Robustness evaluation of carbon ion radiotherapy for ventricular tachycardia

Graphical abstract

Highlights

• •In silico robustness evaluation of carbon ion radiotherapy for ventricular fibrillation.
• •Target coverage decreased by 13.8–19.0% when not accounting for cardiac motion.
• •Sufficient robustness required a margin or setup uncertainty of 5 mm.
• •Electrocardiographic-gated CT-based treatment plans were considered as ideal.
• •Case-specific robustness evaluation is recommended due to motion variability.

Abstract

Favorable clinical outcomes of carbon-ion radiotherapy (CIRT) for ventricular arrhythmia have been reported, and optimal treatment strategies are being explored. We assessed the robustness of CIRT plans against cardiac motion in ventricular tachycardia. Electrocardiographic (ECG)-gated computed tomography (CT) images at end-diastole and end-systole were used for planning, and dose distributions were recalculated on the opposite phase to evaluate dosimetric changes. Cardiac motion reduced target coverage by about 13.8–19.0%, whereas a 5-mm margin or robust optimization reduced the decrease approximately to 5%. An optimization approach that accounts for cardiac motion using ECG-gated CT provided a favorable balance for clinical application.

1. Introduction

In recent years, stereotactic arrhythmia radioablation (STAR) has emerged as a promising noninvasive treatment for ventricular tachycardia (VT) [1], [2], [3], [4]. The underlying mechanisms are thought to differ over time: in the acute phase, STAR may induce conduction slowing [5], modify gap junction [6], [7] expression, and alter sodium channel activity [8], [9], whereas in the chronic phase, fibrosis and scar formation [10] are considered the predominant contributors to arrhythmia suppression. Particle therapy, characterized by its distinct Bragg peak [11], [12], has gained attention as an advanced modality because of its potential to deliver highly conformal dose distributions. Comparative planning studies have primarily focused on proton therapy, which offers dosimetric advantages over photon-based techniques. Recently, Amino et al. reported a case of carbon ion radiotherapy (CIRT) in a patient with premature ventricular contraction-induced dilated cardiomyopathy, demonstrating a marked reduction in VT episodes [13]. CIRT may offer further benefits beyond proton therapy owing to its sharper lateral penumbra and higher biological effectiveness [12]. However, despite these theoretical advantages, the robustness of dose distribution in CIRT remains a key concern. Therefore, this study aimed to conduct an assessment of treatment-planning robustness for cardiac CIRT, providing preliminary insights into its clinical feasibility and dosimetric stability.

2. Materials and methods

We retrospectively analyzed electrocardiographic (ECG)-gated cardiac computed tomography (CT) images from 15 patients with cardiac failure who underwent imaging at our institution. The study protocol was approved by the Institutional Review Board (Approval No. 2025–72). All contrast-enhanced, ECG-gated CT scans included two cardiac phases—systolic and diastolic. For each case, virtual targets corresponding to the left ventricular apex (Target 1) and the mid-ventricular region (Target 2) were defined on either the end-diastolic or end-systolic CT image used for treatment planning. Target delineation was performed by an experienced medical physicist with expertise in cardiac radiotherapy, following the methodologies reported by van der Ree et al.[14] and Knutson et al. [15]. Examples of target delineation are shown in the Supplementary Fig. S1. In this study, considering the target sizes reported by Knutson et al. [15] and Widesott et al. [7], Target 1 and Target 2 were assumed to exist independently. Treatment planning was performed using scanning carbon ion beams delivered from a single beam angle. All plans were generated using RayStation version 2023B (RaySearch Laboratories, Stockholm, Sweden). Unless otherwise specified, all reported doses represent relative biological effectiveness (RBE)–weighted doses. The prescribed dose was 25 Gy delivered in a single fraction [3], [13]. Physical dose were calculated using the pencil beam algorithm (CarbonPencilBeam v6.0), and biological dose distributions were computed using the modified microdosimetric kinetic model (MKM) [16], [17]. To account for interphase cardiac motion, 10 planning scenarios were created. In the internal target volume (ITV) plans, treatment plans were generated by prescribing the dose to an ITV created using isotropic margins. Four margin sizes were evaluated: 1-, 2-, 3-, and 5-mm. Robust plans were generated using a min–max optimization approach [18], with cardiac motion compensated by incorporating isotropic setup uncertainties of 1-, 2-, 3-, and 5-mm. A range uncertainty of ±2% was also applied. For both the ITV plans and Robust plans, treatment planning was performed using a single CT phase (end-diastole or end-systole). In contrast, the 2-phase robust plan utilized CT images from both phases and applied min–max optimization to compensate for interphase target displacement. For comparison, a plan optimized without accounting for cardiac motion was defined as the Plan w/o. Details of each optimization strategy are provided in the Supplementary Material A and Fig. S2. In this study, all treatment plans were generated by applying optimization parameters exclusively to the target volumes. This approach was adopted to exclude the influence of the optimization algorithm and to purely elucidate the robustness of target dose among different cardiac motion–management strategies. Each plan created on one CT phase was recalculated on the corresponding alternate phase to evaluate interphase dose variations. To assess radiation exposure to organs at risk (OARs), the mean dose to the adjacent lung region and the left ventricle was compared across all plans relative to the Plan w/o. Because CT images with a limited field of view (FOV) were used, the whole lung was not included in the imaging volume. Therefore, the lung region was defined as the intersection of the lung volume with a 1-cm isotropic expansion of the left ventricle. Examples of OAR delineation are shown in the Supplementary Fig. S3.

