Geometrical variability of esophageal tumors and its implications for accurate radiation therapy - Chapter 9: Dosimetric benefits of mid-position compared with internal target volume strategy for esophageal cancer radiation

(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: 133PDF page: 133PDF page: 133PDF page: 133

9

Dosimetric benefits of mid-position compared

with internal target volume strategy for esophageal

cancer radiation therapy

P. Jin, M. Machiels, K.F. Crama, J. Visser, N. van Wieringen, A. Bel, M.C.C.M. Hulshof, and T. Alderliesten

A version of this chapter has been accepted for publication in

International Journal of Radiation Oncology * Biology * Physics. 2018; In press

(3)

Abstract

Purpose

Both mid-position (MidP) and internal target volume (ITV) strategies can take the respiration-induced target motion (RTM) into account. This study aimed to compare these two strategies in terms of clinical target volume (CTV) coverage and dose to organs at risk (OARs) for esophageal cancer radiation therapy (RT).

Materials and methods

Fifteen esophageal cancer patients were retrospectively included for neoadjuvant RT planning. Per patient, a 10-phase 4D-CT was acquired with CTV and OARs delineated on the 20% phase. The MidP-CT was reconstructed based on deformable image registration (DIR) between the 20% phase and the other nine phases; thereby the CTV and OARs delineations were propagated and an ITV was constructed. Both MidP and ITV strategies were used for treatment planning, yielding the planned dose. Next, these plans were applied to the 10-phase 4D-CT to calculate the dose dis-tribution for each phase of the 4D-CT. Based on the DIR, these calculated dose disdis-tributions were warped and averaged to yield the accumulated 4D-dose. Subsequently, we compared, in terms of CTV coverage and dose to OARs, the planned dose with the accumulated 4D-dose and also the MidP strategy with the ITV strategy.

Results

The differences between the planned dose and accumulated 4D-dose were limited and clinically irrelevant. In 14 patients, both MidP and ITV strategies showed V95%>98% for the CTV.

Com-pared to the ITV strategy, the MidP strategy showed a significant reduction of approximately 10% in the dose-volume histogram parameters for the lungs, heart, and liver (p<0.001, Wilcoxon signed-rank test).

Conclusions

Compared to the ITV strategy, the MidP strategy in treatment planning can lead to an approxi-mately 10% reduction in the dose to OARs with an adequate CTV coverage for esophageal cancer RT.

(4)

999999999

9.1 Introduction

Preoperative radiation therapy (RT) with concurrent chemotherapy can improve survival among patients with potentially curable esophageal or gastroesophageal-junction (GEJ) cancer [17]. The planning target volume (PTV) is used to encompass the clinical target volume (CTV) in the presence of tumor delineation and inter-/intrafractional tumor position variations. Respiration-induced target motion (RTM) is the primary source of intrafractional uncertainty for RT of esopha-geal or GEJ cancer. This is most pronounced in the cranial–caudal (CC) direction and its average amplitude is 1–4 mm in the proximal esophagus and 5–8 mm in the distal esophagus and GEJ [165,210,214]. However, the interfractional variability of RTM for esophageal and GEJ tumors is limited (mean variation⩽1.4 mm), therefore a single 4D computed tomography (CT) would be sufficient for estimation and prediction of the RTM over the treatment course [214].

The internal target volume (ITV) is commonly used for taking into account the RTM uncer-tainty in treatment planning, which covers the whole RTM of the CTV [74]. Alternatively, the mid-ventilation (MidV) or mid-position (MidP) strategy can be used, where the RTM can be in-cluded as a random positioning error in the probability-based CTV-to-PTV margin [137]. For the lung stereotactic body RT (SBRT), liver SBRT, and pancreatic cancer RT, the MidV/MidP strategy reduced the PTV volume by 25%, 34%, and 14%, respectively compared with the ITV strategy [127,202,227,228]. Consequently, the mean dose to organs at risk (OARs) was signif-icantly reduced by 0.5–2.0 Gy, which was only reported in a few dosimetry studies of comparing the two strategies [202,228]. The use of the MidV/MidP strategy has shown good locoregional control and overall survival for lung SBRT [129] and the possibility of dose escalation in the gross tumor volume (GTV) for liver SBRT [229].

For esophageal cancer RT, the dosimetric impact of using the MidV/MidP strategy cannot be directly derived from the findings in other tumor sites because of the different RTM pattern and different OARs [214]. Most studies on the use of 4D-CT and motion management in esophageal cancer RT focused on the RTM amplitude [70,165,170,198,199,210,214]. Only few studies discussed the use of ITV for esophageal cancer RT [68,71,230] and neither ITV nor MidV/MidP strategies have yet been analyzed in terms of their dosimetric impacts.

The implementation of an ITV strategy is hindered by the labor-intensive CTV delineation [71], which is also prone to delineation uncertainties especially without the aid of fiducial markers [231]. For the MidV/MidP strategy, a region and direction-dependent quantification of the RTM amplitude is required for margin calculation [125,210,214]. However, the absence of fiducial markers in clinical practice makes this quantification tough [165,210,214].

Due to the potentially reduced dose to OARs with possibly lower toxicity [50–53,56] and the need of dose escalation [48,49] as shown in other tumor sites [129,228,229], it is essential to investigate the implementation and dosimetric impact of using the MidV/MidP strategy for esophageal cancer RT. In this study, we therefore aimed to apply the MidP strategy in treatment

(5)

planning for free-breathing esophageal cancer RT and to compare it with the ITV strategy in terms of PTV volumes, CTV coverage, and dose to the OARs, using 4D-dose accumulation based on deformable image registration (DIR).

9.2 Materials and methods

Patients, imaging data, and manual delineations

We retrospectively included 15 patients with esophageal cancer, who were consecutively treated between December 2015 and May 2016 (Table 9.1). The requirement of additional ethical ap-proval for human subject involvement was waived by the local medical ethics review committee.

