Skip to main content
NCBI home page
Physics and Imaging in Radiation Oncology logoLink to Physics and Imaging in Radiation Oncology
. 2026 Mar 1;38:100939. doi: 10.1016/j.phro.2026.100939

Reduction of organ of interest dose in proton therapy for oesophageal cancer through optimized setup robustness settings and online adaptation

Eva van Weerd a,⁎, Anne Lisa Wolf a, Steven JM Habraken a,b, Yvonne LB Klaver a,b, Mischa S Hoogeman a,c
PMCID: PMC12969121  PMID: 41810038

Graphical abstract

graphic file with name ga1.jpg

Open in a new tab

Keywords: Adaptive proton therapy, Intensity modulated proton therapy, Esophageal cancer, Anatomical changes, Organs of interest, Treatment planning

Highlights

  • •

    Adequate target coverage was achieved in 80% of patients with 5 mm setup robustness.

  • •

    Reduced setup robustness resulted in significant organ of interest dose reduction.

  • •

    Mean heart dose decreased by 2.6 Gy (RBE) when reducing robustness from 8 to 5 mm.

  • •

    Online plan adaptation preserved clinically acceptable target coverage.

  • •

    Online plan adaptation reduced mean heart dose by 3.7 Gy (RBE)

Abstract

Background and purpose

Oesophageal radiotherapy may cause side effects due to high organs of interest (OOIs) doses. While proton therapy (PT) reduces OOI dose compared to photon therapy, setup robustness settings used may limit OOI sparing. This study assessed how optimizing robustness settings for adequate target coverage affected OOI dose, and evaluated the potential of online adaptive PT (OAPT).

Materials and methods

Twenty oesophageal cancer patients treated with intensity-modulated PT were analysed. Plans with 8 (clinical standard), 7, 6 and 5 mm setup robustness settings were generated on the planning 4D-CT (pCT) and robustly evaluated on weekly repeat 4D-CTs (rCTs). OAPT plans with individualized robustness settings were optimized and robustly evaluated on rCTs. Clinical target volume (CTV) V95% and OOI doses were assessed, and two-year mortality estimated using a validated model.

Results

A 5 mm setup robustness setting maintained CTV V95% ≥ 98% for 80% of patients, with 92.4–97.7% in the remaining 20%. With OAPT, V95% ≥ 98% was achieved in 75%, with 97.6–97.9% in the remaining 25%. Mean heart dose decreased from 11.8 Gy (RBE) (Inter Quartile Range (IQR) 10.9–13.9) to 9.2 Gy (RBE) (IQR 8.2–11.6) when reducing settings from 8 to 5 mm. OAPT provided a reduction of 3.7 Gy (RBE), reducing two-year mortality by 6.9% compared to the 8 mm setting.

Conclusions

Reducing setup robustness settings to 5 mm was feasible for most patients and reduced OOI dose. Large anatomical changes required plan adaptation. OAPT allowed further OOI dose reductions while preserving acceptable target coverage.

1. Introduction

With over 500,000 new cases in recent years, oesophageal cancer (OC) is the 11th most common cancer worldwide, and its incidence is rising [1], [2]. Treatment typically consists of chemotherapy or chemoradiation followed by surgery, or chemoradiation alone [3], [4], [5].

Radiotherapy can cause acute and late side effects, including radiation pneumonitis and cardiac complications, due to high doses to organs of interest (OOIs) [6], [7]. Proton therapy (PT) can reduce these side effects by sparing surrounding OOIs due to its favourable depth-dose distribution compared to photon therapy [8], [9], [10], [11]. However, PT for OC is challenging due to its sensitivity to density changes in the beam path caused by respiratory and cardiac motion, anatomical variations, and interplay effects between motion and scanned beam delivery [12], [13]. These uncertainties are typically managed using active or passive motion management strategies combined with appropriate setup robustness settings [13], [14]. Previous studies indicate that an 8 mm setup robustness setting generally ensures adequate target coverage [15], [16], [17]. However, the feasibility of reducing these settings has not yet been investigated, although smaller robustness settings may lower OOI doses and potentially reduce side effects.

Online adaptive proton therapy (OAPT) may enable even smaller robustness settings while maintaining target coverage, by allowing daily adaptation to interfractional changes [18], [19]. Herbst et al. demonstrated this potential in OC patients, showing that reducing setup robustness from 7 to 2 mm did not increase the need for manual adaptation [20]. In addition, OAPT may reduce increases in mean heart dose (MHD) observed over the course of treatment [15], [16].

This study aimed to determine the minimum setup robustness settings required in our clinical practice for adequate target coverage, to assess their impact on OOI doses, and to evaluate whether combining an alternative evaluation approach with OAPT could further reduce OOI doses in OC patients treated with intensity-modulated proton therapy (IMPT).

2. Materials and methods

2.1. Patient data

This retrospective analysis included 20 patients with mid- or distal oesophageal adenocarcinoma or squamous cell carcinoma. Patients were treated with IMPT at Holland Proton Therapy Centre (HollandPTC) between October 2021 and March 2022, and provided written informed consent. The local institutional review board waived the need for formal review (approval number P18.053). Patient characteristics are summarized in Table 1.

Table 1.

Demographic and clinical characteristics of the included patients.