Quantitative results were analyzed using the Statistical Package for the Social Sciences (SPSS), version 28 (IBM SPSS Statistics for Windows; IBM Corp., Armonk, NY, USA). Because normality could not be confirmed by the Shapiro–Wilk test, the Friedman test followed by Bonferroni correction was applied to assess significant differences among the three treatment planning methods and in the evaluation of respiratory motion. A p-value < 0.05 was considered statistically significant. Results are presented as the median (interquartile range).

3. Results

Fig. 1 illustrate the changes in V_95% for Targets 1 and 2, respectively. In the Plan w/o, V_95% decreased by an average of 13.7% (8.3% − 23.9%) for Target 1 and 11.8% (5.3% − 21.6%) for Target 2. Increasing either the ITV margin size or the robustness parameter reduced the variation in V_95%. Specifically, V_95% changes were 0.1% (0.0% − 1.2%) and 0.0% (0.0% − 0.1%) for Targets 1 and 2, respectively, with the ITV 5-mm plan, and 2.4% (0.7% − 4.3%) and 1.0% (0.1% − 4.6%) with the Robust 5-mm plan. In the 2-phase robust plan, reductions in V_95% were 0.1% (0.0 – 0.4%) for Target 1 and 0.0% (0.0% − 1.5%) for Target 2. For Target 1, the lung D_mean increased by 6.7% (5.3% − 13.4%), 1.5% (1.3% − 3.8%), and 4.3% (3.3% − 8.8%) for the ITV 5 mm, two-phase robust, and Robust 5 mm plans, respectively. For Target 2, the left ventricle D_mean increased by 8.1% (6.3% − 10.1%), 1.4% (0.9% − 2.6%), and 5.5% (4.3% – 6.9%) under the same conditions. Results for OARs other than those shown here are presented in Supplementary Fig. S4.

Fig. 1.

Changes in target coverage after dose recalculation and variations in OAR doses relative to the Plan_w/o. Target coverage changes are expressed as variations in V95%, while OAR doses are represented by the mean dose (Dmean) to the lung ROI and the left ventricle, which were most impacted for each target. * indicates a significant difference (p < 0.05) compared to Plan_w/o.

Fig. 2 shows the dose distribution of a representative case that supports the results presented in Fig. 1. In Plan w/o, the decrease in target coverage (V_95%) was 21.8% for Target 1 and 27.7% for Target 2 in the case shown in Fig. 1 (Patient 1). In the ITV 5 mm plans, dose spread tended to be relatively uniform in all directions surrounding the target, whereas in the two-phase robust plans and the Robust 5 m m plans, dose spread was non-isotropic and exhibited directional dependence. The dose distributions of all treatment plans evaluated in this study are shown in Supplementary Fig. S5 and S6. The DVH curve for this case is shown in the Supplementary Fig. S7 and S8.

Fig. 2.

Representative dose distributions of the planned dose and the recalculated dose. The dose distribution shows the dose of a treatment plan generated at end-systole (Plan dose), recalculated on the end-diastolic phase (Recalc. dose). Reduced target coverage was observed in the region indicated by the red arrow. Dose spread to the OAR was observed in the region indicated by the yellow arrow. All doses are expressed as RBE-weighted doses. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

4. Discussion

This study assessed the robustness of treatment planning for cardiac CIRT. Our analysis indicated that an approximate displacement of 5 mm should be considered to maintain adequate target coverage, either through isotropic margin expansion or robust optimization. However, this degree of compensation was also associated with an increase exceeding 5% in the mean dose to OARs, emphasizing the trade-off between target coverage and OAR sparing. 2-phase robust optimization using ECG-gated CT effectively limited target dose reductions to approximately 1%–2%, while producing only a modest increase in OAR dose compared with plans assuming a 5-mm motion margin (ITV 5 mm or Robust 5 mm). This approach appeared clinically advantageous for preserving dose conformity and minimizing unnecessary exposure to OARs.

