- Email: [email protected]
Action Levels on Dose and Anatomic Variation for Adaptive Radiation Therapy Using Daily Offline Plan Evaluation: Preliminary Results
Practical Radiation Oncology (2018) xx, 1-6
www.practicalradonc.org
Basic Original Report
Action Levels on Dose and Anatomic Variation for Adaptive Radiation Therapy Using Daily Offline Plan Evaluation: Preliminary Results Baoshe Zhang PhD*, Sung-Woo Lee PhD, Shifeng Chen PhD, Jinghao Zhou PhD, Karl Prado PhD, Warren D’Souza PhD, Byongyong Yi PhD Department of Radiation Oncology, University of Maryland School of Medicine, Baltimore, Maryland Received 3 January 2018; revised 5 July 2018; accepted 9 August 2018
Abstract Purpose: This study aimed to develop action levels for replanning to accommodate dosimetric variations resulting from anatomic changes during the course of treatments, using daily cone beam computed tomography (CBCT). Methods and materials: Daily or weekly CBCT images of 20 patients (10 head and neck, 5 lung, and 5 prostate cancers) who underwent resimulation per physicians’ clinical decisions, mainly from the comparison of CBCT scans, were used to determine action levels. The first CBCT image acquired before the first treatment was used as the reference image to rule out effects of dose inaccuracy from the CBCT. The Pearson correlation of clinical target volume (CTV) was used as a parameter of anatomic variation. Parameters for action levels on dose and anatomic variation were deduced by comparing the parameters and clinical decisions made for replanning. A software tool was developed to automatically perform all procedures, including dose calculations, using the CBCT and plan evaluations. Results: Replans were clinically decided based on either significant dose or anatomic changes in 13 cases. The 7 cases that did not require replanning showed dose differences <5%, and the Pearson correlation of the CTV was >75% for all fractions. A difference in planning target volume dose >5% or a difference in the image correlation coefficient of the CTV <0.75 proved to be indicators for replanning. Once the results of the CBCT plan met the replanning criteria, the software tool automatically alerted the attending physician and physicist by both e-mail and pager so that the case could be examined closely. Conclusions: Our study shows that a dose difference of 5% and/or anatomy variation at 0.75 Pearson correlations are practical action levels on dose and anatomic variation for replanning for the given data sets. Ó 2015 American Society for Radiation Oncology. Published by Elsevier Inc. All rights reserved.
Sources of support: This work had no specific funding. Conflicts of interest: The authors have no conflicts of interest to disclose. * Corresponding author. Department of Radiation Oncology, University of Maryland School of Medicine, Baltimore, MD 21043. E-mail address: [email protected] (B. Zhang). https://doi.org/10.1016/j.prro.2018.08.006 1879-8500/Ó 2018 American Society for Radiation Oncology. Published by Elsevier Inc. All rights reserved.
2
Practical Radiation Oncology: --- 2018
B. Zhang et al
Introduction One fundamental assumption about patient setup in radiation therapy is that the initial simulation represents a target position that will be reproducible throughout daily treatments. Based on this assumption, image guided radiation therapy (IGRT) is used to verify whether the patient setup position at the time of each treatment reproduces that at the time of simulation. Radiation therapy is often administered in 20 to 40 daily fractions over 4 to 8 weeks, with the exception of nonconventional radiation therapy techniques, such as palliative radiation therapy or hypofractionated stereotactic radiation therapy. During such a lengthy treatment course, patients may experience weight change and/or tumors may shrink or grow as a result of tumor cell death or disease progression, respectively. If treatment continues with the original treatment plan despite such changes, the delivered dose will not be the same as the expected dose, which may result in underdosing the target or overdosing critical structures. Adapting treatment plans to effectively achieve the treatment goals when such variations are significant is imperative. Resimulation and replanning are crucial clinical decisions, although often quite time-consuming. Changes in a patient’s tumor and weight are usually gradual processes that can be monitored through daily cone beam computed tomography (CBCT) imaging. The concept of adaptive radiation therapy (ART) was introduce by Yan et al1 in 1997, and since then, the need for a feedback control strategy for patient variation through the adaptive adjustment of treatment planning or treatment delivery has been widely recognized. ART may be achieved either offline,2 with online adaptation and replanning,3,4 or by adaptively adjusting beam delivery.5,6 Although daily imaging techniques were initially developed for daily patient setup, these are now widely used to implement ART.7e11 Despite many technologies and various ways of performing ART, authors have not found reports on the clinical action levels for resimulation using dose and anatomic variations, which are critical in the treatment decision-making process. The aim of our study was to develop action levels for dose and anatomic variations (ALDAVs), using CBCT images acquired during daily IGRT that quantitatively predict when to resimulate the patient and then replan. To rule out the effect of dosimetric uncertainties in CBCT-based planning, the dose distribution of the first CBCT images is regarded as the baseline, and subsequent CBCT images are compared with that baseline. Once differences exceed a preset level, the system will flag for
potential resimulation and replanning. Although CBCT images suffer from image truncation and CT number inaccuracy, use of the same CBCT scanner, the same imaging acquisition protocol, and the same image reconstruction algorithms will factor out these adverse effects. Moreover, several studies have shown that using site-specific CT number-to-density tables instead of a conventional standard CT-to-density table can improve the accuracy of the dose directly calculated from the CBCT images to approximately 2%.12e18
Methods and Materials Consistencies in both dosimetric and geometric (anatomic) characteristics of CBCT images acquired during treatment were used to implement an automatic computation framework to predict when to resimulate the patient and then replan. The first CBCT image acquired before the first treatment was used as the reference image or as the ground truth for patient treatment positioning. The subsequent CBCT images were geometrically and dosimetrically compared against the first CBCT image. An image correlation coefficient of the clinical target volume (CTV), described in the following, was used to determine the geometric change of the tumor, and the dose distribution on CBCT images was used for the dosimetric change check. For lung cases, the CTV based on IGTV is used. Clinically determined isocenters of the CBCT scans are used in calculating the doses and the CTV comparisons. The CTV is rigidly registered to the CBCT of the treatment day. To avoid CBCT image quality disparity, a prerequisite was that CBCT images must be acquired by the same imaging protocol (eg, mAs, kVp, and filtration) and the same CBCT scanner model and reconstructed by the same algorithm. All the CBCT images were acquired with a half-fan filter, 3mm slice thickness, and 16-cm scan length. The RayStation V6 (RaySearch) planning system with collapse cone algorithm was used for dose calculations, and 3 types of Varian linear accelerators (TrueBeam, Trilogy and IX, Varian, Palo Alto, CA) and their on board imaging systems were used in treating and imaging the patients.