Sex
 Male, n (%) 19 (95)
 Female, n (%) 1 (5)
Age, years, median (range) 67 (52–79)
Histology, n (%)
 Adenocarcinoma 17 (85)
 Squamous cell carcinoma 3 (15)
TNM classification, n (%)
 T1N0M0 1 (5)
 T2NxM0 1 (5)
 T2N0M0 1 (5)
 T2N1M0 1 (5)
 T3N0M0 4 (20)
 T3N1M0 6 (30)
 T3N2M0 6 (30)
Location primary tumour, n (%)
 Mid 3 (15)
 Distal 16 (80)
 Mid/Distal 1 (5)
CTV volume, cc, median (range) 257 (113–572)
Amplitude, mm, median (range) 10 (5–20)
Neoadjuvant therapy 15 (75%)
Definitive chemoradiotherapy 5 (25%)
Open in a new tab

n = number of patients.

2.2. Imaging and registration

Non-contrast planning 4D-computed tomography (pCT) and weekly repeat 4D-CTs (rCTs) (SOMATOM Definition Edge, Siemens Healthineers, Germany) were acquired, following our standard protocol. Four to five rCTs were acquired per patient, depending on the fractionation schedule. In five patients only three rCTs were available due to discontinuation of routine rCT or cancellation of the final rCT. The pCTs were rigidly registered to each rCT based on bony anatomy, followed by deformable image registration (DIR) using the ANACONDA algorithm in RayStation (version 10b, RaySearch Laboratories, Sweden).

2.3. Delineation and treatment planning

Gross tumour volume (GTV) and clinical target volume (CTV) were delineated on the 20% breathing phase of the pCT, following Thomas et al. [21]. Contours were deformably propagated to the ten breathing phases and the average CT (avgCT) of the pCT and the rCTs and reviewed by an experienced radiation therapist (RTT). An internal clinical target volume (ITV) was generated, for optimization purposes.

Eighty plans were generated in RayStation (v10b) for ProBeam 4.0 (Varian Medical Systems, a Siemens Healthineers company, USA), according to the clinical protocol, with setup robustness settings ranging from 8 (clinical standard) to 5 mm.

This study explored settings to a lower limit of 5 mm based on Hawkins et al., who described this as the minimal margin for photon radiotherapy with CBCT guidance for this group of patients [22]. In RayStation v10b, setup robustness settings larger than 5 mm generate additional shift scenarios for each direction, substantially increasing computation time. To limit this, a 5 mm setup robustness setting was combined with various isotropic ITV expansions (3, 2, 1 and 0 mm). Throughout this article the term ‘setup robustness setting’ refers to the combination of setup robustness and ITV expansion. Robust optimization incorporated full inhale (Fin) and full exhale (Fex) breathing phases, and a 3% range uncertainty, based on institution-specific range measurements.

Plans were optimized on the avgCT including posterior and right oblique beams, with an optional left oblique beam if plan robustness was insufficient. Automatic spot and energy layer spacing was applied and adjusted per patient to remain within machine constraints. Optimizations used 60 iterations, with spots below 3 MU removed after 20 iterations. Dose was calculated using Monte Carlo v5.1 (3 mm3 grid).

To mitigate interplay effects, a range shifter was used to increase spot size, and four-layered repainting was applied for tumour motion larger than 5 mm. Tumour motion was derived from DIR vector fields between Fin and Fex, using CTV as focus region of interest. The 90th percentile of the maximum motion was calculated in three directions. Patients with a CTV motion amplitude exceeding 20 mm are deemed ineligible for PT and were not included in this study. The 4D-robust evaluation, comprising 28 scenarios, was conducted using the full setup robustness setting, following the Dutch consensus [14], and performed on both Fin and Fex [13].

To evaluate plans over the treatment course, nominal plans were calculated, and 4D-robust evaluation was applied on all rCTs with a setup robustness setting of 2 mm and a range uncertainty of 3%. The 2 mm setting accounts for residual errors after online setup correction [23]. For these plans, no plan adaptations were performed, as they were not considered necessary in clinical practice for this patient cohort.

2.4. Online adaptive

OAPT is not yet clinically available at our institute and was therefore simulated using existing clinical software. It was assumed no 4D-CT would be acquired during treatment due to time constraints, additional daily dose and practical feasibility. Therefore, intrafraction respiratory tumour motion was incorporated into the setup robustness settings, and 3D-robust evaluation was performed on the avgCT, including 28 scenarios, according to institutional clinical practice. Respiratory-induced tumour motion was quantified on the 4D-pCT following the clinical protocol (section 2.3). These values were incorporated as random error for the margin calculation, as 0.4 times the 90th percentile of the maximum motion in three directions [24].

Residual errors in the treatment chain and DIR-related uncertainties were also included in the margin calculation. Residual errors were based on institution-specific quality assurance measurements, and DIR uncertainties were derived from Weistrand et al. [25] (Supplementary Material, Table S1). Evaluation metrics of DIR performance in this patient cohort were not explicitly evaluated as part of this study. The quantified uncertainties were translated into setup robustness settings using the van Herk margin recipe [26], resulting in individualized anisotropic setup robustness settings applied for each patient. Across the study population, the mean values were 4 mm (right-left), 5 mm (anterior-posterior) and 6 mm (cranio-caudal). Range uncertainty and interplay mitigation strategies were applied according to the clinical protocol (section 2.3).

