Geometrical variability of esophageal tumors and its implications for accurate radiation therapy - Chapter 10: General discussion

(1)

UvA-DARE is a service provided by the library of the University of Amsterdam (https://dare.uva.nl)

Geometrical variability of esophageal tumors and its implications for accurate

radiation therapy

Jin, P.

Publication date

2019

Document Version

Other version

License

Other

Link to publication

Citation for published version (APA):

Jin, P. (2019). Geometrical variability of esophageal tumors and its implications for accurate

radiation therapy.

General rights

It is not permitted to download or to forward/distribute the text or part of it without the consent of the author(s) and/or copyright holder(s), other than for strictly personal, individual use, unless the work is under an open content license (like Creative Commons).

Disclaimer/Complaints regulations

If you believe that digital publication of certain material infringes any of your rights or (privacy) interests, please let the Library know, stating your reasons. In case of a legitimate complaint, the Library will make the material inaccessible and/or remove it from the website. Please Ask the Library: https://uba.uva.nl/en/contact, or a letter to: Library of the University of Amsterdam, Secretariat, Singel 425, 1012 WP Amsterdam, The Netherlands. You will be contacted as soon as possible.

(2)

527155-L-bw-Jin 527155-L-bw-Jin 527155-L-bw-Jin 527155-L-bw-Jin Processed on: 18-12-2018 Processed on: 18-12-2018 Processed on: 18-12-2018

Processed on: 18-12-2018 PDF page: 153PDF page: 153PDF page: 153PDF page: 153

10

(3)

10.1 Summary

The studies presented in this thesis investigated the geometrical variability of esophageal tumors, including observer variability in tumor delineation, interfractional tumor position variation, and intrafractional tumor motion. In addition, we also considered the clinical implications of these geometrical uncertainties in esophageal cancer radiation therapy (RT).

The study in Chapter 2 quantified the inter- and intra-observer variation of the gross tumor volume (GTV) delineation in the planning computed tomography (pCT) with and without the presence of fiducial markers. The use of fiducial markers significantly reduced both inter- and intra-observer variation of the GTV delineation and, thereby, can also reduce the variation in clinical target volume (CTV) delineation.

In Chapter 3, the markers and pre-treatment cone-beam computed tomography (CBCT) scans were used to quantify the interfractional tumor position variation relative to bony anatomy. When the bony anatomy-based setup verification was used, the interfractional tumor position variation was found to be most pronounced in the cranial–caudal (CC) direction as compared to the left– right (LR) and anterior–posterior (AP) direction. In the cardia it is large in all three directions, whereas larger systematic and random errors were observed in the distal esophagus and cardia when the carina was used for setup verification compared to bony anatomy-based setup verifica-tion. Therefore, in Chapter 4, we concluded that the carina-based setup verification is not favor-able for esophageal cancer RT. Another interfractional uncertainty is the variability of gastroin-testinal gas volume, which can degrade the delivered dose distribution. Our findings in Chapter

5 suggest that a density override in treatment planning could mitigate this dose degradation. Chapters 6 to 9 focused on the induced esophageal tumor motion. The

respiration-induced tumor motion was quantified using fiducial markers and four-dimensional CT (4D-CT) scans. In Chapter 6, the respiration-induced tumor motion was shown to be mainly large in the CC direction and pronounced in the distal esophagus as well as cardia. The phantom study pre-sented in Chapter 7 investigated the feasibility and image quality of retrospectively reconstructing 4D-CBCT. Since this resulted in good image quality, we subsequently applied the retrospectively reconstructed 4D-CBCT to quantify the interfractional variability of the respiration-induced tu-mor motion. In Chapter 8, the interfractional variability of amplitude and trajectory shape of the respiration-induced tumor motion was found to be small. Based on these findings, in Chapter

9, use of the mid-position (MidP) strategy was compared with the internal target volume (ITV)

strategy in terms of target coverage and dose to the organs at risk (OARs). Due to the smaller probability-based planning target volume (PTV) in the MidP strategy, the dose to the OARs was significantly reduced without compromising the CTV coverage.

This chapter discusses the different approaches used to mitigate the geometrical uncertain-ties and their clinical implications in esophageal cancer RT. In addition, future perspectives of esophageal cancer RT are addressed.

(4)

10

10 10.2 Geometrical uncertainties and mitigation strategies

In the previous chapters, we demonstrated the necessity of using online bony anatomy-based setup verification and daily inspection of anatomical changes. We also recommended the use of fiducial markers and 4D-CT for treatment planning to reduce delineation uncertainty and to account for respiration-induced tumor motion. However, in addition to the geometrical uncertainties dis-cussed in this thesis, other less-substantial geometrical uncertainties exist. Moreover, there are alternative strategies to mitigate the geometrical uncertainties. We will discuss these uncertain-ties and strategies in the following subsections.