Image correlation coefficient The Pearson correlation coefficient (PCC)19 was used to determine changes in the CTV. A 3-dimensional correlation convolution between 2 images, f and g, was defined by extending the 2-dimensional PCC as follows:
PPP x y z f ðx þ i; y þ j; z þ kÞ f ½gðx; y; zÞ g rijk Z qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi PPP PPP 2 x y z f ðx; y; zÞ f x y z ½gðx; y; zÞ g
Practical Radiation Oncology: --- 2018
where f(x,y,z) and g(x,y,z) represent the intensity (e.g., CT number or Hounsfield unit) at coordinate x, y, z; i,j,k are relative position shifts from x,y,z; and f and g are the mean values of the intensity matrices f and g, respectively. In fact, if 2 images are not registered properly (e.g., as a result of improper online treatment registration), this formula can be used as a cost function or part of a cost function to find the correct image registration and/or evaluate the setup error.
Automatic dose calculation using CBCT A total of 222 sets of daily or weekly CBCT images from 20 patients (10 head and neck, 5 advanced lung, and 5 prostate cancers) who underwent resimulation were analyzed to determine variations in anatomy and dose. All patients were immobilized per the University of Maryland Practice Guidelines (not published), both for simulations and treatments. Either intensity modulated radiation therapy or volumetric modulated arc therapy is used as the treatment delivery technique. The decision to resimulate was made by the patient’s physicians, based mainly on visual comparisons of the CBCT scans and the planning CT scans. CBCT images of daily setups were used to calculate the dose distribution and evaluate anatomic changes. The clinical treatment plan was copied from the planning CT to the CBCT images through the rigid registrations of clinical setups. A predetermined and disease siteespecific
Action levels for ART
3
CT density table for the CBCT was used for dose calculation. To rule out uncertainties of dose calculations and pixel values, variations between the first CBCT and the CBCT of the day were compared.
Structure of the system and clinical process An offline strategy calculated the dose distribution of CTV or gross tumor volume using the acquired CBCT images and then was compared with those with the reference CBCT images (ie, first CBCT image) at a later time. Contours of the structures, including the targets and organs at risk (OARs), were moved to the CBCT, and the beams were also copied from the planning CT to the corresponding CBCT accordingly. Because of the limited field of view of the CBCT image, some structures may remain outside the CBCT. The truncated portions of structures were not considered for geometric comparison or dosimetric comparison. The planning CT and the CBCT from the treatment day were registered with rigid image registration that was centered to the isocenter of the treatment day, as determined by therapists. Our clinical implementation was composed of a data processing pipeline and a database. The data processing pipeline (Fig 1) consisted of 4 major modules: the CBCTcollecting module, the plan registration module, the dose calculation module, and the evaluation and action module. The data pipeline was executed periodically without human intervention. The CBCT-collecting module was
Figure 1 System diagram of our clinical implementation: cone beam computed tomographyecollecting module, plan registration module, dose calculation module, evaluation and action module.
4
B. Zhang et al
triggered periodically to use a standard Digital Imaging and Communications in Medicine (DICOM)20 query to search for newly acquired CBCT DICOM images of the patient of interest and then retrieve the CBCT images and related image registration DICOM files and save them to a database. The plan registration module imported these DICOM CBCT images and registrations into the treatment planning system and associated them with the corresponding treatment plan and images. A site-specific CTto-density table was then assigned to these CBCT images. Through the registration matrix, the treatment plan was copied along with the contours to these CBCT images. The dose calculation module calculated the dose distribution from the copied treatment plan directly onto these CBCT images and the PCC between the first and current CBCT image sets. Variations of doseevolume parameters (eg, D95, D90, D10, and D5) were also compared. The evaluation and action module analyzed trends in the PCC and dose-volume histogram variations to determine whether the patient’s radiation oncologist and radiation physicists should more closely review these changes. To avoid uncertainties caused by patient setup, the system also has a function to shift the isocenter position within a cube of 1 1 1 cm3 around the treatment isocenter to find the highest PCC and the lowest dosimetric indices differences.
