. Figure 1 illustrates a series of cine MV images acquired during the lung SBRT treatment, where the tumor is readily seen in each frame. To extract the tumor motion trajectory, a template image (white box in Fig 1) encompassing the entire tumor was first created manually from the first frame of the cine MV images. Tumor position in the following frames was then determined by searching for the maximal cross-correlative region to match the template image. To qualitatively evaluate whether the algorithm correctly tracked the tumor motion, the actual tumor positions in the cine MV images were visually compared against the tumor contour (red contour in Fig 1) that moves with the calculated shifts in a timeresolved fashion. If apparent (visually detectable) discrepancies between the actual tumor position and the tumor contour were observed, the tracking results (tumor motion trajectories) were considered “un-usable” and were discarded. If no apparent discrepancies were found, the tracking results were considered as “usable” for further analysis. The accuracy of the “usable” tracking results has been investigated previously and was found to be within a millimeter. 10 The validation of tracking algorithm for
Figure 2 Example of the superior-inferior tumor motion trajectories extracted from cine megavoltage (MV) images of a single beam in 5 fractions.
e62
F. Zhang et al
Figure 3 Example of a tumor motion probability density distribution (PDF) derived from cine megavoltage (MV) images and the definition of Max RMV. Original PDF is shown in dark blue bars and the smoothed PDF is shown in red curve. Light blue lines are the 0.5% cutoff probability threshold that is used to calculate Max RMV.
motion measurement using different imaging modality (cine MRI and cine MV) has been demonstrated in our previous publication. 15 Both vertical and horizontal tumor motions were extracted from the cine MV images, but only the vertical motion was analyzed. Because all beams were coplanar, vertical motion in cine MV was essentially the superiorinferior (SI) motion. Figure 2 shows an example of the SI tumor motion trajectories of the same beam from 5 treatment fractions, in which intrafractional and interfractional variations of tumor motion were clearly seen. Mean displacement of each tumor motion trajectory was set to zero. This is so that the pretreatment patient positioning
Practical Radiation Oncology: January-February 2014
matched the mean tumor position between the CBCT and planning CT. An overall mean and max tumor motion range throughout the entire course of treatment, labeled as Mean RMV and Max RMV, respectively, was determined from cine MV images for each patient using all “usable” tumor motion trajectories from all treatment fractions. Mean RMV was simply calculated as the average distance between all respiratory peaks and respiratory valleys at immediate neighbors. To determine Max RMV, an overall tumor motion probability density distribution (PDF) was first generated for each patient using all motion trajectories of the patient that were derived from cine MV images of all beams from all fractions. The resultant tumor motion PDF reveals the probability of finding the tumor at a specific location relative to the mean tumor position. It should be noted that an occasional deep inspiration or cough may result in significantly large but uncharacteristic tumor displacements. The dosimetric effects of these extremely low probability incidences are expected to be negligible. To avoid such uncertainty and determine an effective Max RMV, a cutoff probability threshold of 0.5% was first applied to the tumor motion PDF, prior to the calculation of Max RMV. Figure 3 illustrates an example of the overall tumor motion PDF and the determination of the effective Max RMV. The cutoff threshold is an empirical value purely based on our clinical experience. The purpose of applying the threshold is to remove the effect of sudden movement of the tumor, such as cough, on the measurement of the tumor motion range. For simplicity and convenience, Max RMV was still used instead of effective Max RMV throughout this article. For 4DCT, tumor motion trajectories were automatically determined using the same cross-correlation algorithm as was used for cine MV images. Tumor motion range in the SI direction was calculated as the distance
Figure 4 Differences between Mean RMV and R4DCT (A) and between Max RMV and R4DCT (B) as a function of R4DCT. 4DCT, 4dimensional computed tomography; MV, megavoltage.
Practical Radiation Oncology: January-February 2014
Uncertainties in 4DCT tumor motion assessment
e63
Figure 5 Correlations between Mean RMV and R4DCT (A) and between Max RMV and R4DCT (B). Solid lines are for all patients; dotted lines are for patients whose tumor motion ranges are less than 5 mm based on 4DCT. 4DCT, 4-dimensional computed tomography; MV, megavoltage.
between the most superior and the most inferior tumor positions, labeled as R4DCT. Tumor motion ranges measured on cine MV (Mean RMV, Max RMV) were compared with those measured on 4DCT (R4DCT) using a paired signed rank test with a significance level of 0.05. Correlations between Mean and Max RMV and R4DCT were studied for all patients and for patients whose tumors moved less than 5 mm on the 4DCT.