Treatment plans were robustly optimized on the avgCT of the pCT to determine optimization objectives which served as baseline plans for reoptimization on the avgCT of each consecutive rCT, following a strategy similar to Winkel et al. [27]. The weight of the CTV minimum dose objective was increased by a factor of 1.5, based on preliminary testing on a subset of rCTs, to improve target coverage while avoiding excessive dose to surrounding OOIs. Optimization involved determining spot positions, followed by two optimization rounds without modifying the objectives, to ensure plan convergence. The same setup robustness settings used for optimization were used for 3D-robust evaluation, including DIR-related delineation uncertainties, as the strict time constraints of an OAPT workflow limit the feasibility of thoroughly reviewing and adapting deformable contours [28].

2.5. Plan evaluation

Target and OOI dose guidance was applied according to the PROTECT-trial (Supplementary Material, Table S2) [29]. Target coverage was considered adequate if CTV V95% ≥ 98% in the voxel_wise minimum (Vmin). Dose to parallel organs was evaluated in the nominal plan, and body and spinal cord dose were evaluated in the voxel-wise maximum (Vmax). Patients received neoadjuvant chemoradiotherapy (41.4 Gray (Gy) (RBE1) in 23 fractions, CROSS regimen) or definitive chemoradiotherapy (50.4 Gy (RBE) in 28 fractions), with concurrent carboplatin and paclitaxel [4], [30]. Two-year mortality was estimated using a validated NTCP-model for MHD, as employed for patient selection in the Netherlands [31], [32].

2.6. Data analysis

For each patient, target coverage and OOI doses were averaged across all rCTs. Results for the patient cohort were reported as median values with interquartile range (IQR). Differences in OOI doses between plans with setup robustness settings of 8 and 5 mm, and 8 mm and OAPT were tested using the Wilcoxon signed-rank test, a non-parametric test for paired data (SPSS v27, IBM). P < 0.05 was considered statistically significant.

3. Results

Both planning strategies met CTV V95% ≥ 98%, body and spinal cord constraints in the robustness evaluation on the pCT.

For the 8 mm setup robustness setting, CTV V95% ≥ 98% was achieved in 73/78 rCTs, with four rCTs falling below 95% coverage but none below 90%. For the 5 mm setting, coverage was achieved in 62/78 rCTs, with seven rCTs below 95% and two below 90%. Mean CTV V95% ≥ 98% across all rCTs was maintained in 90% (18) of patients with 8 mm and 80% (16) with 5 mm, with lowest individual V95% values of 95.1% and 92.4%, respectively (Fig. 1). Two patients failed to meet mean CTV V95% ≥ 98% for any setting. For patients 1 and 19, the 5 mm setting provided better coverage than 6 mm, likely due to large anatomical changes and beam-specific dose distribution rather than the robustness setting itself.

Fig. 1.

Fig. 1

Open in a new tab

CTV coverage across robustness settings. Mean CTV coverage (V95% (%)) in Vmin for Fin and Fex for various robustness settings and OAPT. The black line represents the V95% ≥ 98% dose constraint. Closed symbols represent mean coverage in Fin, open symbols represent mean coverage in Fex. Crosses indicate mean coverage for OAPT in the Vmin on the avgCT. Mean coverage values were calculated per patient as the mean across all rCTs.

In OAPT plans, 58/78 rCTs achieved CTV V95% ≥ 98%, with none of the rCTs below 95%. Mean CTV V95% was maintained in 75% (15) of patients with a lowest individual value of 97.6%. Body and spinal cord constraints were met for all rCTs in both strategies.

Reducing setup robustness settings from 8 to 5 mm decreased median MHD from 11.8 Gy (RBE) (IQR 10.9–13.9) to 9.2 Gy (RBE) (IQR 8.2–11.6) (Fig. 2). Based on these values, the median estimated two-year mortality was reduced by 3.5%. Largest reductions were seen in low-dose metrics: median lung V5Gy decreased from 23.1% (IQR 17.5–40.6) to 18.0% (IQR 13.6–33.8), and median heart V25Gy from 21.4% (IQR 20.2–26.3) to 16.2% (IQR 13.6–20.4). OAPT resulted in a median MHD of 8.1 Gy (RBE) (IQR 7.4–9.2), lung V5Gy of 16.6% (IQR 11.2–29.6), and heart V25Gy of 13.7% (IQR 11.3–15.2), representing median reductions of 3.7 Gy (RBE), 6.5%, and 7.7%, respectively, compared with the 8 mm plans. The decrease in MHD corresponded to a median reduction in estimated two-year mortality of 6.9%. Smaller differences were observed for liver, spleen, and kidneys, due to their greater distance from the target (Supplementary Material, Fig. S1). OOI doses differed significantly between the 8 mm and 5 mm settings, and between the 8 mm setting and OAPT (p < 0.05).

Fig. 2.

Fig. 2

Open in a new tab

Distribution of dose metrics for OOIs. Boxplots showing the distribution of different dose metrics. The central line represents the median, the boxes indicate the interquartile range (IQR), and the whiskers extend to the minimum and maximum values within 1.5 × IQR. Outliers are displayed as circles. The y-axis represents either dose in Gy (RBE) or volume in percentage, depending on the metric: MHD (Gy (RBE)), Heart V40Gy (%), Heart V25Gy (%), MLD (Gy (RBE)), Lungs V20Gy (%), and Lungs V5Gy (%).