Results Figure 2 shows the distribution of variations of dose and the PCC using the CBCT scans taken at the time of resimulation requested by clinicians. Among 20 patients, 7 patients (3 head and neck, 2 prostate, and 2 lung cancers; green solid symbols) were determined not to need the new plan because of little difference in dose (<2%) between the original plan and the same plan on the new CT (no-new plan-group, no-action group). For the other 13 cases (dashed or dotted symbols), however, the treating physicians decided to use the new plan using the new CT after comparing the dose distribution of the old plan with that of the new CT (new-plan group, action group). As shown in Figure 2, all cases in the action group are in the region of a PCC <0.75 or dose change >5%, and all cases in the no-action group are in the region of a PCC >0.75 and dose change <5%. This suggests that action levels at which to consider re-CT for replanning are PCC <0.75 or dose change >5% (P < .001; c2 analysis). Here, Dxx,n is the CBCT’s Dxx (e.g., D95) of the n-th treatment fraction. Dose-volume histogram variation is defined as (Dxx,n/Dxx,1 e 1). Two typical cases of deviation are presented here as an illustration: a head and neck case (Fig 3) and a lung case (Fig 4). For the head and neck case, daily CBCT scans were acquired for the first 3 treatment days, followed by weekly CBCT scans, but daily CBCT scans were acquired
Practical Radiation Oncology: --- 2018
Figure 2 Maximum value of dose-volume histogram variations among D95, D90, D10, and D5 versus image correlation coefficient in 20 patients (5 with prostate cancer [A], 5 with lung cancer [C], and 10 with head and neck cancer [:]) at the time that replanning was requested clinically. Solid green symbols represent patients who did not need replanning; orange striped symbols represent patients whose predicted replanning was earlier than the actual clinical replanning; and purple checked symbols represent patients whose predicted replanning agreed with actual clinical replanning. (A color version of this figure is available at https://doi.org/10.1016/j.prro.2018.08.006.)
for the lung case. For the head and neck case, in 2 weeks, the PCC for the CTV decreased to 0.77, and the D10 and D5 of CTV experienced approximately a 20% change. The patient had lost about 4.5 kg of weight, and replanning was ordered by a treating physician. For the lung case, the changes in D95, D90, D10, and D5 were <5%, but the PCC fell below 0.65. This case went to replanning as a result of significant tumor shrinkage. The entire procedure, including data processing pipeline, evaluation, and alert reporting, is fully automated. On average, it takes approximately 10 minutes to compute
Figure 3 Temporal variation of indices of a patient with head and neck cancer. Solid line Z PCC; dotted lines Z dose-volume histogram variation of cone beam computed tomography planning target volume doses (D95, D90, D10, and D5) of the n-th fraction versus those of the first fraction.
Practical Radiation Oncology: --- 2018
Figure 4 Temporal variation of indices of a lung patient. Solid line Z Pearson’s correlation coefficient; dotted lines Z dosevolume histogram variation of cone beam computed tomography planning target volume doses (D95, D90, D10, and D5) of the n-th fraction versus those of the first fraction.
the dose distribution of a volumetric arc therapy plan on a CBCT image set and another 5 minutes to find the highest PCC around the treatment isocenter between the first and the newly acquired CBCT images. In our clinic, approximately 60 to 70 CBCT image sets are acquired per day on average. Retrieval of all CBCT DICOM image sets from the Varian Aria oncology information system takes approximately 30 minutes, and <2 hours is needed to process these tasks using parallel computation.
Discussion CBCT gained its popularity as an adaptive planning modality not only because of its geometric alignment capabilities but also because of its potential dosimetric evaluation functionality. As an IGRT tool, CBCT provides the capacity to capture such changes regularly during the treatment course in the form of serial snapshots and is a means of verifying accurate and precise radiation delivery.21 Once these changes exceed the predefined tolerance levels, physicians and physicists can more carefully review those CBCT images and evaluate changes to decide whether resimulation and replanning will be beneficial to the patient’s care. Schaly et al reported a threshold as the gamma criteria of 3-mm distance to agreement and 30 Hounsfield unit differences.22 This method provides 82% to 100% sensitivity and a 30% false-positive rate. This method only considers the anatomic change of the body. Hvid et al suggested daily dose surveillance of organs at risk, using CBCT images of head and neck cases.23 No action levels have been suggested. The method described in this paper uses a standard DICOM data stream to create an efficient data pipeline
Action levels for ART
5
from raw CBCT images to calculated geometric and dosimetric metrics. This system not only compares a patient’s anatomic changes based on CBCT images but also analyzes the dosimetric change trend based on CBCT scans starting from the first day of treatment. As a result, the system is able to provide radiation oncologists with accurate and concise information to determine whether resimulation and replanning are indicated and whether any treatment parameter should be adjusted accordingly. Furthermore, the system is configurable; once a change exceeds its predefined action levels, physicians and physicists will be automatically alerted through pagers and emails to review the patient’s treatment and evaluate the treatment plan. One big question in resimulation and replanning is how to determine the threshold level at which action is required. Attending physicians clinically determined to use replans with resimulation images for 13 patients (the action group of the 20 cases of preliminary study), but 7 patients (the no-action group) continued to use old plans. All patients in the action group demonstrated either PCCs < 0.75 or >5% in dose differences, whereas all Group B patients showed PCCs >0.75 and <5% dose differences (Fig 2). Suggested action levels of a PCC <0.75 or >5% in dose differences are reasonable criteria for replanning. The parameters of all 13 patients who ended up with new plans were analyzed for each fraction. Seven cases showed that the parameters reached the action level at or near the re-CT date, whereas 6 cases reached the action level earlier. This indicates that 6 patients might have benefitted if the decision to replan had been made earlier. This study used 20 cases of 3 anatomic sites. Because of the limited number of cases, it is not clear whether the ALDAV’s metric parameters are anatomy dependent. Furthermore, because the study is designed to use the resimulation cases only, the sensitivity of the ALDAV was not acquired. Patient daily variations, such as patient relaxation, bladder/rectal filling, and change in weight, are not taken in to account when analyzing the data; however, all patient setups of each anatomic site are per the previously commented Practice Guideline. A half fan image protocol is used to prevent image truncations. None of the CBCT images suffered from image truncations, except in inferosuperior directions. Because of this limitation, conclusions of this report may not be applied if significant anatomic changes happen outside the CBCT range, unless multiple CBCT scans are performed for every fraction. Nevertheless, this study will be a good starting point for a large-scale study with multi-institution collaboration to answer these questions. Current practice on determining re-CT is still a somewhat subjective clinical decision. This study suggests the utility of an automatic method with data-driven objective parameters. This study focused only on the target dose and the volume using rigid registration. Applying deformable image registration may further improve decision-making and would be
6
B. Zhang et al
especially useful in developing action levels on the OARs.