Per patient, one planning 4D-CT was acquired (LightSpeed RT 16; General Electric, Wauke-sha, WI) with patients breathing freely in head-first supine position using an arm and knee support (CIVCO Medical Solution, Coralville, IA). All 4D-CT scans were sorted into 10 breathing phases (phase binning, 0–90%) and the average intensity projection CT (AIP-CT) was automatically de-rived (Advantage 4D software; General Electric). The in-plane pixel size was 1.0 mm or 1.3 mm depending on the field of view. The slice thickness was 2.5 mm. No serious artifacts were observed in the 4D-CT scans.

Per patient, the gross target volume (GTV) was retrospectively delineated on the 20% phase of the 4D-CT by an experienced radiation oncology resident with the aid of all available diagnos-tic information including diagnosdiagnos-tic positron emission tomography and/or CT scans, endoscopy, and endoscopic ultrasound. We chose the 20% phase as reference due to its proximity to the MidP, which may result in less effort to propagate the delineations to the other phases than when using one of the other phases as reference. The CTV was generated by extending the GTV in the CC direction with a 20-mm margin in the cardiac region or a 35-mm margin above the GEJ and pe-ripherally with coverage of the para-esophageal fatty tissue and involved lymph node stations.

The OARs (i.e., lungs, heart, liver, kidneys, and spinal cord) were delineated initially on the AIP-CT by the radiation therapists as part of the clinical treatment planning. By performing DIR between the AIP-CT and the 20% phase of the 4D-CT, the OAR delineations were propagated to the 20% phase of the 4D-CT. All DIRs in the present study were done using ADMIRE version 2.0 (Elekta AB, Stockholm, Sweden), which applies a non-linear matching method with block-wise normalized-sum-of-squared-differences metric.

MidP-CT reconstruction

Per patient, a MidP-CT was reconstructed based on DIR, in line with the approach introduced for lung cancer RT [232]. First, the 20% phase was deformably registered with each of the other nine phases, which yielded nine deformation vector fields (DVFs). These DVFs consisted of the

(6)

999999999

Ta b le 9. 1: P at ie nt in fo rm at io n an d r es p ira tio n-in d uc ed m ot io n am p lit ud e of s ub -v ol um es o f t he C TV . P atie n t Sex A ge T um o r loc atio n an d ty pe V olum e of C T V [cm 3] R espir atio n -in duc ed m otio n amplitude of C T V [mm ] P ro xim al es oph agus M iddle es oph agus Di st al es oph agus C ar di a L R -L L R -R A P -A A P -P C C L R -L L R -R A P -A A P -P C C L R -L L R -R A P -A A P -P C C L R -L L R -R A P -A A P -P C C 1 72 M D is tal A C 104.88 1.5 1.2 1.7 0.4 2.3 0.8 0.7 1.5 0.6 3.1 3.1 1.7 3.1 2.4 4.2 -2 60 F D is tal A C 110.53 -2.0 3.0 2.2 0.5 5.5 2.1 1.5 3.4 2.1 10.9 3 55 M D is tal A C 360.46 1.3 3.0 2.1 0.4 3.8 1.1 1.3 1.3 0.4 6.9 1.6 1.8 2.8 0.6 7.4 2.4 1.7 5.9 1.0 8.3 4 69 M D is tal A C 156.35 -2.2 1.7 2.6 0.5 5.8 1.3 1.7 4.7 1.4 7.6 5 61 F M idd le S C C 56.62 -1.4 0.8 1.6 0.7 3.3 2.7 1.4 3.0 1.2 6.3 -6 76 M D is tal A C 148.39 -1.7 1.6 2.9 1.7 5.2 1.7 1.4 4.0 3.1 7.2 7 48 M D is tal A C 140.61 -2.8 1.7 3.2 0.4 4.8 2.6 1.7 3.6 2.9 5.7 8 46 M D is tal A C 111.11 -2.3 2.7 3.4 1.4 8.0 2.3 2.4 4.6 2.1 11.0 9 84 M D is tal S C E C 100.63 -2.5 3.1 3.1 0.9 7.9 3.6 1.9 5.1 0.6 8.6 3.2 2.4 7.1 6.7 19.5 10 59 M D is tal S C C 41.72 1.1 1.2 1.6 0.9 1.8 1.1 0.9 1.9 1.1 2.9 1.5 2.1 2.4 1.8 3.1 -11 61 M D is tal A C 191.85 -1.5 1.5 2.0 0.5 5.3 3.9 1.9 3.5 0.4 4.6 1.4 1.2 5.3 2.1 5.2 12 55 M D is tal A C 412.55 0.9 2.0 1.4 1.2 1.6 1.0 2.6 2.0 1.1 3.6 2.0 1.4 3.4 1.3 5.3 3.2 1.7 5.6 6.2 6.7 13 75 M D is tal A C 232.42 -3.1 1.9 3.4 0.3 3.9 2.5 1.1 5.8 0.6 4.4 14 67 M D is tal A C 182.50 -3.6 2.4 3.3 1.4 4.9 2.0 2.4 5.1 4.4 4.9 15 62 M D is tal A C 204.68 1.0 2.9 2.7 0.4 4.8 1.9 2.4 1.9 0.4 6.6 2.4 4.3 2.8 0.5 7.9 3.8 2.1 4.5 4.8 8.4 A b br ev ia tion s: M = m ale ,F = fe m ale; A C = ade no ca rc inom a, S C C = squa mo u s ce ll ca rc inom a, S C E C = sm al lc el le sop h ag eal ca rc inom a; C T V = cl inical ta rg et vo lume; L R -L = left –r ig h t amp litude at the mo st left side ,L R -R = left –r ig h t amp litude at the mo st ri gh t side ,A P -A = an te rior –p o st er ior amp litude at the mo st an te rior side ,A P -P = an te rior –p o st er ior amp litude at the mo st p o st er ior side ,C C = cr anial –ca udal.