Heart and lung doses generally decreased progressively with lower robustness settings, with OAPT providing additional reduction (Fig. 3). In three cases, OAPT slightly increased MHD (n = 2) or MLD (n = 2) compared with the reduced robustness settings, with one patient showing increases in both metrics. Similar patterns were observed for heart V25Gy and lungs V5Gy (Supplementary Material, Fig. S2).

Fig. 3.

Fig. 3

Open in a new tab

Mean heart dose and mean lung dose per patient. MHD (Gy (RBE)) and MLD (Gy (RBE)) per patient for different setup robustness settings and OAPT. Mean dose values were calculated per patient as the mean across all rCTs.

When comparing MHD on the pCT to the mean dose over the treatment course, plans following the clinical protocol showed greater variability than OAPT (Fig. 4). In 35% of patients, MHD increased by more than 1 Gy (RBE), with a maximum of 3.5 Gy (RBE). The highest increase in estimated two-year mortality was 4.1%. In contrast, OAPT kept MHD stable, closely aligning pCT calculations. For MLD, differences between strategies were minimal. Median lungs V5Gy were comparable, but the lower mean for OAPT reflected a subgroup of patients with reduced lungs V5Gy exposure during treatment.

Fig. 4.

Fig. 4

Open in a new tab

Differences in OOI dose over the treatment course compared to the planning CT. Differences in OOI dose (Gy (RBE)) or volume (%) over the treatment course (Tx) relative to the pCT, between the plans following the clinical protocol and OAPT. All plans were recalculated on the avgCT of the rCTs, with pCT dose also calculated on the avgCT. Boxplots show differences in MHD (Gy (RBE)), MLD (Gy (RBE)), and lungs V5Gy (%). The horizontal line within each box represents the median, x the mean, box edges the interquartile range, and whiskers extend to 1.5 times the interquartile range. Outliers are displayed as individual points. Positive values indicate an increase in dose, and negative values a decrease in dose compared to pCT.

4. Discussion

We evaluated the minimum setup robustness setting required for adequate target coverage in IMPT for OC and assessed the impact on OOI dose for both the clinical protocol and OAPT. Reducing the robustness setting from 8 to 5 mm significantly improved OOI sparing while preserving adequate CTV coverage in 80% of patients without plan adaptation. In the remaining patients, clinically relevant coverage loss was observed on individual rCTs, with reductions up to 12%. OAPT restored or substantially improved target coverage in these patients, resulting in clinically acceptable coverage in all cases, generally with additional or comparable OOI sparing.

The clinical 8 mm robustness setting achieved mean CTV V95% ≥ 98% in 90% of patients, consistent with previous studies [15], [17]. OAPT maintained coverage in 15 (75%) patients. In the remaining coverage was close to 98% (lowest 97.6%), which would be considered clinically acceptable. To assess the impact of this underdosage on OOI dose, these plans were normalized to 98% target coverage. The results show that this normalization had a negligible effect on the measured OOI doses (Supplementary Material, Table S3). The limited underdosages observed were expected to result from anatomical changes creating imbalances in the objective list. Patients with coverage below the criterion should be monitored, and adjustments to the objective list may be required. Although the weight of the CTV minimum-dose objective was increased for OAPT, no corresponding OOI dose increase was observed. Potential effects on OOIs remain a consideration for clinical implementation.

Smaller robustness settings significantly reduced OOI doses, particularly for lower heart and lung dose metrics associated with pneumonia and cardiopulmonary side effects [8], [11]. OAPT provided additional OOI sparing beyond what could be achieved by reducing setup robustness settings alone. OAPT corrected for interfractional OOI position changes, resulting in more consistent MHD throughout treatment and contributing to a lower estimated two-year mortality, consistent with findings reported by Visser et al. for offline-adapted IMPT plans [16].

To ensure clinical applicability, different planning strategies were used for the clinical protocol and OAPT plans, representing current workflow and a feasible OAPT approach. In the OAPT plans, tumour motion was incorporated into the robustness settings. This method was considered valid, as ITV-based optimization and 4D-robust evaluation are considered conservative and may overestimate target motion, potentially limiting OOI sparing with OAPT [33]. Applying this method to clinical protocol plans in a subset of patients (n = 5), reduced OOI doses even without daily adaptation (Supplementary Material, Fig. S3). However, in one patient, a baseline shift of the diaphragm resulted in insufficient target coverage, indicating that this approach alone may be inadequate to ensure adequate target dose. In contrast, in the OAPT plan, target coverage was restored through adaptation, while achieving an additional reduction in OOI dose. In the clinical plan, target coverage was maintained through 4D robust optimization and evaluation (Supplementary Material, Fig. S4).

OAPT optimization required 60–90 min per plan, the main bottleneck for clinical implementation. Fast optimization and dose calculation are essential to minimize the impact of intrafraction changes, as tumour drift, which was not explicitly considered in this study [34]. Several studies highlighted the need for advancements in automation to enable clinical OAPT implementation for OC patients [20], [35], [36]. Several strategies to reduce OAPT calculation time, have shown promising results [37], [38], [39].

A limitation of this study is the use of weekly rCTs, which captures anatomical changes over treatment but not daily variations. Although CT images provide high-quality data for OAPT workflows, only a limited number of PT centres have in-room CT available. Cone Beam Computed Tomography (CBCT) is commonly used for patient positioning, but currently lacks sufficient image quality for proton dose calculation. Ongoing research aims to enable CBCT-based OAPT in the future [40], [41].