Conclusions An automated method to determine the need for patient replanning was designed and tested. The software tool collects the CBCT scans and automatically evaluates changes in dose and target shape. The efficiency of this method was tested retrospectively on studies from 20 patients with cancer and found to be practical. Based on our clinical experience, a PCC <0.75 or dose difference >5% has led to replanning, and based on this, we have established PCC <0.75 and dose difference >5% as action levels for consideration of replanning by the attending physicians.
Acknowledgments The authors thank Dr. Nancy Knight for critical comments and professional opinions during manuscript preparation.
References 1. Yan D, Vicini F, Wong J, Martinez A. Adaptive radiation therapy. Phys Med Biol. 1997;42:123-132. 2. Yan D, Lockman D, Brabbins D, Tyburski L, Martinez A. An offline strategy for constructing a patient-specific planning target volume in adaptive treatment process for prostate cancer. Int J Radiat Oncol Biol Phys. 2000;48:289-302. 3. Mohan R, Zhang X, Wang H, et al. Use of deformed intensity distributions for on-line modification of image-guided IMRT to account for interfractional anatomic changes. Int J Radiat Oncol Biol Phys. 2005;61:1258-1266. 4. Wu Q, Chi Y, Chen PY, Krauss DJ, Yan D, Martinez A. Adaptive replanning strategies accounting for shrinkage in head and neck IMRT. Int J Radiat Oncol Biol Phys. 2009;75:924-932. 5. Court LE, Tishler RB, Petit J, Cormack R, Chin L. Automatic online adaptive radiation therapy techniques for targets with significant shape change: A feasibility study. Phys Med Biol. 2006;51:24932501. 6. Burridge N, Amer A, Marchant T, et al. Online adaptive radiotherapy of the bladder: Small bowel irradiated-volume reduction. Int J Radiat Oncol Biol Phys. 2006;66:892-897.
Practical Radiation Oncology: --- 2018 7. Lei Y, Wu Q. A hybrid strategy of offline adaptive planning and online image guidance for prostate cancer radiotherapy. Phys Med Biol. 2010;55:2221-2234. 8. Thilmann C, Nill S, Tücking T, et al. Correction of patient positioning errors based on in-line cone beam CTs: clinical implementation and first experiences. Radiat Oncol. 2006;1:16. 9. Sorcini B, Tilikidis A. Clinical application of image-guided radiotherapy, IGRT (on the Varian OBI platform). Cancer Radiother. 2006;10:252-257. 10. Nakagawa K, Yamashita H, Shiraishi K, et al. Verification of intreatment tumor position using kilovoltage cone-beam computed tomography: A preliminary study. Int J Radiat Oncol Biol Phys. 2007;69:970-973. 11. Ding GX, Duggan DM, Coffey CW, et al. A study on adaptive IMRT treatment planning using kV cone-beam CT. Radiother Oncol. 2007;85:116-125. 12. Chen S, Le Q, Mutaf Y, et al. Feasibility of CBCT-based dose with a patient-specific stepwise HU-to-density curve to determine time of replanning. J Appl Clin Med Phys. 2017;18:64-69. 13. Guan H, Dong H. Dose calculation accuracy using cone-beam CT (CBCT) for pelvic adaptive radiotherapy. Phys Med Biol. 2009;54: 6239-6250. 14. Hatton J, McCurdy B, Greer PB. Cone beam computerized tomography: The effect of calibration of the Hounsfield unit number to electron density on dose calculation accuracy for adaptive radiation therapy. Phys Med Biol. 2009;54:N329-N346. 15. Yang Y, Schreibmann E, Li T, Wang C, Xing L. Evaluation of onboard kV cone beam CT (CBCT)-based dose calculation. Phys Med Biol. 2007;52:685-705. 16. Yoo S, Yin FF. Dosimetric feasibility of cone-beam CT-based treatment planning compared to CT-based treatment planning. Int J Radiat Oncol Biol Phys. 2006;66:1553-1561. 17. Rong Y, Smilowitz J, Tewatia D, Tome WA, Paliwal B. Dose calculation on kV cone beam CT images: An investigation of the Hu-density conversion stability and dose accuracy using the sitespecific calibration. Med Dosimetry. 2010;35:195-207. 18. Dunlop A, McQuaid D, Nill S, et al. Comparison of CT number calibration techniques for CBCT-based dose calculation. Strahlenther Onkol. 2015;191:970-978. 19. Rodgers JL, Nicewander WA. Thirteen ways to look at the correlation coefficient. Am Statistician. 1988;42:59-66. 20. NEMA PS3/ISO 12052, Digital Imaging and Communications in Medicine (DICOM) Standard: National Electrical Manufacturers Association, Rosslyn, VA, USA. Available at: http://medical.nema. org. Accessed October 1, 2017. 21. Goyal S, Kataria T. Image guidance in radiation therapy: Techniques and applications. Radiol Res Pract. 2014;2014:705604. 22. Schaly B, Kempe J, Venkatesan V, Mtichell S. Using gamma index to flag changes in anatomy during image-guided radiation therapy of head and neck cancer. J Appl Clin Med Phys. 2017;18:79-87. 23. Hvid C, Elstrom U Jensen K, Grau C. Cone-beam computed tomography (CBCT) for adaptive image guided head and neck radiation therapy. Acta Onclogica. 2018;57:552-556.
www.practicalradonc.org
Basic Original Report
Action Levels on Dose and Anatomic Variation for Adaptive Radiation Therapy Using Daily Offline Plan Evaluation: Preliminary Results Baoshe Zhang PhD*, Sung-Woo Lee PhD, Shifeng Chen PhD, Jinghao Zhou PhD, Karl Prado PhD, Warren D’Souza PhD, Byongyong Yi PhD Department of Radiation Oncology, University of Maryland School of Medicine, Baltimore, Maryland Received 3 January 2018; revised 5 July 2018; accepted 9 August 2018
Abstract Purpose: This study aimed to develop action levels for replanning to accommodate dosimetric variations resulting from anatomic changes during the course of treatments, using daily cone beam computed tomography (CBCT). Methods and materials: Daily or weekly CBCT images of 20 patients (10 head and neck, 5 lung, and 5 prostate cancers) who underwent resimulation per physicians’ clinical decisions, mainly from the comparison of CBCT scans, were used to determine action levels. The first CBCT image acquired before the first treatment was used as the reference image to rule out effects of dose inaccuracy from the CBCT. The Pearson correlation of clinical target volume (CTV) was used as a parameter of anatomic variation. Parameters for action levels on dose and anatomic variation were deduced by comparing the parameters and clinical decisions made for replanning. A software tool was developed to automatically perform all procedures, including dose calculations, using the CBCT and plan evaluations. Results: Replans were clinically decided based on either significant dose or anatomic changes in 13 cases. The 7 cases that did not require replanning showed dose differences <5%, and the Pearson correlation of the CTV was >75% for all fractions. A difference in planning target volume dose >5% or a difference in the image correlation coefficient of the CTV <0.75 proved to be indicators for replanning. Once the results of the CBCT plan met the replanning criteria, the software tool automatically alerted the attending physician and physicist by both e-mail and pager so that the case could be examined closely. Conclusions: Our study shows that a dose difference of 5% and/or anatomy variation at 0.75 Pearson correlations are practical action levels on dose and anatomic variation for replanning for the given data sets. Ó 2015 American Society for Radiation Oncology. Published by Elsevier Inc. All rights reserved.
Sources of support: This work had no specific funding. Conflicts of interest: The authors have no conflicts of interest to disclose. * Corresponding author. Department of Radiation Oncology, University of Maryland School of Medicine, Baltimore, MD 21043. E-mail address: [email protected] (B. Zhang). https://doi.org/10.1016/j.prro.2018.08.006 1879-8500/Ó 2018 American Society for Radiation Oncology. Published by Elsevier Inc. All rights reserved.
2
Practical Radiation Oncology: --- 2018
B. Zhang et al
Introduction One fundamental assumption about patient setup in radiation therapy is that the initial simulation represents a target position that will be reproducible throughout daily treatments. Based on this assumption, image guided radiation therapy (IGRT) is used to verify whether the patient setup position at the time of each treatment reproduces that at the time of simulation. Radiation therapy is often administered in 20 to 40 daily fractions over 4 to 8 weeks, with the exception of nonconventional radiation therapy techniques, such as palliative radiation therapy or hypofractionated stereotactic radiation therapy. During such a lengthy treatment course, patients may experience weight change and/or tumors may shrink or grow as a result of tumor cell death or disease progression, respectively. If treatment continues with the original treatment plan despite such changes, the delivered dose will not be the same as the expected dose, which may result in underdosing the target or overdosing critical structures. Adapting treatment plans to effectively achieve the treatment goals when such variations are significant is imperative. Resimulation and replanning are crucial clinical decisions, although often quite time-consuming. Changes in a patient’s tumor and weight are usually gradual processes that can be monitored through daily cone beam computed tomography (CBCT) imaging. The concept of adaptive radiation therapy (ART) was introduce by Yan et al1 in 1997, and since then, the need for a feedback control strategy for patient variation through the adaptive adjustment of treatment planning or treatment delivery has been widely recognized. ART may be achieved either offline,2 with online adaptation and replanning,3,4 or by adaptively adjusting beam delivery.5,6 Although daily imaging techniques were initially developed for daily patient setup, these are now widely used to implement ART.7e11 Despite many technologies and various ways of performing ART, authors have not found reports on the clinical action levels for resimulation using dose and anatomic variations, which are critical in the treatment decision-making process. The aim of our study was to develop action levels for dose and anatomic variations (ALDAVs), using CBCT images acquired during daily IGRT that quantitatively predict when to resimulate the patient and then replan. To rule out the effect of dosimetric uncertainties in CBCT-based planning, the dose distribution of the first CBCT images is regarded as the baseline, and subsequent CBCT images are compared with that baseline. Once differences exceed a preset level, the system will flag for
potential resimulation and replanning. Although CBCT images suffer from image truncation and CT number inaccuracy, use of the same CBCT scanner, the same imaging acquisition protocol, and the same image reconstruction algorithms will factor out these adverse effects. Moreover, several studies have shown that using site-specific CT number-to-density tables instead of a conventional standard CT-to-density table can improve the accuracy of the dose directly calculated from the CBCT images to approximately 2%.12e18
Methods and Materials Consistencies in both dosimetric and geometric (anatomic) characteristics of CBCT images acquired during treatment were used to implement an automatic computation framework to predict when to resimulate the patient and then replan. The first CBCT image acquired before the first treatment was used as the reference image or as the ground truth for patient treatment positioning. The subsequent CBCT images were geometrically and dosimetrically compared against the first CBCT image. An image correlation coefficient of the clinical target volume (CTV), described in the following, was used to determine the geometric change of the tumor, and the dose distribution on CBCT images was used for the dosimetric change check. For lung cases, the CTV based on IGTV is used. Clinically determined isocenters of the CBCT scans are used in calculating the doses and the CTV comparisons. The CTV is rigidly registered to the CBCT of the treatment day. To avoid CBCT image quality disparity, a prerequisite was that CBCT images must be acquired by the same imaging protocol (eg, mAs, kVp, and filtration) and the same CBCT scanner model and reconstructed by the same algorithm. All the CBCT images were acquired with a half-fan filter, 3mm slice thickness, and 16-cm scan length. The RayStation V6 (RaySearch) planning system with collapse cone algorithm was used for dose calculations, and 3 types of Varian linear accelerators (TrueBeam, Trilogy and IX, Varian, Palo Alto, CA) and their on board imaging systems were used in treating and imaging the patients.