The present results align with those of Wang et al., who quantified cardiac deformation during the cardiac cycle and reported that most STAR targets exhibited motion within 5 mm, although some exceeded 10 mm [19]. Consistent with their observations, our data suggest that a 5 mm margin generally ensures adequate target coverage. However, substantial interpatient variability persists, with several cases demonstrating inadequate coverage even after robust optimization with 5-mm margins. This finding underscores the importance of patient-specific motion assessment before treatment planning to achieve optimal dosimetric outcomes.

Regarding cardiac STAR, Shah et al. reported treatment plans using proton therapy and demonstrated the potential for reduced OAR toxicity compared with photon-based STAR [6]. Although few studies have directly compared CIRT with photon STAR for VT, planning investigations in other anatomical sites suggest that CIRT may similarly achieve improved OAR sparing while maintaining adequate target coverage.

Because the magnitudes of expected uncertainties in proton beam therapy and CIRT are generally considered to be comparable [20], [21], the findings of the present study may also be applicable to proton STAR. However, compared with CIRT, proton beam therapy exhibits distinct dose distribution characteristics, including a larger penumbra and the absence of a fragmentation tail [12], [22]. These differences may result in a broader lateral dose spread and a sharper distal dose fall-off than those observed in the present study, which differs from the dose spread trends identified for carbon-ion beams.

Various techniques for ECG-gated irradiation have been explored in cardiac STAR [23], [24], [25]. However, reports addressing ECG-gated irradiation in particle therapy—particularly in cardiac CIRT—remain scarce. Because particle therapy is inherently more sensitive to tissue density variations and target motion [26], [27], and because the interplay effect between scanning beams and moving targets is a well-recognized issue [28], [29], [30], developing motion-robust and gating strategies tailored to the physical characteristics of particle beams is essential for clinical implementation.

This study has several limitations. First, ECG-gated CT images with a limited FOV were used. Consequently, a comprehensive evaluation of OARs in cardiac CIRT could not be achieved. Ideally, all OARs in cardiac treatment other than the heart and lungs should have been evaluated. However, this study represents an extremely preliminary medical physics investigation of cardiac CIRT, and further detailed examination will be essential. Second, only two cardiac phases were analyzed; future studies should incorporate large–FOV, multiphase ECG-gated CT acquisitions specifically designed for treatment planning. Third, respiratory motion was not considered in this analysis. Previous studies have shown that respiratory motion can cause notable cardiac shifts [31], [32], [33]. Shah et al. considered cardiac and respiratory motion using averaged images obtained from four-dimensional CT (4DCT) [6]. Given the high sensitivity of particle beams to tissue changes associated with motion, it would be ideal to account for motion using both 4DCT and ECG-gated CT. In this study, cardiac motion was addressed using ECG-gated CT, whereas respiratory motion was not considered. Future investigations combining 4DCT-based respiratory motion assessment with ECG-gated CT-based cardiac motion assessment may provide a more accurate evaluation of particle beam robustness in STAR. Another limitation concerns the RBE model used. Currently, no experimentally validated RBE model exists for CIRT in arrhythmia radioablation. Determining the optimal biological dose and establishing a validated RBE model for cardiac CIRT represent essential areas for future investigation. In general, targets in STAR are defined based on the American Heart Association 17-segment model [34]. Further investigations of the motion characteristics and treatment planning for each of these 17 segments would contribute to a deeper understanding of STAR. In this study, the pencil beam algorithm was used for dose calculation. Because it is well recognized that the pencil beam algorithm has limitations in dose calculation accuracy [35], particularly at lung–tissue interfaces, the evaluated dose at the lung boundary in this study may have been affected. In future studies, re-evaluation using Monte Carlo simulations [35], [36] is expected to provide further important insights.

In conclusion, cardiac motion cannot be neglected in cardiac CIRT. The present study showed that compensating for approximately 5 mm of motion is required to ensure adequate target coverage. However, with future advancements in ECG-gated irradiation and other motion-management strategies, it may become possible to deliver cardiac CIRT under smaller motion conditions, thereby minimizing irradiated volumes and potentially reducing OAR exposure.

Data availability statement for this work

Research data are stored in an institutional repository and will be shared upon request to the corresponding author.

CRediT authorship contribution statement

Yuya Miyasaka: Conceptualization, Data curation, Formal analysis, Investigation, Writing – original draft. Mayumi Ichikawa: Data curation, Writing – review & editing. Hikaru Souda: Data curation, Formal analysis, Writing – review & editing. Yuki Tominaga: Conceptualization, Writing – review & editing. Yuhi Wakisaka: Conceptualization, Writing – review & editing. Hongbo Chai: Investigation, Writing – review & editing. Miyu Ishizawa: Formal analysis, Writing – review & editing. Masashi Koto: Project administration, Resources, Writing – review & editing. Takeo Iwai: Investigation, Project administration, Writing – review & editing.

Declaration of competing interest

The authors 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

The following are the Supplementary data to this article: