Impact of additional PRV constraint to Direct OAR robust optimization in IMPT for targets near serial organs at risk

Abstract

Purpose

In intensity-modulated proton therapy (IMPT), the benefit of adding planning organ-at-risk volume (PRV)-like nominal constraints to direct robust optimization for serial organs at risk remains uncertain when target-OAR separation is minimal. Using paraspinal chordoma as a model, we evaluated whether an additional nominal canal/thecal sac (PRV-like) constraint improves OAR sparing or compromises target coverage.

Methods

Ten patients with paraspinal chordomas were planned using two IMPT strategies: direct cord robust optimization alone (Cord-RO) and direct cord optimization with additional nominal canal/thecal sac constraint (Canal-RO). PTV-based helical tomotherapy (HT) was generated as a secondary benchmark. Prespecified institutional criteria were used for planning, prioritizing OAR constraints over coverage; when standard coverage goals were not achieved, GTV D98 ≥ 59 Gy(RBE) was accepted as the fallback criterion. Endpoints included target coverage, cord/canal doses, and robustness. Statistical comparisons used Wilcoxon signed-rank, Friedman, and Cochran’s Q tests, two-sided, p < 0.05.

Results

All plans met plan-specific OAR constraints. Cord-RO achieved superior target coverage (median HR-CTV D98 [Gy(RBE)]: 63.05 vs. 57.38 vs. 59.46; p ≤ 0.002) and met the fallback objective (GTV D98 ≥59 Gy[RBE]) in 10/10 cases versus 4/10 and 7/10 (p=0.011) in Canal-RO and HT respectively. Robustness favored Cord-RO (median worst-case CTV D95: 90% vs 85%, p=0.018). Nominal cord D0.03cc was similar; worst-case cord D0.03cc was slightly lower with Canal-RO.

Conclusion

When targets abut serial OARs, our results suggest that direct OAR-based robust optimization without additional PRV-like constraints improve target coverage and robustness while maintaining acceptable OAR doses. PRV constraints may be reserved for re-irradiation or when target-OAR separation is adequate.

Introduction

Managing uncertainty is a fundamental challenge in intensity-modulated proton therapy (IMPT), particularly when targets abut serial organs at risk (OARs). CTV-based robust optimization has largely replaced PTV-based approaches for target planning, offering superior coverage under setup and range perturbations [1–3]. However, the optimal strategy for constraining serial OARs adjacent to targets remains uncertain. In clinical practice, serial OARs are typically managed using direct OAR robust optimization, and some workflows additionally apply a nominal PRV-type constraint (e.g., canal/thecal sac) to introduce an explicit geometric buffer at the target–OAR interface. While robust optimization has been extensively evaluated for target volumes, whether adding PRV-type constraints to a directly robustly optimized serial OAR provides meaningful additional protection, or instead compromises target coverage without incremental safety, has not been systematically studied. This question is particularly relevant in settings where even small coverage trade-offs may affect local control.

Paraspinal chordomas provide an ideal clinical model, as targets frequently abut the spinal cord and thecal sac. Durable local control depends on maximal safe resection and high-dose radiotherapy [4–6]; however, en bloc resection is frequently not feasible in the mobile spine [7–9], and metallic instrumentation complicates planning [10, 11]. Target under-dosing is associated with inferior local control [6, 12, 13], making this a clinically consequential setting to test competing optimization strategies.

A recent PTCOG consensus statement also highlighted substantial inter-institutional variability in spinal cord constraint implementation, reflecting the absence of a standardized approach [14]. We therefore used paraspinal chordoma to compare two IMPT strategies that share identical direct cord robust optimization, differing only by the addition of a nominal PRV-like constraint: i) direct spinal cord robust optimization alone (Cord-RO), and (ii) direct spinal cord robust optimization with an additional nominal canal/thecal sac (PRV-like) constraint (Canal-RO). As a secondary comparator, (iii) PTV-optimized, PRV-based helical tomotherapy(HT) plans were generated to contextualize IMPT plan quality against a high-quality photon technique.