Delineation variability

The advances in highly conformal techniques of dose delivery, such as intensity-modulated RT (IMRT) and volumetric-modulated arc therapy (VMAT), require an accurate target definition. An inadequate target definition can result in an underdose to the tumor, which can lead to a higher locoregional recurrence rate; however, an excessive target definition can increase the incidence of toxicity and complication due to the increased dose to the OARs. The delineation variability of the GTV and CTV was believed to be the main source of geometrical uncertainty and, therefore, may be the weakest link in RT accuracy [244–246].

In Chapter 2, it was found that the inter- and intra-observer delineation variation in GTV of the esophageal tumor was large without the aid of fiducial markers, especially in the longitudinal (i.e., CC) direction (>1 cm). Initially, the use of18_{F-fluorodesoxyglucose positron emission}

tomogra-phy (PET)-CT was expected to reduce the observer variation in the esophageal tumor delineation because this was found true for both lung and pancreatic tumor delineations [142,247–252]; on the contrary, the observer variation in esophageal tumor delineation was still found to be large even though PET-CT was added for guidance [64,120]. Due to the low resolution and lack of universal thresholds of uptakes, the use of PET-CT only retains its superiority in the diagnostic regime. Currently, the use of fiducial markers is regarded as the gold standard for GTV delineation on CT [64,231].

In fact, it is the delineation variation in the CTV, rather than the GTV, that is of importance in the conventional fractionated esophageal cancer RT. Since the CTV in the longitudinal direction is generated by extending the GTV with a margin of 30–35 mm, the GTV delineation variation can be propagated to the CTV. Although the involved nodes will also be included in the CTV if they are not covered by the standard GTV-to-CTV margin, this accounts for only a small fraction of the patients. Moreover, the GTV delineation accuracy can also play a role in the case of a boost dose in the GTV. As a result, we can assume that the variation of CTV delineation is also substantial in the CC direction and will be reduced when the fiducial markers are used in the GTV extent determination.

(5)

Nevertheless, there might be extra CTV delineation variation between observers due to a dif-ferent interpretation of the anatomical and diagnostic information, even if the GTV is delineated in exactly the same way [253]. The use of both a consensus guideline and a digital delineation atlas have shown to result in a significant reduction in CTV delineation variation for rectal cancer RT [254]. It is also recommended that the observers strictly follow the delineation guideline for esophageal and gastroesophageal junction (GEJ) tumors [253].

In addition to the use of PET-CT and markers, magnetic resonance imaging (MRI) has also been investigated for target definition and delineation in RT treatment planning due to its flexibil-ity of varying scanning parameters to obtain a high contrast between different soft tissues. Initially, the fusion of MRI and CT was often used in the target volume delineation of many tumor sites such as brain cancer, head-and-neck cancer, and pelvic cancer [150,251,255–259]. Recently, the addition of MRI was found to reduce the overall target delineation variation for pancreatic cancer [152,250]. For esophageal cancer RT, diffusion-weighted (DW) MRI can facilitate the GTV delineation in the longitudinal direction, whereas the regular T2-weighted imaging did not improve the GTV delineation [153]. Moreover, the combination of fiducial markers and regu-lar T1/T2-weighted MRI was considered to be feasible, showing good visibility of the markers [154,260,261]. In this case, it is possible to take advantage of the markers in reducing delineation variation as well as the superiority of MRI in image quality.

Setup verification

In Chapters 3 and 4, we presented the interfractional setup errors based on bony-anatomy reg-istration and carina regreg-istration, where we asserted that bony anatomy-based setup verification is currently the most appropriate approach. Although the target-based registration would be the ideal approach to eliminate the target setup errors, we believe that the fiducial markers implanted at the tumor borders and in the center of the tumor are the most representative surrogate for the macroscopic tumor position. However, the result of automatic rigid registration based on a vol-ume of interest, including the markers and the tumor, was mostly not clinically acceptable. The main cause for this was that the deformation in the target volume was prohibitively large and com-plex for rigid registration to find a good match between the two sets of markers [176]. Although a manual rigid registration was attempted when the automatic rigid registration did not work, the manual registration did not result in an acceptable match either, and it was rather time consuming and subjective, due to the considerable deformation. Therefore, we concluded that, unfortunately, marker-based setup verification, in spite of the clinical benefits, is not clinically feasible.