Image correlation coefficient The Pearson correlation coefficient (PCC)19 was used to determine changes in the CTV. A 3-dimensional correlation convolution between 2 images, f and g, was defined by extending the 2-dimensional PCC as follows:
PPP x y z f ðx þ i; y þ j; z þ kÞ f ½gðx; y; zÞ g rijk Z qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi PPP PPP 2 x y z f ðx; y; zÞ f x y z ½gðx; y; zÞ g
Practical Radiation Oncology: --- 2018
where f(x,y,z) and g(x,y,z) represent the intensity (e.g., CT number or Hounsfield unit) at coordinate x, y, z; i,j,k are relative position shifts from x,y,z; and f and g are the mean values of the intensity matrices f and g, respectively. In fact, if 2 images are not registered properly (e.g., as a result of improper online treatment registration), this formula can be used as a cost function or part of a cost function to find the correct image registration and/or evaluate the setup error.
Automatic dose calculation using CBCT A total of 222 sets of daily or weekly CBCT images from 20 patients (10 head and neck, 5 advanced lung, and 5 prostate cancers) who underwent resimulation were analyzed to determine variations in anatomy and dose. All patients were immobilized per the University of Maryland Practice Guidelines (not published), both for simulations and treatments. Either intensity modulated radiation therapy or volumetric modulated arc therapy is used as the treatment delivery technique. The decision to resimulate was made by the patient’s physicians, based mainly on visual comparisons of the CBCT scans and the planning CT scans. CBCT images of daily setups were used to calculate the dose distribution and evaluate anatomic changes. The clinical treatment plan was copied from the planning CT to the CBCT images through the rigid registrations of clinical setups. A predetermined and disease siteespecific
Action levels for ART
3
CT density table for the CBCT was used for dose calculation. To rule out uncertainties of dose calculations and pixel values, variations between the first CBCT and the CBCT of the day were compared.
Structure of the system and clinical process An offline strategy calculated the dose distribution of CTV or gross tumor volume using the acquired CBCT images and then was compared with those with the reference CBCT images (ie, first CBCT image) at a later time. Contours of the structures, including the targets and organs at risk (OARs), were moved to the CBCT, and the beams were also copied from the planning CT to the corresponding CBCT accordingly. Because of the limited field of view of the CBCT image, some structures may remain outside the CBCT. The truncated portions of structures were not considered for geometric comparison or dosimetric comparison. The planning CT and the CBCT from the treatment day were registered with rigid image registration that was centered to the isocenter of the treatment day, as determined by therapists. Our clinical implementation was composed of a data processing pipeline and a database. The data processing pipeline (Fig 1) consisted of 4 major modules: the CBCTcollecting module, the plan registration module, the dose calculation module, and the evaluation and action module. The data pipeline was executed periodically without human intervention. The CBCT-collecting module was
Figure 1 System diagram of our clinical implementation: cone beam computed tomographyecollecting module, plan registration module, dose calculation module, evaluation and action module.
4
B. Zhang et al
triggered periodically to use a standard Digital Imaging and Communications in Medicine (DICOM)20 query to search for newly acquired CBCT DICOM images of the patient of interest and then retrieve the CBCT images and related image registration DICOM files and save them to a database. The plan registration module imported these DICOM CBCT images and registrations into the treatment planning system and associated them with the corresponding treatment plan and images. A site-specific CTto-density table was then assigned to these CBCT images. Through the registration matrix, the treatment plan was copied along with the contours to these CBCT images. The dose calculation module calculated the dose distribution from the copied treatment plan directly onto these CBCT images and the PCC between the first and current CBCT image sets. Variations of doseevolume parameters (eg, D95, D90, D10, and D5) were also compared. The evaluation and action module analyzed trends in the PCC and dose-volume histogram variations to determine whether the patient’s radiation oncologist and radiation physicists should more closely review these changes. To avoid uncertainties caused by patient setup, the system also has a function to shift the isocenter position within a cube of 1 1 1 cm3 around the treatment isocenter to find the highest PCC and the lowest dosimetric indices differences.