Results Table 1 summarizes the tumor motion measurements for all patients. On average of all patients, insignificant differences were found between Mean RMV and R4DCT (P = .67, mean [± SD] difference = −0.7 [± 1.6] mm). The difference between RMV and R4DCT was found inversely proportional to R4DCT (Y = −0.4X + 0.6, r = 0.76), as shown in Fig 4A. Significant differences were found between Max RMV and R4DCT (P = .03, mean [± SD] difference = 1.9 [± 1.6] mm). Max RMV was greater than R4DCT in all patients; the difference between the 2 showed no correlation with R4DCT (Y = −0.02X + 1.9, r = 0.05), as shown in Fig 4B. Figure 5 shows the correlation between Mean RMV and R4DCT and between Max RMV and R4DCT, which can be expressed as Y = 0.7X (r = 0.88) and Y = 0.8X (r = 0.50), respectively. These results indicate that in general Mean RMV tends to be smaller than R4DCT, while Max RMV tends to be greater than R4DCT. The same analysis performed on tumors that moved less than 5 mm on 4DCT revealed the following correlations: Y = 1.3X (r = 0.83) and Y = 1.7X (r = 0.49), implying both Mean and Max RMV tend to be greater than R4DCT when R4DCT is less than 5 mm.
Discussion In this study, we investigated the correlation between tumor motion measured from 4DCT at CT simulation and tumor motion measured from cine MV images during beam-on throughout the entire course of lung SBRT treatment. In general the tumor motion range measured from 4DCT does estimate the mean tumor motion range exhibited throughout the course of treatment. Differences between the 2 exist, but are not significant. However, it was found that the tumor motion range measured on 4DCT was consistently smaller than the maximum tumor motion range measured throughout the course of treatment across all patients. This systematic difference between the 2 is presumably due to patient breathing variations, which is expected and confirmed with our study via statistical analysis of tumor motion throughout the course of treatment. In addition, this study quantitatively revealed the relationship between the 2, as illustrated in Fig 5B. Interestingly, it was found that this systematic difference is more substantial for tumors that move small amplitudes versus those that move large amplitudes. As a result, this systematic difference is rather persistent, around 1.9 mm, and is independent of the tumor motion range measured from 4DCT. While it is well known that tumor motion exhibits large intrafractional and interfractional variations, 16 there is still lack of proper strategies to manage these variations. There have been a few studies conducted to investigate the clinical impact of these variations, especially in the application of lung SBRT, and to explore potential solutions. For example, Cai et al 6 and Simon et al 17 demonstrated in separate studies that 4DCT can lead to errors in delineation of tumor ITV and tumor motion PDF due to breathing variations. Ge et al 18 in a recent
e64
F. Zhang et al
publication proposed to minimize these uncertainties in determining ITV for lung SBRT by combining information from 3DCT and 4DCT. In this study, we quantitatively evaluated the relationships in tumor motion between 4DCT and cine MV images. The results provide additional, and more importantly, statistical information about tumor motion variations during lung SBRT, which may potentially aid the development of a simple yet strategic method to manage these uncertainties during the application of lung SBRT. Tumor respiratory motion is inherently complex. Tumor motion measurement based on only 1 effective breathing cycle extracted from 4DCT is evidently insufficient. One potential solution may include measuring statistical characteristics of tumor motion such as mean and max motion range, motion PDF, and motion variability, to provide detailed information about the tumor motion. It is believed that these statistical characteristics are more reproducible from day to day than a single measurement of tumor motion range. In fact, our previous studies have shown that tumor motion PDF maintains high reproducibility between fractions. 19 Tumor motion in only the SI direction was investigated in this study. Tumor motion in the anterior-posterior (AP) and medial-lateral (ML) directions can be nontrivial for some patients, especially in the AP direction. However, imaging AP and right-left tumor motion using cine MV requires beam angle to be 90 degrees/270 degrees and 0 degrees/180 degrees, respectively. Due to various clinical considerations, it is not guaranteed that these beams can be used in treatment planning. As a result, we cannot obtain enough “usable” cine MV images to extract the AP and ML tumor motion trajectories. However, the differences between R4DCT and RMV in the SI direction as observed in this study were presumably due to patient’s intrafractional and interfractional breathing variations; its effects can be similarly applied to tumor motion in the AP and ML directions. Therefore, we expect a similar trend, though not necessarily of the same magnitude, of correlations and differences to exist between R4DCT and RMV in the AP and ML directions to that observed in the SI direction. The cine MV images were measured at a frequency of 1 frame/second in this study. The temporal resolution is therefore relatively low for accurately measuring tumor respiratory motion. It was set clinically this way to prevent the system from slowing down or crashing due to image overloading. This suboptimal frequency may affect the accuracy of the motion trajectories and the tumor motion PDF. However, its impact is expected to be small because of the following: (1) this low frequency can be largely compensated by using a large number of data points. In this study the PDF was typically calculated using 12 tumor motion trajectories (3 “usable” beams per fraction and 4 total fractions), which correspond to approximately a total of 12 minutes of image acquisition time and a total of 720 data points; and (2) our previous study 12 showed that the
Practical Radiation Oncology: January-February 2014
accuracy of the PDF exponentially increases as the image acquisition rate increases, but reaches a plateau when the frame rate is greater than 2 frames/second. Therefore, the estimated error in the PDF calculation at a frame rate of 1 frame/second is approximately 1.5%. For future studies, especially to study tumor motion variability based on cine MV, a higher frame rate (≥ 2 frames/second) is desirable. This is clinically feasible because most systems can set the frame rate up to 8 frames/second.