Dose accumulation was performed by averaging over repeat CTs, representing a pragmatic approach [42]. Although DIR has been reported in the literature for dose accumulation, it is associated with uncertainties in regions with low soft-tissue contrast. Previous work has demonstrated that both deformable and non-deformable accumulation strategies introduce uncertainties, and that the true delivered dose is likely to lie between these two approaches [42].

We used a setup robustness setting combined with ITV expansion during optimization, but full settings were used for robust evaluation. Optimization without ITV expansion might yield slightly better results, but differences are expected to be small.

Setup robustness settings below 5 mm were not investigated, and further research is needed to assess feasibility considering residual errors, inter- and intrafraction motion. Notably, Herbst et al. reported using a 2 mm setup robustness setting for OAPT in oesophageal cancer, without observing an increase in the manual adaptation rate compared with offline adapted plans [20]. In OAPT, lowering DIR uncertainty could further reduce setup robustness settings. Recent studies suggest that deep learning–based approaches can enhance DIR performance [43], and combining DIR with deep learning segmentation may yield more accurate contours [36], [44]. Additionally, minimizing motion could reduce robustness settings. Notably, the motion incorporated in the OAPT margin recipe was based on the pCT, while patient breathing amplitude may vary during treatment.

Clinically used interplay mitigation strategies were applied, supporting the clinical relevance of our findings, although the interplay effect was not explicitly assessed.

While reducing setup robustness settings improved OOI sparing, plan adaptation remained essential in three patients with large anatomical changes, potentially increasing OOI doses.

Our findings may contribute to identifying which patients may benefit from OAPT. Despite significant dose reductions, the clinical benefit remains unclear and the additional treatment time required must be weighed against expected reduction of side effects [45], [46].

Finally, the NTCP model used for patient selection in the Netherlands has limitations. It was validated on photon data and lacks some key prognostic factors [47]. Therefore, we focused on reporting OOI dose differences as primary outcomes.

In conclusion, this study demonstrated that reducing setup robustness settings is feasible for most OC patients treated with IMPT. In cases with large anatomical changes, offline or online adaptation may be necessary to maintain adequate target coverage. Smaller setup robustness settings significantly reduced OOI doses. Combining reduced setup robustness with OAPT and 3D-robust evaluation allowed further OOI dose reductions compared with margin reduction alone, while maintaining clinically acceptable target coverage.

Declaration of Generative AI and AI-assisted technologies in the writing process

During the preparation of this work the author used ChatGPT in order to improve readability and language. After using this tool, the author reviewed and edited the content as needed and take full responsibility for the content of the publication.

CRediT authorship contribution statement

Eva van Weerd: Writing – original draft, Visualization, Validation, Project administration, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Anne Lisa Wolf: Writing – review & editing, Validation, Supervision, Software, Resources, Methodology, Conceptualization. Steven J.M. Habraken: Writing – review & editing, Methodology. Yvonne L.B. Klaver: Writing – review & editing, Methodology. Mischa S. Hoogeman: Writing – review & editing, Supervision, Methodology, Conceptualization.

Declaration of competing interest

The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: Steven Habraken reports a relationship with Varian Medical Systems Inc that includes: funding grants. Steven Habraken reports a relationship with Dutch Research Council that includes: funding grants. Steven Habraken reports a relationship with RaySearch Laboratories AB that includes: funding grants. Steven Habraken reports a relationship with Elekta AB that includes: funding grants. Mischa Hoogeman reports a relationship with Siemens Healthineers AG that includes: funding grants. Mischa Hoogeman reports a relationship with RaySearch Laboratories AB that includes: funding grants. Mischa Hoogeman reports a relationship with Elekta AB that includes: funding grants. Mischa Hoogeman reports a relationship with Varian Medical Systems Inc that includes: funding grants. Mischa Hoogeman reports a relationship with Accuray Inc that includes: consulting or advisory and funding grants. Co-author is member of the scientific council and chair of the working group clinical investigations of the Varian Flash Forward Consortium − YK If there are other authors, they declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A

Supplementary data to this article can be found online at https://doi.org/10.1016/j.phro.2026.100939.

1

All doses are reported as Gy (RBE), assuming a relative biological effectiveness (RBE) of 1.1 in accordance with ICRU 93 guidelines.

Contributor Information

Eva van Weerd, Email: e.van.weerd@hollandptc.nl.

Anne Lisa Wolf, Email: a.wolf@hollandptc.nl.

Steven J.M. Habraken, Email: s.j.m.habraken@lumc.nl.

Yvonne L.B. Klaver, Email: y.klaver@hollandptc.nl.

Mischa S. Hoogeman, Email: m.hoogeman@hollandptc.nl.