Materials and methods

Study design and patient selection

Ten consecutive patients with mobile-spine paraspinal chordoma were included in this dosimetric comparative planning study following institutional review board approval. Nine had undergone prior debulking intralesional surgery with spinal stabilization using metallic instrumentation. In all patients, the tumor was in close proximity to the spinal cord or thecal sac, with the target volume directly abutting them in eight cases.

Simulation and image acquisition

Patients were simulated in site-specific positions: prone with vacuum-bag immobilization for thoracic/lumbar lesions, or supine with thermoplastic mask for cervical lesions. Planning CT was acquired with 3-mm slice thickness and SEMAR reconstruction for implant patients [15]. Planning MRI included volumetric T2-weighted, fat-suppressed, diffusion-weighted, and contrast-enhanced sequences with metal artifact reduction protocols for implant patients [16]. Orthogonal radiographs verified implant geometry, and megavoltage CT was co-registered with the planning CT in artifact-prone regions.

Target and organ-at-risk delineation

Target volumes were delineated by the treating radiation oncologist. The GTV encompassed gross and/or postoperative residual disease. The high-risk CTV (CTV-HR) was a 3-mm geometric GTV expansion modified to respect anatomic barriers; the low-risk CTV (CTV-LR) encompassed at-risk regions per institutional practice, similarly edited for anatomic boundaries. The spinal cord was contoured as neural tissue on fused MRI; CT myelography was used when artifact limited cord visualization. The thecal sac/spinal canal (“canal”) was additionally contoured as a PRV-type avoidance structure for Canal-RO and HT.

Treatment planning strategies

Three plans were generated per patient; dose constraints are detailed in Table 1. The prescription dose was 70.2 Gy(RBE) in 32fractions to the high-risk CTV. When CTV goals were unachievable owing to cord abutment, GTV D98 ≥ 59 Gy(RBE) was applied as the prespecified fallback (Table 1). All IMPT plans used pencil beam scanning and were generated using RayStation version 12 (RaySearch Laboratories, Stockholm, Sweden) with 3–5 beams per plan; beam number and angles were identical between Cord-RO and Canal-RO plans. The maximum dose contribution from any single beam was limited to 60% of the total prescription dose. Both IMPT strategies employed CTV-based minimax robust optimization with ± 2 mm setup uncertainty and range uncertainties of 3.5–4% (4% applied in patients with metallic implants). Robust evaluation used RayStation’s predefined 16-scenario (Eight cuboidal setup uncertainty vertices with 2 range perturbations) setup–range perturbation set (± 2 mm setup; ±3.5%, or ± 4% in implant cases). Worst-case cord dose was additionally evaluated using ± 4 mm setup uncertainty with the same range uncertainty, as a conservative margin relative to the ± 2 mm optimization uncertainty. Worst-case values were the minimum CTV D95 and maximum OAR dose across scenarios; scenario pass rate was the proportion meeting CTV D95 ≥ 95%. HT plans were evaluated under nominal planning conditions only; no equivalent robust uncertainty analysis was applied, consistent with conventional photon PTV-based planning workflows. Spinal cord and canal dose constraints were prioritized over target coverage during optimization.

Table 1.