137

(7)

3D voxel displacements from the 20% phase to the other nine phases of the 4D-CT. Next, the nine DVFs were averaged to calculate a mean DVF indicating the 3D voxel displacement from the 20% phase to a time-averaged MidP. Subsequently, 10 DVFs from the MidP to all 10 phases were calculated (Fig. 9.1a). Using these 10 DVFs, the 10 phases of the 4D-CT were warped and then averaged in intensity, resulting in the MidP-CT (Fig. 9.1b). The MidP-CT reconstruction was done using in-house developed software based on the Insight Toolkit (ITK version 4.12).

4D-CT MidP-CT 0% 30% 80% 30% 60% 90% 10% 70% MidP 50% 40% 80% 20% 0% 30% 60% 90% 10% 70% MidP 50% 40% 80% 20% 0% =mean( ) = + (a) =inverse( ) 0% 30% 80% MidP-CT DVF (b) DVFs P A R L Cranial Caudal (c) LR amplitude CC amplitude AP amplitude CTV CTV sub-volume

at the most left side = mean( ) at the most right side = mean( ) LR amplitude:

at the most anterior side = mean( ) at the most posterior side = mean( ) AP amplitude:

CC amplitude = mean( ) CT transverse slice

Fig. 9.1: (a) Illustration of mid-position (MidP) reconstruction based on the deformation vector ﬁelds (DVFs) derived

from deformable image registration between the 20% phase and the other nine phases of four-dimensional (4D)

com-puted tomography (CT).(b) Example of DVFs from MidP to three phases and those phases of 4D-CT overlaid on the

MidP-CT.(c) Schematic drawing of quantifying the respiration-induced target motion amplitudes in the left–right (LR),

anterior–posterior (AP), and cranial–caudal (CC) directions for a sub-volume of the clinical target volume (CTV).

Based on these DIRs, the CTV and OARs delineations on the 20% phase were thereby auto-matically propagated to the other nine phases and the MidP-CT. The envelope of all 10 CTVs on the 10-phase 4D-CT was taken to yield the ITV on the MidP-CT (Monaco version 5.19; Elekta AB).

Margin recipe

The CTV-to-PTV and ITV-to-PTV margins were calculated based on the margin recipe proposed in [125,137] to ensure that 95% of the prescribed dose is received by 90% of the population. The PTV for esophageal cancer RT is located in a mixed anatomical environment (i.e., mix of lung and soft tissue); it is however complicated to embed different penumbras in one margin recipe.

(8)

999999999

Instead, we applied the simplified formula: 2.5Σ + 0.7σ, where Σ and σ are the square roots of the quadratic sum of systematic and random errors, respectively. Assuming online cone-beam CT (CBCT)-based bony anatomy setup verification, for the ITV strategy, the systematic and ran-dom errors consisted of the delineation variation (Σdelineation) and the interfractional target

posi-tion variaposi-tion relative to bony anatomy, i.e., the residual errors after the rigid CBCT-based bony anatomy registration (Σinterand σinter). For the MidP strategy, in addition to Σdelineation, Σinter, and

σinter, we included the RTM as a random error (σRTM). The rotational setup errors were neglected

in the present study.

Since the delineation variation, interfractional target position variation, and RTM were found to be most pronounced in the CC direction and in the distal esophagus [176,210,214,231], we applied an anisotropic and region-specific (proximal, middle, distal esophagus, and cardia) mar-gin. For this, the CTV/ITV was divided into 2–4 sub-volumes depending on how many regions were covered by the CTV/ITV (example: Fig. 9.2). An anisotropic region-specific margin was ap-plied in each sub-volume to construct a regional PTV (Fig. 9.3). By taking the union of all regional PTVs, the PTV with region-specific margins was then constructed (Monaco).

Fig. 9.2: Illustration of the four sub-volumes of the clinical target volume (CTV) in patient 12 on one sagittal mid-position

computed tomography slice.

Although neither the ITV nor the MidP strategy requires an implementation of markers, we assumed that markers were implanted in all patients to use a reduced delineation uncertainty in the margin recipe, based on an earlier study quantifying the inter-observer delineation variation [231]. We set Σdelineationat 3.0 mm and 5.0 mm in the CC direction for the most cranially and caudally

located sub-volumes, respectively, and 1.6 mm in the left–right (LR) and anterior–posterior (AP) directions. For Σinterand σinter, values from an earlier study [176] were applied (Table 9.2).

For the MidP strategy, the σRTMis approximated by 0.358 times the RTM amplitude [125].

To this end, we used the DVFs derived from the DIR between the 20% phase and the other nine phases to quantify the RTM amplitude of each voxel on the CTV surface. Per CTV sub-volume, the amplitude of RTM was calculated as illustrated in Fig. 9.1c.

(9)

PTVMidP-RTM PTVITV-only PTVMidP-full PTVITV-full

0 5 10 15 20 0 5 1 0 1 5 2 0 2 5 3 0 LR 0 5 10 15 20 0 5 1 0 1 5 2 0 2 5 3 0 CC 0 5 10 15 20 0 5 1 0 1 5 2 0 2 5 3 0 AP P ro x im a l e s o p h a g u s 0 5 10 15 20 0 5 1 0 1 5 2 0 2 5 3 0 0 5 10 15 20 0 5 1 0 1 5 2 0 2 5 3 0 0 5 10 15 20 0 5 1 0 1 5 2 0 2 5 3 0 M id d le e s o p h a g u s 0 5 10 15 20 0 5 1 0 1 5 2 0 2 5 3 0 0 5 10 15 20 0 5 1 0 1 5 2 0 2 5 3 0 0 5 10 15 20 0 5 1 0 1 5 2 0 2 5 3 0 D is ta l e s o p h a g u s 0 5 10 15 20 0 5 1 0 1 5 2 0 2 5 3 0 0 5 10 15 20 0 5 1 0 1 5 2 0 2 5 3 0 0 5 10 15 20 0 5 1 0 1 5 2 0 2 5 3 0 C a rd ia

Respiration-induced motion amplitude [mm]

P T V m a rg in [ m m ]

Fig. 9.3: Planning target volume (PTV) margin versus respiration-induced target motion (RTM) amplitude in the left–right

(LR), cranial–caudal (CC), and anterior–posterior (AP) directions and in four regions in the esophagus for four diﬀerent

PTVs: incorporating the RTM uncertainties only, using the mid-position (MidP) strategy (PTVMidP-RTM) and the internal

target volume (ITV) strategy (PTVITV-only); incorporating all region-speciﬁc and anisotropic geometrical uncertainties,

using the MidP strategy (PTVMidP-full) and the ITV strategy (PTVITV-full).