Appendix A. Supplementary data

The following are the Supplementary data to this article:

Supplementary Data 1
mmc1.pdf (414.6KB, pdf)

References

  • 1.World Cancer Research Fund. Oesophageal cancer statistics, https://www.wcrf.org/preventing-cancer/cancer-statistics/oesophageal-cancer-statistics/#oesophageal-cancer-deaths/; 2022 [accessed 14 February 2025].
  • 2.Liu C.Q., Ma Y.L., Qin Q., Wang P.H., Luo Y., Xu P.F., et al. Epidemiology of esophageal cancer in 2020 and projections to 2030 and 2040. Thorac Cancer. 2023;14:3–11. doi: 10.1111/1759-7714.14745. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 3.Hoeppner J., Brunner T., Schmoor C., Bronsert P., Kulemann B., Claus R., et al. Perioperative chemotherapy or preoperative chemoradiotherapy in esophageal cancer. N Engl J Med. 2025;392:323–335. doi: 10.1056/NEJMoa2409408. [DOI] [PubMed] [Google Scholar]
  • 4.Shapiro J., van Lanschot J.J.B., Hulshof M.C.C.M., van Hagen P., van Berge Henegouwen M.I., Wijnhoven B.P.L., et al. Neoadjuvant chemoradiotherapy plus surgery versus surgery alone for oesophageal or junctional cancer (CROSS): long-term results of a randomised controlled trial. Lancet Oncol. 2015;16:1090–1108. doi: 10.1016/S1470-2045(15)00040-6. [DOI] [PubMed] [Google Scholar]
  • 5.Jensen G.L., Hammonds K.P., Haque W. Neoadjuvant versus definitive chemoradiation in locally advanced esophageal cancer for patients of advanced age or significant comorbidities. Dis Esophagus. 2023;36 doi: 10.1093/dote/doac050. [DOI] [PubMed] [Google Scholar]
  • 6.Thomas M., Defraene G., Lambrecht M., Deng W., Moons J., Nafteux P., et al. NTCP model for postoperative complications and one-year mortality after trimodality treatment in oesophageal cancer. Radiother Oncol. 2019;141:33–40. doi: 10.1016/j.radonc.2019.09.015. [DOI] [PubMed] [Google Scholar]
  • 7.Chuong M.D., Hallemeier C.L., Li H., Ronald Zhu X., Zhang X., Tryggestad E.J., et al. Executive summary of clinical and technical guidelines for esophageal cancer proton beam therapy from the particle therapy co-operative group thoracic and gastrointestinal subcommittees. Front Oncol. 2021;11 doi: 10.3389/fonc.2021.748331. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8.Lin S.H., Hobbs B.P., Verma V., Tidwell R.S., Smith G.L., Lei X., et al. Randomized phase IIB trial of proton beam therapy versus intensity-modulated radiation therapy for locally advanced esophageal cancer. J Clin Oncol. 2020;38:1569–1579. doi: 10.1200/JCO.19.02503. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 9.Zhou P., Du Y., Zhang Y., Zhu M., Li T., Tian W., et al. Efficacy and safety in proton therapy and photon therapy for patients with esophageal cancer: a meta-analysis. JAMA Netw Open. 2023;6 doi: 10.1001/jamanetworkopen.2023.28136. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 10.Lin S.H., Merrell K.W., Shen J., Verma V., Correa A.M., Wang L., et al. Multi-institutional analysis of radiation modality use and postoperative outcomes of neoadjuvant chemoradiation for esophageal cancer. Radiother Oncol. 2017;123:376–381. doi: 10.1016/j.radonc.2017.04.013. [DOI] [PubMed] [Google Scholar]
  • 11.Frederiks M.L., van Etten B., Kelder W., Dieters M., Beukema J.C., Ijsbrandy C., et al. Proton radiotherapy significantly reduces pneumonia in patients with esophageal cancer. Int J Radiat Oncol Biol Phys. 2025;122:313–324. doi: 10.1016/j.ijrobp.2024.12.036. [DOI] [PubMed] [Google Scholar]
  • 12.Grassberger C., Dowdell S., Lomax A., Sharp G., Shackleford J., Choi N., et al. Motion interplay as a function of patient parameters and spot size in spot scanning proton therapy for lung cancer. Int J Radiat Oncol Biol Phys. 2013;86:380–386. doi: 10.1016/j.ijrobp.2013.01.024. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 13.Tryggestad E.J., Liu W., Pepin M.D., Hallemeier C.L., Sio T.T. Managing treatment-related uncertainties in proton beam radiotherapy for gastrointestinal cancers. J Gastrointest Oncol. 2020;11:212–224. doi: 10.21037/jgo.2019.11.07. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 14.Korevaar E.W., Habraken S.J.M., Scandurra D., Kierkels R.G.J., Unipan M., Eenink M.G.C., et al. Practical robustness evaluation in radiotherapy – a photon and proton-proof alternative to PTV-based plan evaluation. Radiother Oncol. 2019;141:267–274. doi: 10.1016/j.radonc.2019.08.005. [DOI] [PubMed] [Google Scholar]
  • 15.Anakotta R.M., van der Laan H.P., Visser S., Ribeiro C.O., Dieters M., Langendijk J.A., et al. Weekly robustness evaluation of intensity-modulated proton therapy for oesophageal cancer. Radiother Oncol. 2020;151:66–72. doi: 10.1016/j.radonc.2020.07.015. [DOI] [PubMed] [Google Scholar]
  • 16.Visser SO, Ribeiro C, Dieters M, Mul VE, Niezink AGH, van der Schaaf A, et al. Robustness assessment of clinical adaptive proton and photon radiotherapy for oesophageal cancer in the model-based approach. Radiother Oncol 2022;177:197–204. doi: 10.1016/j.radonc.2022.11.001. [DOI] [PubMed]