This marker-involved automatic soft-tissue-based rigid registration has, however, been success-fully performed for setup verification in pancreatic cancer [161]. This is probably because the de-formation of the target volume in pancreatic cancer was not as large as that in the case of esophageal cancer. The standard deviation (SD) of the residuals in the linear regression of the pairwise

(6)

10

tance between the markers was up to 1.5 mm in the pancreatic cancer case; this is much smaller than the SD we found in the esophageal cancer case (up to 6.7 mm). Moreover, the pairwise dis-tance between the markers was up to 37.9 mm in these pancreatic cancer patients versus 177.9 mm in the investigated esophageal cancer patients. These data confirm that the target volume defor-mation was smaller in pancreatic cancer than in esophageal cancer, which may be due to the fact that the pancreatic tumor is solid and less elongated.

Nevertheless, the automatic soft-tissue-based rigid registration in esophageal cancer patients has recently been employed in two studies [262,263]. In these studies, fiducial markers were also implanted at the tumor borders and in the middle of the tumor to aid the registration. The air present in the esophagus or stomach, and the lateral displacement of the esophagus, were found to be the cause of the difficulties in the soft-tissue-based rigid registration [263]. Although both studies indicated that the soft-tissue contrast on CBCT was too low to perform an accurate reg-istration, they were still in favor of the soft-tissue-based rigid registration since it can reduce the CTV-to-PTV margin. As discussed in Chapter 3, we arrived at a conflicting conclusion due to the difficulty in performing this registration and assessing the registration results. In our opinion, the bony anatomy registration is currently the most reliable approach for setup verification.

Cardiac activity-induced tumor motion

For the intrafractional uncertainty, we mainly investigated the respiration-induced tumor motion, which is the main source of the intrafractional tumor motion. However, the cardiac activity-induced esophageal tumor motion was not thoroughly discussed. One study indicated that the cardiac activity has a similar weight on the intrafractional esophageal tumor motion as respiration [264]. In that study, 4D-CT was acquired during a deep-inspiration breath-hold (DIBH) and sorted into 10 phases based on the cardiac cycle. Deformable image registration (DIR) was used to register the 10 phases, yielding the voxel-to-voxel motion trajectory [265]. The average maxi-mum motion amplitudes in the LR and AP directions were in fact smaller than 5 mm. However, due to the large voxel size in the CC direction, it was claimed that the findings of the motion in the CC direction were not reliable. Moreover, despite the DIBH, the quantified cardiac activity-induced tumor motion might include a residual motion due to the decrease in lung volume, which could be caused by the continuous oxygen uptake during the DIBH [131,132]. In another study, fiducial markers and a real-time fluoroscopic tracking system were applied to analyze the intrafrac-tional digestive tract motion including the esophagus and GEJ [121]. By analyzing the frequency spectrum of the motion, the cardiac activity-induced motion was found to be much smaller in am-plitude but with a higher frequency than the respiration-induced motion. However, the absolute amplitude of the cardiac activity-induced motion was not derived in that study [121].

To our knowledge, no other reports are available on cardiac activity-induced esophageal tu-mor motion. In most studies that aimed to quantify the esophageal tutu-mor motion using a

(7)

ing motion-sorted 4D-CT [165,170,210], the respiration-induced esophageal tumor motion was a combination of all motions, including respiration-induced and cardiac activity-induced tu-mor motion. As concluded in the above-mentioned study [121], the cardiac activity-induced tu-mor motion has a higher frequency and a smaller amplitude compared to the respiration-induced tumor motion. Therefore, the combined motion could have a frequency similar to that of the respiration-induced motion, and the cardiac activity-induced tumor motion is integrated into the quantified respiration-induced tumor motion. In our opinion, for RT of esophageal cancer with free breathing, there is no need to include an extra margin for cardiac activity-induced tumor mo-tion compensamo-tion.

Breath-holding technique

The use of breath-holding (BH) techniques has become more common in motion management for thoracic and abdominal tumors due to advances in accurate RT in all respects, especially for stereotactic body RT (SBRT). According to a recent review [266], BH is beneficial in almost all thoracic and upper abdominal tumor sites, irrespective of whether or not the RT is hyperfrac-tionated [130,207,266–270]. Compared with other motion management techniques, BH shows superiority in treatment planning workload, interplay effect, size of the PTV margin, and its clin-ical consequence (i.e., reduced toxicity in OARs). However, it requires more patient collabora-tion (such as pre-treatment DIBH training) and not all patients are able to endure holding their breath. Also, it can result in longer treatment time than using real-time tumor tracking or a motion-encompassing strategy (e.g., ITV or MidP strategy).

Although BH has not yet been widely discussed for RT of esophageal cancer, it has been for RT of lung and liver cancer. This is probably because, compared to esophageal tumors, the respiration-induced motion of lung and liver tumors can vary more dramatically over the treatment course [203,271–275]. In one study on lung cancer patients, the reproducibility of respiration-induced tumor motion alone was found to be over 50%, but decreased to 20–30% when the baseline shift was taken into account [273]. In another study, the lung tumor motion measured in the 4D-CT scan approximated the average range of tumor motion, but underestimated the maximum range of tumor motion [275]. Other studies on lung cancer (SB)RT claimed that it is adequate to use a sin-gle 4D-CT to evaluate the respiration-induced tumor motion, because the intrafractional motion amplitude and trajectory shape variability was limited [212,226]. However, based on these find-ings, the unrepresentativeness of 4D-CT lies mainly in the baseline shift, which can probably be compensated for by using tumor-based setup verification with the help of 4D-CBCT [128,212]. Furthermore, if the passive motion-encompassing strategy is used in the (SB)RT of lung cancer, the margin for the respiration-induced tumor motion compensation can be small due to the large penumbra in the lung tissue [125]. As the baseline shift of lung tumor can be eliminated by us-ing tumor-based setup verification, this margin is approximately 6.6 mm for a respiration-induced

(8)

10

tumor motion of 10 mm in amplitude, including the delineation uncertainty [128]. Whereas the margin can be reduced to (at most) 5 mm if the ideal DIBH can be achieved, this margin is then only used to account for delineation uncertainty. In this case, the difference in CTV-to-PTV margin between the DIBH and motion-encompassing strategies could be minor for lung cancer (SB)RT.

For liver cancer patients, both planning 4D-CT and in-room 4D-CBCT have been used to quantify the inter- and intrafractional liver motion. For the majority of patients, the motion am-plitude was found to be consistent over the treatment course, where the differences were (on av-erage) smaller than 2 mm in the LR, CC, and AP directions [182,276]. Similar to the findings for lung tumors, the baseline shift of the liver tumor was found to introduce much more uncer-tainty than the respiration-induced liver tumor motion. However, when fiducial markers and CBCT projections were used, the interfractional variation of peak-to-peak respiration-induced liver tumor motion was found to be large [271]. Other studies used similar methods to monitor the respiration-induced motion of liver and pancreatic tumors; they concluded that both inter-and intrafractional motion variations were large inter-and that using only 4D-CT to account for the respiration-induced tumor motion would not be appropriate [203,272,274]. The inconsistent results reported by the above-mentioned studies might be attributable to differences in the mo-tion quantificamo-tion approaches used. One study showed that the 4D-CBCT could underrepresent the actual target motion because of the averaging effect, whereas the CBCT fluoroscopy projec-tion images with markers can reflect the actual moprojec-tion [277].

Ideally, the use of BH can eliminate the ITV and minimize the CTV-to-PTV margin. In prac-tice, however, one concern of active DIBH is its stability during one BH, reproducibility between the different BH in one fraction, and reproducibility between the BH in different fractions. In one study on liver patients, the diaphragm surface motion was used to represent the liver motion [278]; the authors found that in the CC direction, the average maximum variation of the diaphragm sur-face motion during a BH was only 1.4 mm, the average intrafractional BH reproducibility was 1.5 mm, whereas the average interfractional BH reproducibility was 3.4 mm. In another study, the positions of both diaphragm and fiducial markers implanted in the liver were evaluated for BH reproducibility [267]; the authors revealed that for diaphragm and markers, the average intrafrac-tional reproducibility was 2.5 mm and 2.3 mm, respectively, and the average interfracintrafrac-tional repro-ducibility was 4.4 mm and 4.3 mm, respectively, in the CC direction. Both studies concluded that daily imaging to monitor the interfractional offset is needed for potential additional repositioning after bony anatomy-based setup correction.

For pancreatic cancer patients, substantial tumor motion was observed during an inhalation BH, with an average absolute maximum motion of 4.2 mm and 2.7 mm in the CC and AP direction, respectively [131]. However, it was (on average) only 1.7 mm and 0.8 mm in the CC and AP direction, respectively, implying that it was small for the majority of patients. For patients with liver or pancreatic tumors, the exhalation BH was found to be more stable than the inhalation BH [132,

(9)

267,278]. However, the duration of the exhalation BH might be shorter than that of inhalation BH; this means that more BH sessions are needed during one treatment fraction. Moreover, for thoracic tumors, the exhalation BH might not be preferred since one benefit of inhalation BHs (e.g., DIBH) is to spare the lungs and the heart.