Results Figure 2 shows the distribution of variations of dose and the PCC using the CBCT scans taken at the time of resimulation requested by clinicians. Among 20 patients, 7 patients (3 head and neck, 2 prostate, and 2 lung cancers; green solid symbols) were determined not to need the new plan because of little difference in dose (<2%) between the original plan and the same plan on the new CT (no-new plan-group, no-action group). For the other 13 cases (dashed or dotted symbols), however, the treating physicians decided to use the new plan using the new CT after comparing the dose distribution of the old plan with that of the new CT (new-plan group, action group). As shown in Figure 2, all cases in the action group are in the region of a PCC <0.75 or dose change >5%, and all cases in the no-action group are in the region of a PCC >0.75 and dose change <5%. This suggests that action levels at which to consider re-CT for replanning are PCC <0.75 or dose change >5% (P < .001; c2 analysis). Here, Dxx,n is the CBCT’s Dxx (e.g., D95) of the n-th treatment fraction. Dose-volume histogram variation is defined as (Dxx,n/Dxx,1 e 1). Two typical cases of deviation are presented here as an illustration: a head and neck case (Fig 3) and a lung case (Fig 4). For the head and neck case, daily CBCT scans were acquired for the first 3 treatment days, followed by weekly CBCT scans, but daily CBCT scans were acquired
Practical Radiation Oncology: --- 2018
Figure 2 Maximum value of dose-volume histogram variations among D95, D90, D10, and D5 versus image correlation coefficient in 20 patients (5 with prostate cancer [A], 5 with lung cancer [C], and 10 with head and neck cancer [:]) at the time that replanning was requested clinically. Solid green symbols represent patients who did not need replanning; orange striped symbols represent patients whose predicted replanning was earlier than the actual clinical replanning; and purple checked symbols represent patients whose predicted replanning agreed with actual clinical replanning. (A color version of this figure is available at https://doi.org/10.1016/j.prro.2018.08.006.)
for the lung case. For the head and neck case, in 2 weeks, the PCC for the CTV decreased to 0.77, and the D10 and D5 of CTV experienced approximately a 20% change. The patient had lost about 4.5 kg of weight, and replanning was ordered by a treating physician. For the lung case, the changes in D95, D90, D10, and D5 were <5%, but the PCC fell below 0.65. This case went to replanning as a result of significant tumor shrinkage. The entire procedure, including data processing pipeline, evaluation, and alert reporting, is fully automated. On average, it takes approximately 10 minutes to compute
Figure 3 Temporal variation of indices of a patient with head and neck cancer. Solid line Z PCC; dotted lines Z dose-volume histogram variation of cone beam computed tomography planning target volume doses (D95, D90, D10, and D5) of the n-th fraction versus those of the first fraction.
Practical Radiation Oncology: --- 2018
Figure 4 Temporal variation of indices of a lung patient. Solid line Z Pearson’s correlation coefficient; dotted lines Z dosevolume histogram variation of cone beam computed tomography planning target volume doses (D95, D90, D10, and D5) of the n-th fraction versus those of the first fraction.
the dose distribution of a volumetric arc therapy plan on a CBCT image set and another 5 minutes to find the highest PCC around the treatment isocenter between the first and the newly acquired CBCT images. In our clinic, approximately 60 to 70 CBCT image sets are acquired per day on average. Retrieval of all CBCT DICOM image sets from the Varian Aria oncology information system takes approximately 30 minutes, and <2 hours is needed to process these tasks using parallel computation.
Discussion CBCT gained its popularity as an adaptive planning modality not only because of its geometric alignment capabilities but also because of its potential dosimetric evaluation functionality. As an IGRT tool, CBCT provides the capacity to capture such changes regularly during the treatment course in the form of serial snapshots and is a means of verifying accurate and precise radiation delivery.21 Once these changes exceed the predefined tolerance levels, physicians and physicists can more carefully review those CBCT images and evaluate changes to decide whether resimulation and replanning will be beneficial to the patient’s care. Schaly et al reported a threshold as the gamma criteria of 3-mm distance to agreement and 30 Hounsfield unit differences.22 This method provides 82% to 100% sensitivity and a 30% false-positive rate. This method only considers the anatomic change of the body. Hvid et al suggested daily dose surveillance of organs at risk, using CBCT images of head and neck cases.23 No action levels have been suggested. The method described in this paper uses a standard DICOM data stream to create an efficient data pipeline
Action levels for ART
5
from raw CBCT images to calculated geometric and dosimetric metrics. This system not only compares a patient’s anatomic changes based on CBCT images but also analyzes the dosimetric change trend based on CBCT scans starting from the first day of treatment. As a result, the system is able to provide radiation oncologists with accurate and concise information to determine whether resimulation and replanning are indicated and whether any treatment parameter should be adjusted accordingly. Furthermore, the system is configurable; once a change exceeds its predefined action levels, physicians and physicists will be automatically alerted through pagers and emails to review the patient’s treatment and evaluate the treatment plan. One big question in resimulation and replanning is how to determine the threshold level at which action is required. Attending physicians clinically determined to use replans with resimulation images for 13 patients (the action group of the 20 cases of preliminary study), but 7 patients (the no-action group) continued to use old plans. All patients in the action group demonstrated either PCCs < 0.75 or >5% in dose differences, whereas all Group B patients showed PCCs >0.75 and <5% dose differences (Fig 2). Suggested action levels of a PCC <0.75 or >5% in dose differences are reasonable criteria for replanning. The parameters of all 13 patients who ended up with new plans were analyzed for each fraction. Seven cases showed that the parameters reached the action level at or near the re-CT date, whereas 6 cases reached the action level earlier. This indicates that 6 patients might have benefitted if the decision to replan had been made earlier. This study used 20 cases of 3 anatomic sites. Because of the limited number of cases, it is not clear whether the ALDAV’s metric parameters are anatomy dependent. Furthermore, because the study is designed to use the resimulation cases only, the sensitivity of the ALDAV was not acquired. Patient daily variations, such as patient relaxation, bladder/rectal filling, and change in weight, are not taken in to account when analyzing the data; however, all patient setups of each anatomic site are per the previously commented Practice Guideline. A half fan image protocol is used to prevent image truncations. None of the CBCT images suffered from image truncations, except in inferosuperior directions. Because of this limitation, conclusions of this report may not be applied if significant anatomic changes happen outside the CBCT range, unless multiple CBCT scans are performed for every fraction. Nevertheless, this study will be a good starting point for a large-scale study with multi-institution collaboration to answer these questions. Current practice on determining re-CT is still a somewhat subjective clinical decision. This study suggests the utility of an automatic method with data-driven objective parameters. This study focused only on the target dose and the volume using rigid registration. Applying deformable image registration may further improve decision-making and would be
6
B. Zhang et al
especially useful in developing action levels on the OARs.