PTV determination and treatment planning

For the MidP and ITV strategy, the respective PTVs were denoted as PTVMidP-fulland PTVITV-full,

in which the margin included all above specified uncertainties (Fig. 9.4). However, these two PTVs were only suitable for comparison of the dose to OARs because no other uncertainties but the RTM was simulated in the present study. Consequently, for the CTV-coverage evaluation we applied another two PTVs per patient: PTVMidP-RTM, PTVITV-only(Fig. 9.4). The PTVMidP-RTM

was determined by extending the CTV with a margin incorporating σRTMonly; the PTVITV-only

was identical to the ITV. These two PTVs were only used to investigate the CTV coverage for both MidP and ITV strategies.

Based on the four PTVs, four plans with prescribed dose of 41.4 Gy in 23 fractions were cor-respondingly made on the MidP-CT with a 6-MV single-arc (356°) volumetric-modulated arc therapy (VMAT) technique (Oncentra VMAT version 4.3; Elekta AB). The planned doses were

(10)

999999999

Table 9.2: Systematic and random errors (i.e., Σinterand σinter) of the interfractional target position variation relative to

bony anatomy used in the margin recipe.

Left–Right [mm] Cranial–Caudal [mm] Anterior–Posterior [mm]

Σinter σinter Σinter σinter Σinter σinter

Proximal esophagus 1.5 1.3 4.1 1.5 1.9 1.2

Middle esophagus 3.1 1.5 2.9 2.0 3.2 2.3

Distal esophagus 1.9 1.9 4.2 2.5 1.9 1.4

Cardia 5.4 4.3 4.9 3.2 1.9 2.4

Four plans made on MidP-CT

PTVMidP-RTM PTVITV-only PTVMidP-full PTVITV-full

4D-CT = 10 × 3D-CT 4D-CT = 10 × 3D-CT 4D-CT = 10 × 3D-CT 4D-CT = 10 × 3D-CT 4D-CT = 10 × 3D-CT 4D-CT = 10 × 3D-CT 4D-CT = 10 × 3D-CT 4D-CT = 10 × 3D-CT 4D-CT = 10 × 3D-CT 4D-CT = 10×3D-CT MidP-CT (3D) Applied to Planned doses PTV MidP-R TM PTV ITV -only PTV MidP-full PTV ITV -full Calculated 4D-doses PTVMidP-RTM PTVITV-only PTVMidP-full PTVITV-full 4D-CT = 10 × 3D-CT 4D-CT = 10 × 3D-CT 4D-CT = 10 × 3D-CT 4D-CT = 10 × 3D-CT 4D-CT = 10 × 3D-CT 4D-CT = 10 × 3D-CT 4D-CT = 10 × 3D-CT 4D-CT = 10 × 3D-CT 4D-CT = 10 × 3D-CT 10×3D-DVF Warped 4D-doses PTVMidP-RTM PTVITV-only PTVMidP-full PTVITV-full Accumulated 4D-doses PTVMidP-RTM PTVITV-only PTVMidP-full PTVITV-full CTV CTV CTV CTV

Margin for respiration-induced target motion (RTM) using mid-position

(MidP) strategy

Internal target volume (ITV) Margin for delineation

variation, interfractional position variation, and RTM using MidP strategy

Margin for delineation variation and interfractional position variation using ITV strategy

Dose comparison CTV Dose comparison OARs

Fig. 9.4: Schematic view of the four planning target volumes (PTVs) and calculation of the planned dose and

accu-mulated four-dimensional (4D) dose based on the deformation vector ﬁelds (DVFs). Abbreviations: CT = computed tomography; 3D = three-dimensional; CTV = clinical target volume, OARs = organs at risk.

calculated using the collapsed cone algorithm with a dose grid size of 2 mm. For comparison be-tween the different plans, the D99%(minimum dose to 99% of the volume) in the PTV was

nor-malized to 95% of the prescribed dose. The planning objectives in dose-volume histogram (DVH) parameters are listed in Table 9.3.

4D-dose calculation and accumulation

To simulate the RTM-induced blurred dose distribution, for each of the four plans we recalculated the dose on all 10 phases of the 4D-CT (Oncentra). For each plan, the resulting 10 3D-doses (i.e.,

(11)

Table 9.3: Planning objectives.

Region of interest Dose-volume histogram parameter Constraint

PTV V95% Volume [% of the total volume] receiving⩾ 95% of the prescribed dose, i.e., 39.33 Gy =99%*

D2% Near-maximum dose <107%

V10Gy Volume [% of the total volume] receiving⩾ 10 Gy <50%

Lungs V20Gy Volume [% of the total volume] receiving⩾ 20 Gy <30%

Dmean Mean dose [Gy] <16 Gy

Left/right kidney V18Gy Volume [% of the total volume] receiving⩾ 18 Gy <33%

Spinard cord D2cm3 Minimum dose [Gy] to the most irradiated 2 cm3volume <50 Gy

Heart V30Gy Volume [% of the total volume] receiving⩾ 30 Gy <30%

Liver Dmean Mean dose [Gy] <26 Gy

Abbreviation: PTV = planning target volume.