Conclusions We systematically compared and quantitatively correlated tumor motion measured from 4DCT prior to treatment with tumor motion measured from cine MV images during the course of lung SBRT treatment. We found that tumor motion measured from 4DCT approximates the overall average tumor motion range, but consistently underestimates the overall maximum tumor motion range exhibited during treatment. Findings of our study may lead to a potential strategy for managing uncertainties of 4DCT in the application of lung SBRT.
Acknowledgment The authors thank Irina Vergalasova for editing the manuscript and Jing Hu, BS, CMD, for analyzing part of the cine MV data.
References 1. Mageras GS, Pevsner A, Yorke ED, et al. Measurement of lung tumor motion using respiration-correlated CT. Int J Radiat Oncol Biol Phys. 2004;60:933-941. 2. Nagata Y, Negoro Y, Aoki T, et al. Clinical outcomes of 3D conformal hypofractionated single high-dose radiotherapy for one or two lung tumors using a stereotactic body frame. Int J Radiat Oncol Biol Phys. 2002;52:1041-1046. 3. Onishi H, Araki T, Shirato H, et al. Stereotactic hypofractionated high-dose irradiation for stage I nonsmall cell lung carcinoma: clinical outcomes in 245 subjects in a Japanese multiinstitutional study. Cancer. 2004;101:1623-1631. 4. Keall PJ, Mageras GS, Balter JM, et al. The management of respiratory motion in radiation oncology report of AAPM Task Group 76. Med Phys. 2006;33:3874-3900. 5. Redmond KJ, Song DY, Fox JL, Zhou J, Rosenzweig CN, Ford E. Respiratory motion changes of lung tumor over the course of radiation therapy based on respiration-correlated four dimension computed tomography scans. Int J Radiat Oncol Biol Phys. 2009;75: 1605-1612. 6. Cai J, Read PW, Sheng K. The effect of respiratory motion variability and tumor size on the accuracy of average intensity projection from four-dimensional computed tomography: an investigation based on dynamic MRI. Med Phys. 2008;35:4974-4981. 7. Kupelian PA, Forbes A, Willoughby TR, et al. Implantation and stability of metallic fiducials within pulmonary lesions. Int J Radiat Oncol Biol Phys. 2007;69:777-785.
Practical Radiation Oncology: January-February 2014 8. Shirato H, Harada T, Harabayashi T, et al. Feasibility of insertion/implantation of 2.0-mm-diameter gold internal fiducial markers for precise setup and real-time tumor tracking in radiotherapy. Int J Radiat Oncol Biol Phys. 2003;56:240-247. 9. Roman NO, Shepherd W, Mukhopadhyay N, et al. Interfractional positional variability of fiducial markers and primary tumors in locally advanced non-small-cell lung cancer during audiovisual biofeedback radiotherapy. Int J Radiat Oncol Biol Phys. 2012;83: 1566-1572. 10. Imura M, Yamazaki K, Shirato H, et al. Insertion and fixation of fiducial markers for setup and tracking of lung tumors in radiotherapy. Int J Radiat Oncol Biol Phys. 2005;63:1442-1447. 11. Meyer J, Richter A, Baier K, Wilbert J, Guckenberger M, Flentje M. Tracking moving objects with megavoltage portal imaging: a feasibility study. Med Phys. 2006;33:1275-1280. 12. Tai A, Christensen JD, Gore E, Khamene A, Boettger T, Li XA. Gated treatment delivery verification with on-line megavoltage fluoroscopy. Int J Radiat Oncol Biol Phys. 2010;76:1592-1598. 13. Richter A, Wilbert J, Baier K, Flentje M, Guckenberger M. Feasibility study for markerless tracking of lung tumors in stereotactic body radiotherapy. Int J Radiat Oncol Biol Phys. 2010;78:618-627.
Uncertainties in 4DCT tumor motion assessment
e65
14. Zhang F, Hu J, Kelsey CR, et al. Reproducibility of tumor motion probability distribution function in stereotactic body radiation therapy of lung cancer. Int J Radiat Oncol Biol Phys. 2012;84: 861-866. 15. Chang Z, Cai J, Yin FF. Evaluation of motion measurement using cine MRI for image guided stereotactic body radiotherapy on a new phantom platform. J Radiosurgery and SBRT. 2011:109-115. 16. Benedict SH, Yenice KM, Followill D, et al. Stereotactic body radiation therapy: the report of AAPM Task Group 101. Med Phys. 2010;37:4078-4101. 17. Simon L, Giraud P, Servois V, Rosenwald JC. Initial evaluation of a four-dimensional computed tomography system using a programmable motor. J Appl Clin Med Phys. 2006;7:50-65. 18. Ge H, Cai J, Kelsey C, et al. Quantification and minimization of uncertainties of internal target volume (ITV) for stereotactic-body radiation therapy (SBRT) of lung cancer. Int J Radiat Oncol Biol Phys. 2013;85:438-443. 19. Cai J, Read PW, Altes TA, Molloy JA, Brookeman JR, Sheng K. Evaluation of the reproducibility of lung motion probability distribution function (PDF) using dynamic MRI. Phys Med Biol. 2007;52:365-373.