So far, few studies have applied the (DI)BH in esophageal cancer RT [194,195,279], which generally rely on the active breathing control system [130]. In these studies, a reduction in dose to the lungs and the heart was observed owing to the use of a smaller CTV-to-PTV margin. However, no detailed study has been carried out on the stability during a BH or intra-/interfractional DIBH reproducibility regarding esophageal tumor position. In the above-mentioned study investigating the cardiac activity-induced esophageal tumor motion [264], the residual motion (probably due to both cardiac activity and lung volume decrease) may be 5–10 mm during a DIBH. Therefore, the stability of esophageal tumors during a DIBH might be far from optimal.

In Chapter 8, the interfractional variability of respiration-induced motion of esophageal tu-mors was found to be limited in both amplitude and trajectory using markers and 4D-CBCT [214]. In contrast, in a recent study based on 20 patients with distally located tumors and quantita-tion using cine-MRI, the intrafracquantita-tional esophageal tumor moquantita-tion was found to be highly variable within the individual patient [280]. However, this study had a small patient cohort (as did the study in Chapter 8). The different conclusions might stem from the different methods used for quantification, similar to the contrasting results found in the above-mentioned studies on motion quantification of liver tumors. The respiration-induced esophageal tumor motion (albeit possibly variable for a few patients), could make a small contribution to the CTV-to-PTV margin, espe-cially when the MidP strategy is used (as in Chapter 9). Despite the fact that the DIBH could slightly reduce the CTV-to-PTV margin, the tradeoff between the complexity of clinical imple-mentation and limited dosimetric advantages should be balanced.

4D-CBCT

Due to potential underestimation of the respiration-induced tumor motion in a free-breathing 3D-CBCT [281], respiratory-correlated CBCT (i.e., 4D-CBCT) was introduced into RT clinical prac-tice as an addition to the linear accelerator (LINAC)-mounted CBCT, to reduce the respiration-induced blur and gain time-resolved information [211,282]. Nowadays, 4D-CBCT is commonly applied in free-breathing SBRT for lung and liver cancer, especially when a motion-encompassing strategy (such as the MidP strategy) is used [128,218,241]. In these applications, the 4D-CBCT has three main purposes: setup verification, respiration-induced tumor motion estimation, and inspection of other anatomical changes. As mentioned in Chapter 6, the 4D-CBCT is usually acquired with a half-arc (200°) rotation and the imaging time could be up to 4 min to keep an acceptable image quality. However, in this case the field of view (FOV) is probably too small to include the whole body cut in the transverse plane [283]. This means that (sometimes) the

(10)

10

changes in body contour, lungs, and liver cannot be monitored. If a full arc with shifted detector is used (as in the acquisition of 3D-CBCT for the thoracic and abdominal region) the required imaging time will be up to 8 min to retain sufficient image quality; however, this is not desired in a clinical RT workflow. In Chapter 6, we also explored the possibility to acquire the 4D-CBCT with a full arc, balancing the image quality and acquisition duration while retaining the imaging dose. Although the visibility of any fiducial markers in those tested 4D-CBCT was sufficient for the marker-based 4D registration between each of the 10 breathing phases to the planning CT, the overall image quality was not good. In the last decade, many advances have been made in 4D-CBCT techniques for a better image quality and/or lower imaging dose [284–294]. Thus, the 4D-CBCT is a promising imaging modality to be implemented in RT for image guidance.

Recently, a study including eight lung cancer patients found that the 4D-CBCT always under-represents the real motion, irrespective of the reconstruction algorithm [277]. This has not been revealed by studies using the dynamic phantom because, in those studies, a constant breathing mo-tion amplitude was used. When patient data are used, the underrepresentamo-tion is caused by the averaging effect of the projections binned within the same breathing phase (i.e., the sorted bin). Moreover, the incidence of a sudden large amplitude cannot be picked up by the 4D-CBCT and motion artifacts can easily be induced by an irregular breathing pattern. Consequently, it would be necessary to train the patients for regular breathing with less intrafractional tumor motion vari-ation [295]. Further, VMAT may be preferable because of its short dose delivery time, implying that the intrafractional tumor motion variation could play a less important role in the geometrical uncertainty [296].

Respiratory gating and tumor tracking

Alternatives to using a BH technique are respiratory gating and real-time tumor tracking. These latter methods can result in a small CTV-to-PTV margin, require no extensive patient collabora-tion, and can minimize the interplay effect. However, they are associated with a relatively heavy treatment workload with long treatment time [266,297]. Both the gating and tracking methods require either an external or internal surrogate to trigger the beam delivery or to track the tumor motion. Due to the requirement of an additional real-time imaging system [298] and limited addi-tional dosimetric benefits compared to BH [228,299–301], respiratory gating and real-time tumor tracking are not as commonly used as the passive motion-encompassing and active BH strategies for motion management using a standard LINAC.