*_{Note: this is because the D}

99%in the PTV was normalized to 95% of the prescribed dose (i.e., 39.33 Gy) for dose comparison

between different plans.

4D-dose) were warped to the MidP and averaged using the 10 DVFs (i.e., DVFs from the MidP to the 10 phases, which were generated for the MidP-CT reconstruction) to yield a mean 4D-dose (ITK-based in-house software), which was referred to as the accumulated 4D-dose (Fig. 9.4).

Dose comparison

The 3%/3-mm gamma analysis, which is commonly used for patient specific quality assurance, was performed to compare the planned dose with the accumulated 4D-dose to the whole patient (3D Slicer version 4.8) [233,234]. The corresponding pass rate (percentage of γ>1, i.e., dose difference <3% of the prescribed dose or distance to agreement <3 mm) was calculated to inves-tigate its correlation with RTM amplitude using linear regression. Moreover, the planned dose was compared with the accumulated 4D-dose in terms of the DVH parameters for either the CTV or OARs depending on the plan.

To compare accumulated 4D-doses between the MidP and ITV strategies, the 3%/3-mm gam-ma analysis was performed and the voxel-to-voxel dose difference between the two accumulated 4D-doses was also calculated. Moreover, for both planned doses and accumulated 4D-doses, the DVHs of the CTV were compared between the plans based on PTVMidP-RTMand PTVITV-only;

the DVHs of the OARs were compared between the plans based on PTVMidP-fulland PTVITV-full

(Fig. 9.4). The statistical significance of the differences was tested with the Wilcoxon signed-rank test (significance level at 0.05). All statistical analyses were performed in R [144].

Deformable image registration validation

The propagated delineations of the CTV and OARs were visually inspected by an experienced ra-diation oncology resident and an experienced rara-diation therapist, respectively. For five patients

(12)

999999999

(patients 1, 2, 3, 10, and 13), the propagated CTVs were compared with the manually delineated CTVs on the nine phases of 4D-CT in terms of the Dice coefficient and the average distance be-tween the vertices of the propagated CTV mesh and the triangular faces of the manually delineated CTV mesh (i.e., average vertex-to-face distance). The manual delineation of the CTVs was done independently by the radiation oncology resident (i.e., without the presence of the propagated CTVs). The five patients were selected to create a group of diversity in tumor extent, location, and RTM. In addition, the geometrical accuracy of DIR was validated using the lung 4D-CT dataset and manually labelled landmarks, according to the guideline of American Association of Physics in Medicine (AAPM) Radiation Therapy Committee Task Group No. 132 [235].

9.3 Results

Table 9.1 shows for each patient which of the four esophageal regions were covered by the CTV and gives per covered region the RTM amplitude of the CTV in the LR, CC, and AP directions. The mean± standard deviation (SD) of RTM amplitudes in the CC direction was 4.7 ± 2.0 mm and 7.6±4.0mmforthe mostcraniallyandcaudally locatedsub-volumesof theCTV,respectively.

PTVMidP-RTMvolumes were smaller than PTVITV-onlyvolumes: mean± SD reduction of 6.5

± 3.5% for PTVMidP-RTMcompared to PTVITV-only. Compared to the PTVITV-full, the volume of

PTVMidP-fullwas reduced by 12.0± 2.2%.

A linearly decreasing relationship (R2⩾0.82) was found between the RTM amplitude of the caudal sub-volume of the CTV and the overall passing rate of 3%/3-mm gamma analysis which represented the difference between the planned dose and accumulated 4D-dose (Fig. 9.5). Mostly, the pass rate was high (>98%) when the RTM amplitude was⩽11.0 mm. The differences between the planned doses and accumulated 4D-doses, which did not meet the 3%/3-mm tolerance, were mostly located in the caudal regions of the PTVs where the RTM was most pronounced (Fig. 9.5). Although the V95%of the planned dose in the CTV was always >98% for both strategies (Table

2), for the accumulated 4D-dose that was only not the case for patient 9 when using the PTVMidP-RTM-based plan (V95%= 95.7%). The region receiving a dose below 95% of the prescribed dose

(i.e., 39.33 Gy) was mainly located in the caudal part of the CTV (Fig. 9.6). The reason for this could be that patient 9 had an average respiration-induced motion amplitude of the cardiac sub-volume of the CTV of 19.5 mm in the CC direction (Table 9.1), resulting in an up to 10-mm difference in the CC direction between the most caudal slices of PTVITV-onlyand PTVMidP-RTM.

Overall, the V95%, Dmean, and D2%of the accumulated 4D-doses in the CTV were, albeit

signif-icantly, only <1% smaller for the PTVMidP-RTM-based plan compared to the PTVITV-only-based

plan (p⩽0.001, Table 2). Moreover, the pass rate (mean ± SD) of the 3%/3-mm gamma analysis comparing the PTVMidP-RTM-based and PTVITV-only-based accumulated 4D-doses was 85± 13%.

For the OARs, there was no significant difference in DVH parameters between the accumulated 4D-dose and planned dose. For both planned dose and accumulated 4D-dose, the DVH

(13)

Ta b le 9.4 : P la nn in g o b je ct iv es . D VH p ar am et er R efe re n ce va lue P la nn ed dos e A ccum ul at ed 4D -dos e M idP IT V Diffe re n ce p -v alue M idP IT V Diffe re n ce p -v alue PTVMidP-RTM -based or PTVITV-only -based plan V 95% [%] > 98% 99.7 ± 0.2 99.9 ± 0.1 0.2 ± 0.2 0.015 99.0 ± 1.1 99.9 ± 0.1 0.9 ± 1.1 < 0.001 C T V D me an [%] -98.9 ± 0.8 99.2 ± 0.7 0.4 ± 0.4 0.003 98.7 ± 0.9 99.2 ± 0.7 0.5 ± 0.3 < 0.001 D 2% [%] < 107% 100.7 ± 1.0 101.2 ± 0.9 0.5 ± 0.6 0.004 100.4 ± 1.1 100.9 ± 0.9 0.5 ± 0.5 0.001 P T V V 95% [%] -99.0 ± 0.0 99.0 ± 0.0 0.0 ± 0.0 -95.8 ± 3.7 96.3 ± 2.8 0.5 ± 0.9 0.015