  • 17.Canters R., van der Klugt; K., Taasti V.T., Buijsen J., Ta B., Steenbakkers I., et al. Robustness of intensity modulated proton treatment of esophageal cancer for anatomical changes and breathing motion. Radiother Oncol. 2024;198 doi: 10.1016/j.radonc.2024.110409. [DOI] [PubMed] [Google Scholar]
  • 18.Eijkelenkamp H., Boekhoff M.R., Verweij M.E., Peters F.P., Meijer G.J., Intven M.P.W. Planning target volume margin assessment for online adaptive MR-guided dose-escalation in rectal cancer on a 1.5 T MR-Linac. Radiother Oncol. 2021;162:150–155. doi: 10.1016/j.radonc.2021.07.011. [DOI] [PubMed] [Google Scholar]
  • 19.Sheng Y., Li T., Lee W.R., Yin F.F., Wu Q.J. Exploring the margin recipe for online adaptive radiation therapy for intermediate-risk prostate cancer: an intrafractional seminal vesicles motion analysis. Int J Radiat Oncol Biol Phys. 2017;98:473–480. doi: 10.1016/j.ijrobp.2017.02.089. [DOI] [PubMed] [Google Scholar]
  • 20.Herbst P., Draguet C., Barragán-Montero A.M., Borderías-Villarroel E., Chocan Vera M., Populaire P., et al. Potential of automated online adaptive proton therapy to reduce margins for oesophageal cancer. Phys Imaging Radiat Oncol. 2025;33 doi: 10.1016/j.phro.2025.100712. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 21.Thomas M., Mortensen H., Hoffmann L., Møller D., Troost E., Muijs C., et al. Proposal for the delineation of neoadjuvant target volumes in oesophageal cancer. Radiother Oncol. 2020;156:102–112. doi: 10.1016/j.radonc.2020.02.018. [DOI] [PubMed] [Google Scholar]
  • 22.Hawkins M.A., Aitken A., Hansen V.N., McNair H.A., Tait D.M. Set-up errors in radiotherapy for oesophageal cancers - is electronic portal imaging or conebeam more accurate? Radiother Oncol. 2011;98:249–254. doi: 10.1016/j.radonc.2010.11.002. [DOI] [PubMed] [Google Scholar]
  • 23.Ribeiro C.O., Visser S., Korevaar E.W., Sijtsema N.M., Anakotta R.M., Dieters M., et al. Towards the clinical implementation of intensity-modulated proton therapy for thoracic indications with moderate motion: Robust optimised plan evaluation by means of patient and machine specific information. Radiother Oncol. 2021;157:210–218. doi: 10.1016/j.radonc.2021.01.014. [DOI] [PubMed] [Google Scholar]
  • 24.Engelsman M., Sharp G.C., Bortfeld T., Onimaru R., Shirato H. How much margin reduction is possible through gating or breath hold? Phys Med Biol. 2005;50:477–490. doi: 10.1088/0031-9155/50/3/006. [DOI] [PubMed] [Google Scholar]
  • 25.Weistrand O., Svensson S. The ANACONDA algorithm for deformable image registration in radiotherapy. Med Phys. 2015;42 doi: 10.1118/1.4894702. [DOI] [PubMed] [Google Scholar]
  • 26.Van Herk M., Remeijer P., Rasch C., Lebesque J.V. The probability of correct target dosage: dose-population histograms for deriving treatment margins in radiotherapy. Int J Radiat Oncol Biol Phys. 2000;47:1121–1135. doi: 10.1016/S0360-3016(00)00518-6. [DOI] [PubMed] [Google Scholar]
  • 27.Winkel D., Bol G.H., Kroon P.S., van Asselen B., Hackett S.S., Werensteijn-Honingh A.M., et al. Adaptive radiotherapy: the Elekta Unity MR-linac concept. Clin Transl Radiat Oncol. 2019;18:54–59. doi: 10.1016/J.CTRO.2019.04.001. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 28.Nenoff L., Buti G., Bobić M., Lalonde A., Nesteruk K.P., Winey B., et al. Integrating structure propagation uncertainties in the optimization of online adaptive proton therapy plans. Cancers. 2022;14:3926. doi: 10.3390/cancers14163926. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 29.Hoffmann L., Mortensen H., Shamshad M., Berbee M., Bizzocchi N., Bütof R., et al. Treatment planning comparison in the PROTECT-trial randomising proton versus photon beam therapy in oesophageal cancer: results from eight European centres. Radiother Oncol. 2022;172:32–41. doi: 10.1016/j.radonc.2022.04.029. [DOI] [PubMed] [Google Scholar]
  • 30.Smit J.K., Muijs C.T., Burgerhof J.G.M., Paardekooper G., Timmer P.R., Muller K., et al. Survival after definitive (chemo)radiotherapy in esophageal cancer patients: a population-based study in the North-East Netherlands. Ann Surg Oncol. 2013;20:1985–1992. doi: 10.1245/s10434-012-2824-2. [DOI] [PubMed] [Google Scholar]
  • 31.Addendum Slokdarmcarcinoom bij Landelijk Indicatie Protocol Protonentherapie Longcarcinoom. 2021.
  • 32.Defraene G., Dankers F.J.W.M., Price G., Schuit E., van Elmpt W., Arredouani S., et al. Multifactorial risk factors for mortality after chemotherapy and radiotherapy for non-small cell lung cancer. Radiother Oncol. 2020;152:117–125. doi: 10.1016/j.radonc.2019.09.005. [DOI] [PubMed] [Google Scholar]