Although respiratory gating is more appealing in the motion mitigation for proton/particle therapy because of a higher sensitivity to motion [302–304], considerable effort has focused on the development of respiratory-gated RT using a LINAC [305,306]. When an external surrogate was used for gating/tracking for lung, liver, and pancreatic cancer patients, a high correlation was found between the surrogate and the tumor motion. In addition, the tumor position was found

(11)

to be reproducible in the end-exhalation breathing phase [307]. However, unpredictable residual tumor motion is still present [308,309] which can affect the accuracy of the triggered imaging and beam delivery. With an internal surrogate (such as the tumor itself or a fiducial marker), despite the better representation of the actual target motion [310], extensive respiratory-correlated imag-ing techniques for both plannimag-ing and daily image guidance are crucial. These are required to cope with the interfractional baseline shift as well as the inter- and intrafractional respiration-induced motion variation associated with most thoracic and abdominal tumors [311,312].

RT with real-time tumor tracking might reduce the treatment time compared with respiratory gated RT. Clinically, real-time tumor tracking has been demonstrated on the standard LINAC with dynamic multi-leaf collimator (MLC) tracking [123,313], the robot-based CyberKnife sys-tem [314], and the gimbal-based Vero system [315,316]. In a standard clinically available LINAC, fiducial markers are normally used as the surrogate for tumor tracking, the motion of which is mon-itored by the LINAC-integrated kV imaging system. This may also be combined with MV imaging by the electronic portal imaging device (EPID) or optical and sparse monoscopic imaging, with a mean geometrical accuracy of <0.5 mm [313,317]. However, so far, these systems are still rarely implemented for lung or liver cancer patients, who would benefit from them most [317]. For lung cancer patients, only real-time MLC tracking combined with an electromagnetic guidance system has been clinically implemented in the LINAC [318]. More clinical implementations of lung and liver SBRT have been reported for the CyberKnife and Vero systems [315,316,319]. The real-time tumor tracking for both systems is realized by monitoring both internal and external surrogates with a correlation model for motion prediction [320]. The geometrical accuracy was found to be high and comparable for the two systems (i.e.,⩽1.0 mm) [321–324].

For esophageal cancer RT, few studies have examined the use of respiratory gating or real-time tumor tracking [325]. The challenge is how to accurately monitor the tumor motion in real time over the whole treatment fraction. Currently, using fluoroscopy or cine-MRI to monitor the im-planted markers or the tumor itself during treatment would be a promising solution to accurate real-time tracking of the esophageal tumor.

Other anatomical changes

One anatomical change not previously mentioned is tumor volume reduction. As the fractionated RT continues, the gross volume of the esophageal tumor might shrink due to its response to RT. In some in-room imaging scans, such as CBCT, this may not be easily evaluated due to the limited soft-tissue contrast. In this case, another imaging modality (e.g., MRI) might be of added value for this purpose. When the prestenotic esophageal dilatation is relieved by tumor regression, the patient probably needs a re-planning for the sake of a lower dose to the OARs.

Body contour changes can also make a difference for esophageal cancer RT. Because of the obstruction in the esophagus due to the esophageal tumor, patients often experience difficulties

(12)

10

in nutrition intake. The RT and food-intake-related body contour change could change the dose distribution. All these possible anatomical changes emphasize the importance of daily imaging for monitoring purposes.

Despite the limited image quality of CBCT, the body contour and most of the crucial OARs (such as lung, heart, and liver) can be clearly visualized. With the help of markers at the cranial and caudal borders of the tumor, we can clearly indicate the position of the tumor in the CBCT. In addition, the possibility of using adapted-CBCT for dose calculation has been reported [326,

327]. Therefore, when severe anatomical changes occur, recalculating the dose based on CBCT could play an important role in RT of esophageal cancer to inspect and verify the effect of these anatomical changes.