PTVMidP-full-based or PTVITV-full-based plan

V 95% [%] > 98% 100.0 ± 0.0 100.0 ± 0.0 0.0 ± 0.0 0.789 100.0 ± 0.0 100.0 ± 0.0 0.0 ± 0.0 0.584 C T V D me an [%] -101.3 ± 0.7 101.5 ± 1.0 0.2 ± 0.8 0.303 101.2 ± 0.9 101.3 ± 1.1 0.2 ± 0.9 0.359 D 2% [%] < 107% 103.2 ± 1.0 103.3 ± 1.2 0.0 ± 1.1 0.599 103.1 ± 0.9 103.1 ± 1.2 0.0 ± 1.0 0.421 P T V V 95% [%] -99.0 ± 0.0 99.0 ± 0.0 0.0 ± 0.0 -97.6 ± 1.3 97.6 ± 1.2 0.1 ± 0.4 0.561 V 10G y [%] < 50% 44.4 ± 19.4 48.4 ± 19.8 4.1 ± 2.1 < 0.001 44.3 ± 19.6 48.3 ± 19.9 4.0 ± 2.0 < 0.001 Lun gs V 20G y [%] < 30% 9.8 ± 6.8 11.2 ± 7.6 1.4 ± 0.9 < 0.001 9.8 ± 7.0 11.2 ± 7.7 1.4 ± 1.0 < 0.001 D me an [G y] < 16 G y 10.0 ± 3.3 10.6 ± 3.4 0.6 ± 0.2 < 0.001 10.0 ± 3.3 10.6 ± 3.4 0.6 ± 0.2 < 0.001 H ea rt V 30G y [%] < 30% 15.9 ± 5.0 17.8 ± 5.4 2.0 ± 1.1 < 0.001 15.9 ± 5.2 17.9 ± 5.5 1.9 ± 1.1 < 0.001 L iv er D me an [G y] < 26 G y 12.8 ± 5.5 13.5 ± 5.7 0.8 ± 0.5 < 0.001 12.8 ± 5.5 13.6 ± 5.6 0.8 ± 0.5 < 0.001 L eft kidne y V 18G y [%] < 33% 1.5 ± 2.8 2.0 ± 3.6 0.5 ± 0.9 0.018 1.5 ± 2.5 2.0 ± 3.4 0.5 ± 0.9 0.030 R ig h t kidne y V 18G y [%] < 33% 0.2 ± 0.6 0.3 ± 1.2 0.1 ± 0.6 1.000 0.1 ± 0.5 0.3 ± 1.0 0.1 ± 0.5 1.000 S p in al cor d D 2cm 3 [G y] < 50 G y 25.3 ± 3.3 25.4 ± 2.6 0.1 ± 1.4 0.934 25.3 ± 3.3 25.4 ± 2.6 0.1 ± 1.4 0.934 A b br ev ia tion s: D VH = do se-vo lume hi sto gr am ; M idP = mid -p o sit ion, IT V = in te rn al ta rg et vo lume; C T V = clinical ta rg et vo lume , P T V = p la nnin g ta rg et vo lume; R T M = re sp ir at ion-ind uc ed ta rg et mot ion . N ot e: The P T V V 95% of the p la nne d do se w as al w ays 99% b eca u se the p la nne d do se w as nor m al iz ed for al lca se s. The P T V V 95% of the ac cum ula te d 4D -do se w as not u se d for ev al ua tion d ue to the sim ula tion of R T M. The ref or e, no ref er enc e val ue w as giv en . The lis t of D VH p ar ame te rs for the C T V u sin g the P T V M idP -ful l -or P T V IT V -ful l_-b as ed p la n is not in te nde d for C T V co ve ra ge ev al ua tion . It is onl y u se d to ve rif y the fair comp ar is on on the D VH p ar ame te rs in the or ga n s at risk b et w ee n the M idP and IT V str at eg ie s.

144

(14)

999999999

95 96 97 98 99 100 0 4 8 12 16 20 R2_{= 0.83} PTVMidP-RTM 95 96 97 98 99 100 0 4 8 12 16 20 R2_{= 0.86} PTVITV-only 95 96 97 98 99 100 0 4 8 12 16 20 R2_{= 0.84} PTVMidP-full 95 96 97 98 99 100 0 4 8 12 16 20 R2_{= 0.82} PTVITV-full Amplitude [mm] Pa ss r at e [% ] γ>1: Accumulated 4D-dose — Planned dose

P2: Pass rate = 99.05% P9: Pass rate = 97.03%

PTVMidP-RTM

CTV PTVITV-only

γ

PTVMidP-RTM-based plan PTVITV-only-based plan

(a)

(b)

Fig. 9.5: (a) Correlation (R2: coeﬃcient of determination) between the respiration-induced target motion (RTM)

am-plitude at the most caudally located sub-volume of the clinical target volume (CTV) and the pass rate of the gamma analysis comparing the planned dose and accumulated four-dimensional (4D) dose based on all four planning target

volumes (PTVs) using mid-position (MidP) and internal target volume (ITV) strategies.(b) Region of γ>1 comparing

the planned dose and accumulated 4D-dose for patients 2 (P2) and 9 (P9).

ters in the lungs, heart, and liver were significantly reduced by approximately 10% on average, with a mean dose reduction up to 1.0 Gy, 1.7 Gy, and 1.9 Gy, respectively, using the PTVMidP-full-based

plan compared with the PTVITV-full-based plan (p<0.001, Table 9.4). For the kidneys and spinal

cord, no such significant difference was observed. The pass rate (mean± SD) of the 3%/3-mm gamma analysis comparing the PTVMidP-full-based and PTVITV-full-based accumulated 4D-doses

was 71± 14%. Moreover, larger dose differences (>4.0 Gy) between the two strategies were mostly found in the most cranially and caudally located regions of the PTVs (e.g., in the trachea and intestines) (Fig. 9.7).