Conclusions An automated method to determine the need for patient replanning was designed and tested. The software tool collects the CBCT scans and automatically evaluates changes in dose and target shape. The efficiency of this method was tested retrospectively on studies from 20 patients with cancer and found to be practical. Based on our clinical experience, a PCC <0.75 or dose difference >5% has led to replanning, and based on this, we have established PCC <0.75 and dose difference >5% as action levels for consideration of replanning by the attending physicians.
Acknowledgments The authors thank Dr. Nancy Knight for critical comments and professional opinions during manuscript preparation.
References 1. Yan D, Vicini F, Wong J, Martinez A. Adaptive radiation therapy. Phys Med Biol. 1997;42:123-132. 2. Yan D, Lockman D, Brabbins D, Tyburski L, Martinez A. An offline strategy for constructing a patient-specific planning target volume in adaptive treatment process for prostate cancer. Int J Radiat Oncol Biol Phys. 2000;48:289-302. 3. Mohan R, Zhang X, Wang H, et al. Use of deformed intensity distributions for on-line modification of image-guided IMRT to account for interfractional anatomic changes. Int J Radiat Oncol Biol Phys. 2005;61:1258-1266. 4. Wu Q, Chi Y, Chen PY, Krauss DJ, Yan D, Martinez A. Adaptive replanning strategies accounting for shrinkage in head and neck IMRT. Int J Radiat Oncol Biol Phys. 2009;75:924-932. 5. Court LE, Tishler RB, Petit J, Cormack R, Chin L. Automatic online adaptive radiation therapy techniques for targets with significant shape change: A feasibility study. Phys Med Biol. 2006;51:24932501. 6. Burridge N, Amer A, Marchant T, et al. Online adaptive radiotherapy of the bladder: Small bowel irradiated-volume reduction. Int J Radiat Oncol Biol Phys. 2006;66:892-897.
Practical Radiation Oncology: --- 2018 7. Lei Y, Wu Q. A hybrid strategy of offline adaptive planning and online image guidance for prostate cancer radiotherapy. Phys Med Biol. 2010;55:2221-2234. 8. Thilmann C, Nill S, Tücking T, et al. Correction of patient positioning errors based on in-line cone beam CTs: clinical implementation and first experiences. Radiat Oncol. 2006;1:16. 9. Sorcini B, Tilikidis A. Clinical application of image-guided radiotherapy, IGRT (on the Varian OBI platform). Cancer Radiother. 2006;10:252-257. 10. Nakagawa K, Yamashita H, Shiraishi K, et al. Verification of intreatment tumor position using kilovoltage cone-beam computed tomography: A preliminary study. Int J Radiat Oncol Biol Phys. 2007;69:970-973. 11. Ding GX, Duggan DM, Coffey CW, et al. A study on adaptive IMRT treatment planning using kV cone-beam CT. Radiother Oncol. 2007;85:116-125. 12. Chen S, Le Q, Mutaf Y, et al. Feasibility of CBCT-based dose with a patient-specific stepwise HU-to-density curve to determine time of replanning. J Appl Clin Med Phys. 2017;18:64-69. 13. Guan H, Dong H. Dose calculation accuracy using cone-beam CT (CBCT) for pelvic adaptive radiotherapy. Phys Med Biol. 2009;54: 6239-6250. 14. Hatton J, McCurdy B, Greer PB. Cone beam computerized tomography: The effect of calibration of the Hounsfield unit number to electron density on dose calculation accuracy for adaptive radiation therapy. Phys Med Biol. 2009;54:N329-N346. 15. Yang Y, Schreibmann E, Li T, Wang C, Xing L. Evaluation of onboard kV cone beam CT (CBCT)-based dose calculation. Phys Med Biol. 2007;52:685-705. 16. Yoo S, Yin FF. Dosimetric feasibility of cone-beam CT-based treatment planning compared to CT-based treatment planning. Int J Radiat Oncol Biol Phys. 2006;66:1553-1561. 17. Rong Y, Smilowitz J, Tewatia D, Tome WA, Paliwal B. Dose calculation on kV cone beam CT images: An investigation of the Hu-density conversion stability and dose accuracy using the sitespecific calibration. Med Dosimetry. 2010;35:195-207. 18. Dunlop A, McQuaid D, Nill S, et al. Comparison of CT number calibration techniques for CBCT-based dose calculation. Strahlenther Onkol. 2015;191:970-978. 19. Rodgers JL, Nicewander WA. Thirteen ways to look at the correlation coefficient. Am Statistician. 1988;42:59-66. 20. NEMA PS3/ISO 12052, Digital Imaging and Communications in Medicine (DICOM) Standard: National Electrical Manufacturers Association, Rosslyn, VA, USA. Available at: http://medical.nema. org. Accessed October 1, 2017. 21. Goyal S, Kataria T. Image guidance in radiation therapy: Techniques and applications. Radiol Res Pract. 2014;2014:705604. 22. Schaly B, Kempe J, Venkatesan V, Mtichell S. Using gamma index to flag changes in anatomy during image-guided radiation therapy of head and neck cancer. J Appl Clin Med Phys. 2017;18:79-87. 23. Hvid C, Elstrom U Jensen K, Grau C. Cone-beam computed tomography (CBCT) for adaptive image guided head and neck radiation therapy. Acta Onclogica. 2018;57:552-556.