10.3 Safety margin

For esophageal cancer RT worldwide, some centers do not use the CTV concept and only ap-ply a large GTV-to-PTV margin of up to 50 mm in the CC direction to cover the relevant lymph nodes, microscopic spread, and possible geometrical uncertainties [17,44,46]. Other centers often prescribe an isotropic CTV-to-PTV margin of 5–10 mm [48,49,78,172,328]. However, these clinically used margins do not seem to be based on published studies discussing interfrac-tional setup errors and respiration-induced motion patterns for esophageal tumors [36,70,165,

168,170,176,210]. There are several reasons why no actual margins for esophageal cancer RT have been published. First, the published margin recipes are mostly based on the simulation of a spherical CTV in combination with a fixed beam penumbra [137,138,329,330]. However, the esophageal tumor has an elongated shape with a complicated and changing anatomical environ-ment. The mixed anatomical environment of lung tissue and soft tissues (such as heart, stomach, and liver) yields a varying penumbra in the whole region; therefore, it is difficult to apply the mar-gin recipe directly to esophageal cancer patients. Another reason is that the imamar-ging modality and setup verification methods used in the RT workflow for esophageal cancer, which can strongly in-fluence the margin, vary widely between institutes. Moreover, most studies concentrated on only one aspect of the geometrical uncertainties and proposed one margin component. Nevertheless, it is not convenient for most clinical institutes to apply it directly, due to the lack of integration with other geometrical uncertainties. Furthermore, both inter- and intrafractional tumor motions were found to be highly direction dependent and distinctive for tumors located in different regions of the esophagus, suggesting an anisotropic region-dependent margin. However, one issue is that the CTV can be extensive (e.g., cover the middle and distal sections of the esophagus, as well as the cardia). Although an anisotropic margin has been applied in most treatment planning systems, using a different margin for different parts of the CTV has not yet been clinically implemented. In

Chapter 9, anisotropic and region-dependent margins were experimentally applied to esophageal

cancer patients using a margin recipe commonly applied for most tumor sites [137]. The CTV was

(13)

manually divided into sub-volumes and one anisotropic margin was applied to each sub-volume, resulting in sub-PTVs. Subsequently, the sub-PTVs were merged into one PTV. This boosted our confidence about the feasibility of implementing a region-dependent margin in clinical practice.

In Chapters 7–9, we found that the respiration-induced tumor motion varies considerably be-tween patients, but is likely to be stable over the treatment course for individual patients. There-fore, a patient-specific margin is required. To obtain this, a 4D-CT is needed for treatment plan-ning. The respiration-induced motion quantification should be done by either using fiducial mark-ers or using reliable DIR software [235]. For online setup verification and daily inspection of anatomical changes, daily CBCT is recommended. Although we have reported the CTV-to-PTV margin based on our studies, techniques, and equipment (as presented in Chapter 9), it is recom-mended for clinics to create their own margins based on their treatment protocol.

10.4 Future perspectives

Much research and new clinical implementations have aimed for more accurate dose delivery, bet-ter prognosis, and less toxicity. For esophageal cancer radiation treatment, the MRI-integrated LINAC and proton therapy could be the most promising technologies in the coming decade.

MRI and MR-guided RT

MRI offers the advantage that it does not involve exposure to radiation, provides good soft-tissue contrast, and has the flexibility to tweak the imaging parameters for different visualization pur-poses. Since cardiac gating and respiratory gating are also available in the MRI acquisition, mo-tion artifacts can be minimized. Initially, the use of MRI was subject to the staging of primary tu-mor and lymph nodes [331–336]. However, the use of MRI can also improve delineation of the esophageal tumors and offer the option to combine MRI with markers [153,154,260,261]. More-over, MRI (particularly the dynamic contrast-enhanced MRI and DW-MRI) can aid the evalua-tion and predicevalua-tion of treatment response to RT with concurrent chemotherapy for esophageal cancer [337–341]. Recently, the MR-guided LINAC has been successfully developed and imple-mented in RT clinical practice [342–346]. Due to its good image quality in soft tissue and ability to image the patient in real-time in a respiratory and cardiac triggered manner, MRI offers more op-portunities for active intrafractional motion management (such as respiratory gating) compared to the CBCT-guided RT [347,348]. Some challenges still exist for MR-guided RT, especially re-lated to an MR-only workflow. Studies on the use of MRI or synthetic CT derived from MRI for accurate treatment planning are currently in progress [349–351]. Despite the current lack of clin-ical implementation of MR-guided RT for esophageal cancer, it might be a promising treatment modality to minimize the geometrical uncertainties for esophageal cancer RT.