Planning strategies, target coverage goals and organ-at-risk constraints

Domain	IMPT: Cord-based IMPT (Cord RO)	IMPT: Canal/PRV-based IMPT (Canal RO)	Helical Tomotherapy (HT)
Planning paradigm	CTV-based robust optimization (RO)	CTV-based robust optimization (RO)	PTV-based optimization (no RO)
Intent	Maximize target coverage while respecting cord constraints; control worst-case cord dose	As Cord plan plus additional nominal avoidance using canal/thecal sac	Photon benchmark using PRV-type canal avoidance
Target coverage goals (nominal)	Ideal: CTV D98 ≥ 98%; CTV D95 ≥ 99% Acceptable: CTV D98 ≥ 90%; CTV D95 ≥ 95%	Same as Cord plan	CTV goals tracked: same as IMPT PTV goal: PTV D95 ≥ 95%
If acceptable CTV not met	GTV D98 ≥ 59 Gy(RBE) prioritized	Same as Cord plan	GTV D98 ≥ 59 Gy(RBE) prioritized
Robust evaluation criterion	Setup robustness: 2 mm, 3.5% range; robust CTV D95 ≥ 95% Worst case cord: 4 mm setup and 3.5% range	Same as Cord plan	Not applicable
Spinal cord constraint (nominal)	Cord D0.03cc < 50 Gy(RBE)	Cord D0.03cc < 50 Gy(RBE)	Cord D0.03cc < 50 Gy(RBE)
Spinal cord constraint (robust evaluation)	Worst-case cord D0.03cc < 55 Gy(RBE)	Worst-case cord D0.03cc < 55 Gy(RBE)	Not applicable
Canal/thecal sac use	Contoured; not optimized	Contoured; constrained in nominal scenario only	Used as PRV-type avoidance structure
Canal/thecal sac constraint	None	Canal D0.03cc < 55 Gy(RBE) (nominal)	Canal max < 55 Gy(RBE)

Abbreviations: RO, robust optimization; IMPT, intensity-modulated proton therapy; HT, helical tomotherapy; PRV, planning organ-at-risk volume; CTV, clinical target volume; GTV, gross tumor volume; PTV, planning target volume; D0.03cc, dose to the hottest 0.03 cc; Gy(RBE), gray (relative biological effectiveness)

Cord-RO and Canal-RO used identical direct spinal cord robust optimization settings, except that Canal-RO additionally applied nominal dose constraints to canal (PRV-like structure). HT plans were generated using Precision v3.2 (Accuray, Sunnyvale, CA) with 2-mm isotropic PTV expansions and nominal canal/thecal sac constraints; robust optimization was not employed.

Statistical analysis

All comparisons employed paired analyses. Given the small sample size (n = 10), non-parametric tests were used: pairwise comparisons between Cord-RO and Canal-RO used Wilcoxon signed-rank tests, and three-way comparisons used Friedman test with post-hoc pairwise testing. The binary fallback endpoint [GTV D98 ≥ 59 Gy(RBE)] was assessed using Cochran’s Q test. All tests were two-sided (p < 0.05).

Results

All plans met plan-specific OAR constraints. Cord-RO achieved significantly higher target coverage than Canal-RO and HT (Fig. 1; Table 2), with superior median HR-CTV D98 (63.05 Gy[RBE]; 89.6%), HR-CTV D95 (67.39 Gy[RBE]; 95.7%), GTV D98 (65.12 Gy[RBE]; 92.5%), and LR-CTV D98 (54.08 Gy[RBE]; 96.6%) (Friedman test, p ≤ 0.002 for each metric) (Fig. 2A and B). The fallback objective of GTV D98 ≥ 59 Gy(RBE) was achieved in 10/10 Cord-RO plans versus 7/10 HT and 4/10 Canal-RO plans (Cochran’s Q, p = 0.011).

Fig. 1.

Axial dose color-wash comparison across planning strategies for a representative patient. The Cord-RO plan demonstrates a steep dose gradient overlapping the thecal sac, whereas in the Canal-RO and HT plans the gradient is displaced outside the thecal sac. This reflects the canal/PRV constraint forcing the optimizer to pull dose away from the thecal sac boundary, with a corresponding compromise in target coverage

Table 2.

Dosimetric comparison across plans (Median, IQR)