  • 33.Krieger M., Giger A., Salomir R., Bieri O., Celicanin Z., Cattin P.C., et al. Impact of internal target volume definition for pencil beam scanned proton treatment planning in the presence of respiratory motion variability for lung cancer: a proof of concept. Radiother Oncol. 2020;145:154–161. doi: 10.1016/j.radonc.2019.12.001. [DOI] [PubMed] [Google Scholar]
  • 34.Heethuis S.E., Borggreve A.S., Goense L., Van Rossum P.S.N., Mook S., Van Hillegersberg R., et al. Quantification of variations in intra-fraction motion of esophageal tumors over the course of neoadjuvant chemoradiotherapy based on cine-MRI. Phys Med Biol. 2018;63 doi: 10.1088/1361-6560/aacfb5. [DOI] [PubMed] [Google Scholar]
  • 35.Draguet C., Populaire P., Vera M.C., Fredriksson A., Haustermans K., Lee J.A., et al. A comparative study on automatic treatment planning for online adaptive proton therapy of esophageal cancer: which combination of deformable registration and deep learning planning tools performs the best? Phys Med Biol. 2024;69 doi: 10.1088/1361-6560/ad80f6. [DOI] [PubMed] [Google Scholar]
  • 36.Populaire P., Marini B., Poels K., Svensson S., Sterpin E., Fredriksson A., et al. Autodelineation methods in a simulated fully automated proton therapy workflow for esophageal cancer. Phys Imaging Radiat Oncol. 2024;32 doi: 10.1016/j.phro.2024.100646. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 37.Jagt; T.Z., Breedveld S., van Haveren R., Heijmen B.J.M., Hoogeman M.S. Online-adaptive versus robust IMPT for prostate cancer: how much can we gain? Radiother Oncol. 2020;151:228–233. doi: 10.1016/j.radonc.2020.07.054. [DOI] [PubMed] [Google Scholar]
  • 38.Lalonde A., Bobić M., Sharp G.C., Chamseddine I., Winey B., Paganetti H. Evaluating the effect of setup uncertainty reduction and adaptation to geometric changes on normal tissue complication probability using online adaptive head and neck intensity modulated proton therapy. Phys Med Biol. 2023;68 doi: 10.1088/1361-6560/acd433. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 39.Bobić M., Lalonde A., Nesteruk K.P., Lee H., Nenoff L., Gorissen B.L., et al. Large anatomical changes in head-and-neck cancers – a dosimetric comparison of online and offline adaptive proton therapy. Clin Transl Radiat Oncol. 2023;40 doi: 10.1016/j.ctro.2023.100625. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 40.Oud M., Breedveld S., Ng Wei Siang K., Cassetta R., Habraken S., Perkó Z., et al. A planning approach for online adaptive proton therapy to cope with cone beam computed tomography inaccuracies. Phys Imaging Radiat Oncol. 2025;34 doi: 10.1016/j.phro.2025.100752. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 41.Galapon A., Jr, Wagenaar D., Langendijk J.A., Both S. Transfer learning for DL-based Synthetic CT after reconstruction algorithm upgrade in a proton therapy clinic. Med Phys. 2025;52 doi: 10.1002/mp.17937. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 42.Regnery S., Leiner L., Buchele C., Hoegen P., Sandrini E., Held T., et al. Comparison of different dose accumulation strategies to estimate organ doses after stereotactic magnetic resonance-guided adaptive radiotherapy. Radiat Oncol. 2023;18:92. doi: 10.1186/s13014-023-02284-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 43.Smolders A., Lomax A., Weber D.C., Albertini F. Deep learning based uncertainty prediction of deformable image registration for contour propagation and dose accumulation in online adaptive radiotherapy. Phys Med Biol. 2023;68 doi: 10.1088/1361-6560/ad0282. [DOI] [PubMed] [Google Scholar]
  • 44.Nenoff L., Amstutz F., Murr M., Archibald-Heeren B., Fusella M., Hussein M., et al. Review and recommendations on deformable image registration uncertainties for radiotherapy applications. Phys Med Biol. 2023;68 doi: 10.1088/1361-6560/ad0d8a. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 45.Oud M., Breedveld S., Giżyńska M., Chen Y.H., Habraken S., Perkó Z., et al. Dosimetric advantages of adaptive IMPT vs. enhanced workload and treatment time – a need for automation. Radiother Oncol. 2024;201 doi: 10.1016/j.radonc.2024.110548. [DOI] [PubMed] [Google Scholar]
  • 46.Borderías-Villarroel E., Barragán-Montero A., Sterpin E. Time is NTCP: should we maximize patient throughput or perform online adaptation on proton therapy systems? Radiother Oncol. 2024;198 doi: 10.1016/j.radonc.2024.110389. [DOI] [PubMed] [Google Scholar]
  • 47.Berbée M., Muijs C.T., Voncken F.E.M., Wee L.S., Sosef M., van Etten B., et al. External validation of a lung cancer-based prediction model for two-year mortality in esophageal cancer patient cohorts. Radiother Oncol. 2024;190 doi: 10.1016/j.radonc.2023.109979. [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplementary Data 1
mmc1.pdf (414.6KB, pdf)

Articles from Physics and Imaging in Radiation Oncology are provided here courtesy of Elsevier

RESOURCES