The automatically propagated delineations of the CTV and OARs were approved by our radi-ation oncology resident and radiradi-ation therapist, respectively. Comparing the propagated CTVs with the manually delineated CTVs, the mean± SD(range) of Dice coefficients were 0.92 ± 0.02(0.84–0.97); the mean±SD(range)oftheaveragevertex-to-facedistanceswere1.2±0.4(0.5–

(15)

Fig. 9.6: Isodose lines of accumulated four-dimensional dose calculated based on the planning target volume (PTV)

with margin incorporating the respiration-induced target motion (RTM) uncertainty only (PTVMidP-RTM) on two coronal

slices of mid-position (MidP) computed tomography for patient 9. The clinical target volume (CTV) and internal target

volume (ITV, i.e., PTVITV-only) are also plotted.

2.4) mm. Based on the approach recommended by the AAPM Task Group No. 132, the mean± SD(maximum) of the absolute errors of the DIR using ADMIRE was 0.5± 0.5(6.7) mm, 0.8 ± 0.9(7.6) mm, and 0.6± 0.9(12.4) mm, in the LR, CC, and AP direction, respectively. Only in four out of 300 landmarks in the superior and inferior lobes of the lungs we found the absolute errors

>5.0 mm in at least one direction.

9.4 Discussion

This study is the first to apply the MidP strategy and to compare it with the ITV strategy in treat-ment planning of esophageal cancer RT. Further, for the first time an anisotropic and region-specific safety margin for esophageal cancer RT is proposed and successfully applied. Compared to the ITV strategy, the MidP strategy ensures adequate CTV coverage and a statistically signifi-cant reduction in the dose to the most relevant OARs.

Considering other published results on DIR accuracy [236], intra-observer variation in manual delineation, and the voxel spacing in the 4D-CT, the DIR algorithm in ADMIRE is affirmed to be highly accurate in the target and the lungs for our patient group based on the validation. The visual approval of the automatically propagated delineations by both the radiation oncology resident and radiation therapist indicated that the performance of DIRs in these regions was trustworthy.

For patients with an RTM amplitude⩽11.0 mm, the difference between the planned dose and accumulated 4D-dose was limited irrespective of the MidP or ITV strategy. This implies that for RT planning of esophageal cancer in most patients, a 4D-dose accumulation using DIR is not nec-essary. Moreover, we chose to use the MidP-CT for treatment planning, because it has a better image quality than the AIP-CT and single-phase scan [232] for the purpose of a voxel-to-voxel comparison between the planned dose and accumulated 4D-dose. Although the blurred AIP-CT might be more commonly used for dose calculation due to breathing-induced density variation,

(16)

999999999

Heart (H) Liver (Lr) Lungs (Ls) Left kidney (LK) Right kidney (RK) Spinal cord (S) CTV (C)

ITV MidP ► ◄ ◄ ◄ 0 10 20 30 40 0 20 40 60 80 100 Volume [%] P2 C H Lr Ls LK S RK ► ◄ ◄ ◄ 0 10 20 30 40 P9 C H Lr Ls LK S RK ► ◄ ◄ ◄ 0 10 20 30 40 P12 C H Lr Ls LK S RK Dose [Gy] Dose diﬀerence [%]: Dose diﬀerence [Gy]:

PTVMidP-full CTV PTVITV-full -20% -8.28 -15%-6.21 -10%-4.14 -2.07-5% 2.075% 10%4.14 15% 6.21 20%8.28 P2 P9 P12 Isodose: PTV ITV -full -based − PTV MidP-full

-based accumulated 4D-dose

P2: Pass rate = 50.27% P9: Pass rate = 71.05% P12: Pass rate = 68.01%

γ>1: PTV ITV -full -based − PTV MidP-full

-based accumulated 4D-dose _PTV

MidP-full CTV PTVITV-full γ (a) (b) (c)

Fig. 9.7: Example of accumulated four-dimensional (4D) dose for patient 2 (P2), 9 (P9), and 12 (P12). (a) Dose-volume

histogram for clinical target volume (CTV) and organs at risk with reference value of CTV V95%(▶), lung V10Gyand V20Gy

(◀), and heart V30Gy(◀). (b) Region of γ>1 comparing the accumulated 4D-doses using the mid-position (MidP)

and internal target volume (ITV) strategies based on planning target volumes (PTVs) incorporating all uncertainties (PTVMidP-fulland PTVITV-full).(c) Dose diﬀerence between the PTVMidP-full- and PTVITV-full-based accumulated 4D-doses.

our 4D-dose accumulation results imply a high accuracy of using the MidP-CT for treatment plan-ning. This is in line with the findings for lung (SB)RT [237–240].

The DIR-based quantification of the RTM amplitude confirmed that the RTM amplitude dif-fers per region of the esophagus as shown in previous studies using rigid registration on fidu-cial markers [165,210,214]. This also provided an alternative approach to quantify the patient-specific and region- and direction-dependent RTM amplitude when no fiducial markers are avail-able. Moreover, combined with the dependence of interfractional tumor position variation on direction and region [176], this signifies the necessity of using an anisotropic and region-specific PTV margin.

(17)

Clinically, when the patient is freely breathing and online bony anatomy-based setup verifica-tion is used during the treatment, the full PTV margin should be used, in which the RTM is one component [125,137]. In this study, we only simulated the presence of RTM and ignored the presence of other uncertainties such as interfractional tumor position variation relative to bony anatomy. However, since the full PTV margin was designed to cover all the uncertainties, the CTV coverage is expected to also be adequate if all the uncertainties were simulated with the aid of DIR and daily CBCTs. In this case, the dose to OARs might slightly deviate from the results in the present study, possibly dominated by the interfractional tumor position variation relative to bony anatomy as shown in liver SBRT [241]. For the present study, further dosimetric eval-uation on all the uncertainties, however, would not change the qualitative results regarding the comparison between the two strategies accounting for RTM.