Parameter	IMPT – Cord plan	IMPT – Canal plan	HT plan	p-value
HR-CTV D98	63.05, 89.6% (59.99–65.03)	57.38, 81.5% (52.99–60.88)	59.46, 84.5% (56.20–61.45)	0.001x
HR-CTV D95	67.39, 95.7% (66.81–67.83)	62.15, 88.3% (60.51–65.08)	64.35, 91.4% (61.98–66.44)	0.001x
GTV D98	65.12, 92.5% (64.21–65.80)	56.35, 80.0% (55.88–61.61)	60.24, 85.6% (59.00–63.05)	0.001 x
LR-CTV D98†	54.08, 96.6% (53.07–54.90)	51.55, 92.1% (49.83–53.03)	53.59, 95.7% (53.33–54.05)	0.002 x
HR-PTV D95	Not applicable	Not applicable	62.25, 88.4% (60.36–64.50)	—
CTV D95 worst-case (%)‡	90% (87–93%)	85% (80–88%)	Not applicable	0.018y
Scenarios passed (%)‡	39% (29–80%)	0% (0–10%)	Not applicable	0.018 y
Spinal cord D0.03 cc (Gy(RBE)) – nominal	47.19 (46.04–48.32)	47.09 (42.88–48.68)	47.13 (42.79–48.98)	0.882 x
Spinal canal D0.03 cc (Gy(RBE)) – nominal	59.34 (57.90–60.65)	53.66 (53.20–54.28)	53.72 (52.56–54.82)	0.002 x
Spinal cord D0.03 cc (Gy(RBE)) – worst-case‡	54.12 (53.08–54.48)	51.69 (50.40–52.78)	Not applicable	0.039 y

† LR-CTV = Low-risk clinical target volume

‡ Worst-case = robust evaluation across setup/range uncertainty scenarios; scenarios passed = proportion of robust scenarios meeting the specified target coverage constraint (CTV D95)

HR-CTV = High-risk clinical target volume; GTV = Gross tumor volume; HR-PTV = High-risk planning target volume; D98/D95 = dose covering 98%/95% of the structure volume; D0.03 cc = maximum dose to 0.03 cm³; Gy(RBE) = gray (relative biological effectiveness); IMPT = intensity-modulated proton therapy; HT = helical tomotherapy; x, Friedman test; y, Wilcoxon signed-rank test

Fig. 2.

Box-and-whisker plots comparing target coverage metrics endpoints across the three planning strategies (Cord-RO, Canal-RO, HT) for all ten patients. (A): HR-CTV D98%, (B) GTV D98%

Robustness metrics favored Cord-RO over Canal-RO (Fig. 3), with higher median worst-case CTV D95% (90% vs. 85%; p = 0.018) and scenario pass rate (39% vs. 0%; p = 0.018); robust evaluation was not applicable for HT. Nominal spinal cord D0.03cc was similar across techniques(~ 47 Gy(RBE); p = 0.882). Worst-case cord D0.03cc was marginally lower with Canal-RO(51.69 vs. 54.12 Gy(RBE); p = 0.039), but both remained within the prespecified robust constraint(< 55 Gy(RBE)) (Fig. 4). Full dosimetric data are summarized in Table 2.

Fig. 3.

(A & B): Scenario-wise DVH robustness evaluation for Cord-RO and Canal-RO. Dose–volume histogram bands for a representative case planned using Cord-RO (A) and Canal-RO (B). Curves represent the nominal plan and evaluated robustness scenarios for the GTV, canal/PRV-like structure, and spinal cord. The vertical dashed line denotes the 59 Gy(RBE) GTV D98 fallback criterion. Cord-RO demonstrates better maintenance of GTV coverage across uncertainty scenarios, whereas Canal-RO results in lower dose to the canal/PRV-like structure and spinal cord, with greater compromise in GTV coverage in this representative abutting-cord geometry

Fig. 4.

Summary comparison of nominal cord D0.03cc, worst-case cord D0.03 cc, and canal D0.03cc across planning strategies, illustrating the divergent dose trade-offs despite equivalent nominal cord doses

Discussion

This study demonstrates that, when targets directly interface with serial OARs and geometric separation is limited, direct cord robust optimization without an additional nominal PRV-like constraint may better preserve target coverage and plan robustness, without compromising nominal OAR doses. Nominal cord D0.03cc was comparable across techniques(~ 47 Gy(RBE)), and worst-case cord D0.03cc, although marginally lower with Canal-RO, remained within the prespecified robust tolerance. Worst-case cord metrics were additionally evaluated using a more conservative ± 4 mm setup uncertainty as a serial-OAR safety check. The modest reduction in worst-case cord dose achieved by Canal-RO represented only an incremental dosimetric gain, but was accompanied by a clinically meaningful loss of target coverage and robustness.