(14)

10

10 Proton therapy

Proton therapy is receiving increasing attention. Proton or particle therapy is superior to the tra-ditional photon therapy because of the ability to deliver a lower dose proximal to the tumor and with no (or limited) dose distal to the tumor. Therefore, it can potentially reduce the dose to the OARs, resulting in minimum toxicity and greater possibility of dose escalation. Various planning studies have shown the dosimetric benefits of proton therapy for esophageal cancer in terms of the significantly reduced dose to the critical OARs, such as lung and heart [352–355]. However, these studies hardly considered the high sensitivity of the quality of plans resulting from proton therapy treatment plan optimization to range, setup, and motion uncertainties. Nevertheless, sev-eral prospective studies confirmed good locoregional tumor control, high tumor response, and less acute toxicity for proton therapy compared with photon RT [356–359]. Based on these find-ings, although there is a lack of randomized clinical trials, proton therapy could be a promising alternative treatment modality for esophageal cancer.

In proton therapy, the high-dose region is basically modulated by the spread of multiple Bragg peaks. In addition to the beam energy, the position of the Bragg peaks is highly dependent on the mass densities of the tissue. Despite the fact that the sharp dose fall-off is an advantage of the pro-ton beam, it also implies that the dose distribution can be easily altered by any slight anatomical change. This suggests that the range uncertainty is important in proton therapy. One of the chal-lenges in proton therapy is how to ensure the robustness of the treatment plan to the geometrical uncertainties [360,361]. For setup and range uncertainties, several approaches have been pro-posed for robust plan optimization [362–365]. However, for intrafractional motion uncertainty, solely relying on plan optimization might not be the optimal solution, despite the feasibility of accounting for respiration-induced tumor motion in the robustness optimization [366]. More-over, other anatomical changes (especially gastrointestinal gas pockets) can induce considerable dosimetric degradation in the proton dose distribution [189]. Therefore, a reliable motion man-agement strategy, accurate daily image guidance, efficient adaptive procedure, and robust plan evaluation methods are required, which are also the main difficulties in the current development of proton therapy for esophageal cancer [303,367,368].

10.5 Clinical implications

Geometrical uncertainty is an important factor influencing the accuracy of RT for esophageal can-cer. The use of fiducial markers is of importance as the first step in the reduction of these uncer-tainties. Based on our studies, they can reduce the target delineation uncertainty and facilitate daily inspection of the target position in the daily CBCT. Moreover, the fiducial markers can be exploited to verify the intrafractional respiration-induced tumor motion when a 4D-CBCT is ap-plicable. Should a recently closed randomized trial on dose escalation to the GTV show positive

(15)

results, using fiducial markers as tumor volume indicators will become more important in improv-ing the geometrical accuracy in esophageal cancer RT.

Despite a preference to use an actual target for online setup verification, bony anatomy reg-istration using the pCT and daily CBCT is more reliable considering the potentially substantial deformation of the target. Daily CBCT also allows to evaluate rough anatomical changes, such as changes in gastrointestinal gas pockets. For patients with substantial gastrointestinal gas pockets present in the pCT, the use of density override in treatment planning is a simple and sound ap-proach to mitigate the potential underdose induced by the decreased gas volume during treatment. Our study indicated that a plan-selection strategy can ensure target coverage and minimize over-dose in the target to the greatest extent, considering the unpredictable changes of the gas volume during the treatment. This strategy relies on a plan library consisting of plans using different den-sity override settings, and plan selection depends on inspection of the daily gas volume. However, there could be additional workload in clinical practice. Therefore, it is recommended to assess the tradeoff between the workload and clinical gain of using this adaptive strategy.

To account for the intrafractional tumor motion (especially the respiration-induced tumor mo-tion) it is preferred to use 4D-CT for treatment planning. The use of 4D-CT allows to estimate the amplitude and trajectory shape of the respiration-induced tumor motion, in particular in combina-tion with the use of fiducial markers. With less artifacts and better image quality than the 4D-CT, the MidP-CT is also favorable for treatment planning. Compared to the ITV-based strategy, using the MidP strategy is preferred in terms of a reduction in dose to OARs, without compromising the target coverage.

To a large extent, both IMRT and VMAT treatment planning are robust against the inter- and intrafractional tumor motion. However, VMAT planning might be superior to IMRT in terms of its robustness to intrafractional tumor motion because of the shorter treatment time. The CTV-to-PTV margin is highly subject to the treatment protocol, such as the pCT modality (e.g., average 3D-CT or MidP-CT), the setup verification approach, and the intrafractional tumor motion man-agement. According to the characteristics of the inter- and intrafraction tumor motion found in our studies, applying anisotropic and region-dependent margins is preferable to a uniform margin. Promising technologies, such as 4D-CBCT of good image quality with accurately calibrated CT numbers, and MR-LINAC with an MR-only workflow, will improve geometrical accuracy in the RT of esophageal cancer. However, it is necessary to first evaluate their clinical benefits for the use of adaptive RT of esophageal cancer.