Due to the quadratic sum of all random errors in the PTVMidP-fullmargin calculation, using the

PTVMidP-RTMoverestimated the influence of RTM (Fig. 9.8). For patients with an RTM

ampli-tude⩽11.0 mm, this overestimate was ⩽1.7 mm, which is negligible compared to the CT slice thickness of 2.5 mm and the dose grid size of 2.0 mm. Although we could use the PTVMidP-full

-based plan to evaluate the target coverage -based on a full PTV excluding the RTM uncertainty to avoid this overestimate, the inevitable inclusion of lung tissue in such a target can result in an underestimate of target coverage due to the dose attenuation. Therefore, the evaluation of target coverage based on the CTV is appropriate in the present study.

The full margin including delineation variation, interfractional position variation, and respiration-induced target motion (RTM)

2.5 Σ delineation 2 + Σinter2 + 0.7 σinter2 + σRTM2

The contribution by RTM in the full margin

0.7 σ inter 2 + σRTM2 − 0.7σ int er

The margin including RTM only

0.7σ RTM

The overestimate of RTM contribution

0.7 σ RTM− σ'() *+ 2 _{+ σ} RTM 2 − σ '()*+

(

)

    CTV CTV PTVMidP-full PTVMidP-RTM Delineation + interfractional position variation

Fig. 9.8: Illustration of the overestimation of the respiration-induced target motion (RTM) contribution for the

mid-position (MidP) strategy when the planning target volume (PTV) includes the RTM only, i.e., PTVMidP-RTM, compared to

the RTM contribution in the actual full margin including other uncertainties, i.e., PTVMidP-full. Σdelineationis the systematic

delineation variation; Σinteris the systematic component of the interfractional target position variation; σinteris the

random component of the interfractional target position variation; σRTMis the random error induced by RTM, which is

0.358 times the RTM amplitude. Abbreviation: CTV = clinical target volume.

Albeit many studies showed that the DVH parameters can predict the risk of a particular end-point (e.g., pneumonitis and cardiotoxicity), it is difficult to conclude the clinical benefits in num-bers due to the different predictive values among different population groups and endpoints [50–

(18)

999999999

52,54–57]. However, it is fair to state that a dose reduction to OARs can result in a lower incidence of certain clinical endpoints. For the heart toxicity, it was found that rates of major coronary events increased linearly with the mean dose to the heart by 7.4% per Gy [242]. Consequently, the sta-tistically significant reductions in dose to the lungs, heart, and liver as found in the present study for the MidP strategy compared to the ITV strategy might be considered potentially clinically rel-evant.

Apart from the OARs, more pronounced dose reductions were observed in the upper abdomen (Fig. 9.7), resulting from the larger PTV margin in the caudal part using the ITV strategy compared with the MidP strategy. Although the dose in this region is normally far below the tolerance, it is preferred to keep the dose as low as possible to reduce the risk of toxicity [243].

Only in one patient (patient 9) we found a slightly inadequate CTV coverage, which could be due to the extremely large RTM amplitude (19.5 mm) in the CC direction. However, the approx-imation of σRTMby 0.358 times the RTM amplitude was designed under the condition of RTM

amplitude <10 mm; this is true for the majority of patients because previous studies have shown that the most pronounced RTM, i.e., at the distal esophagus and GEJ, is on average 5–8 mm. Con-sidering that this margin recipe was initially designed to aim for 90% of the population, the MidP strategy will be safe for the majority of patients. For patients with an extremely large RTM ampli-tude, a different margin recipe, i.e., calculating the σRTMby 0.45 times the RTM amplitude, might

work [125]. However, robustness of the treatment plan always needs to be confirmed in clinical practice.

Although the conventional ITV strategy seems easier to be implemented in the clinical prac-tice, the MidP strategy overall does not add more work compared to the ITV strategy when au-tomatic DIR and MidP-CT reconstruction are available. Further, the improved image quality of the MidP-CT may reduce the delineation uncertainty in both target and OARs, thereby reducing the CTV-to-PTV margin. Thus, clinical implementation of the MidP strategy is feasible and favor-able. However, no matter which strategy is used in clinical practice, the most time-consuming step is to generate a region-specific margin since so far there is no other easier approach than manually dividing the CTV/ITV into different sub-volumes. A more elegant approach might be generat-ing regional random errors directly based on the DVFs for an automatic region-dependent margin expansion. In clinical practice, a more efficient approach to generate the region-specific PTV is needful.

Alternative to the MidP strategy, the use of a breath-hold technique, especially inhalation breath hold, could also be appealing to use as motion management strategy in esophageal cancer RT be-cause of the reduced lung dose compared with free-breathing treatment [194,195]. The reduc-tions in mean dose to the lung were <3 Gy for a total dose of 50–60 Gy. However, the breath-holding time is much shorter than the treatment time, resulting in several breath-breath-holding sessions in each treatment fraction. This will inevitably increase the treatment time. Moreover, esophageal cancer patients often suffer from pulmonary complications, which can pose a challenge to perform

(19)

a breath-hold. In addition, there could be substantial motion and inter-breath-holding position variation in the upper abdomen, especially for the inhalation breath-holds [131,132]. Hence, the feasibility of implementing breath-hold in the clinical practice of esophageal cancer RT requires further research.

In conclusion, using a MidP strategy in treatment planning of esophageal cancer RT is beneficial for patients with moderate RTM in terms of an approximately 10% dose reduction in the OARs without compromising the target coverage. When an accurate DIR software is available in clinical practice, we recommend using the MidP strategy with anisotropic and region-specific margins in treatment planning of esophageal cancer RT.