We included HT as a nominal dosimetric reference (PTV/PRV-based) rather than a robustness comparator, since an analogous robust evaluation framework was not applied to HT. Canal-RO IMPT demonstrated target coverage reductions approaching those observed with PTV/PRV-based photon planning, whereas direct cord robust optimization preserved the intrinsic coverage advantage of IMPT. These findings suggest that introducing a nominal PRV-like canal/thecal sac constraint within a robust IMPT framework may reduce the geometric advantage typically associated with IMPT.

Although demonstrated in paraspinal chordoma, this principle is broadly relevant to IMPT scenarios in which targets abut serial OARs. These findings are clinically significant given evidence that target (GTV D98) under-dosing to < 59 Gy(RBE) is associated with inferior local control [6, 12, 13] in these patients, making it a minimum gross-coverage goal.

The observed trade-offs are mechanistically intuitive. Introducing a PRV-type constraint adds an active OAR constraint and increases constraint hierarchy, driving dose away from the OAR boundary in the nominal scenario; this may partly confound the observed coverage loss. When the target directly contacts the OAR, this produces a steeper interface gradient and reduces target coverage and robustness (Fig. 1). These findings align with recognized limitations of the PRV concept in photon planning, where PRV-based constraints have been shown to produce excessively conservative plans when targets abut critical structures [17–19].

Previous studies have demonstrated the superiority of robust optimization over PTV-based approaches for target coverage [1–3]; however the impact of PRV-based OAR constraints within a robust optimization framework has not been addressed. To our knowledge, this is the first study to directly address this question. Our findings support the use of direct OAR-based robust optimization in these scenarios, while PRV constraints may be reserved for cases with adequate target-OAR separation or for re-irradiation settings, where aggressive OAR sparing is paramount.

This study has several limitations. It was a small, single-institution dosimetric analysis with a heterogeneous cohort spanning cervical, thoracic, and lumbar lesions, with varying immobilization strategies and target-OAR abutment. These factors may have influenced achievable dose gradients and limit the generalizability of the findings. In addition, the study is limited to dosimetric endpoints evaluated and does not include clinical outcomes such as toxicity or local control. Validation in larger cohorts and in other clinical settings where targets abut serial OARs, such as near the brainstem or optic apparatus, would strengthen the broader applicability of these findings.

Conclusion

In this paraspinal chordoma cohort, direct OAR robust optimization alone preserved target coverage and plan robustness while maintaining acceptable OAR doses. For IMPT scenarios in which targets abut serial OARs, these findings suggest that an additional nominal PRV-like constraint may not be routinely necessary.

Abbreviations

RO

Robust optimization

IMPT

Intensity-modulated proton therapy

HT

Helical tomotherapy

PRV

Planning organ-at-risk volume

CTV

Clinical target volume

GTV

Gross tumor volume

PTV

Planning target volume

D0.03cc

Dose to the hottest 0.03 cc

Gy(RBE)

Gray (relative biological effectiveness)

Author contributions

SS, SC, and MA conceptualized the study and its clinical implications. RS and AG collaborated in treatment planning and dosimetric data collection. SS and SC performed the final data analysis and prepared the original draft. DS critically reviewed the manuscript and provided supervision.

Funding

The authors declare that no specific funding was received for this research from any funding agency in the public, commercial, or not-for-profit sectors.

Data availability

The datasets generated and/or analysed during the current study are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

This study was a dosimetric treatment-planning analysis performed using previously generated radiotherapy planning datasets and did not involve prospective recruitment of human participants, patient interaction, or identifiable patient information. According to the Indian Council of Medical Research (ICMR) National Ethical Guidelines for Biomedical and Health Research Involving Human Participants (2017), research involving analysis of anonymous or non-identified datasets is categorized as less than minimal risk. Based on the use of anonymized data and the non-interventional nature of this technical dosimetric study, formal ethics committee approval and individual informed consent were waived. The study was conducted in accordance with applicable institutional research policies and the principles of the Declaration of Helsinki where applicable.

Consent for participation

Not applicable. This study did not involve direct participation of human subjects.

Competing interests

The authors declare no competing interests.