Inter‐fraction dose and LET robustness in LET‐optimized proton therapy for pediatric, adolescent and young adult intracranial tumors

Authors: Alfredo Mirandola, Matteo Bagnalasta, Giuseppe Magro, Alessia Bazani, Silvia Molinelli, Vittoria Pavanello, Eleonora Rossi, Stefania Russo, Luca Trombetta, Alessandro Vai, Francesca Colombo, Ester Orlandi, Mario Ciocca, Sabina Vennarini

Abstract

Background

Although pediatric, adolescent and young adults (AYA) patients are known to be eligible for proton therapy, the presence of high linear energy transfer (LET) in critical structures, such as the brainstem, could lead to an increased risk of toxicity. To address this issue, dose‐averaged LET (LETd)‐optimized (LO) planning strategies have been proposed as a means to mitigate LET‐related risks while maintaining dose quality. However, the balance between dose distribution quality, LETd optimization, and plan robustness must be evaluated across the entire treatment course. As the treatments here investigated are delivered in multiple fractions, the assessment of the benefits of LETd optimization and its robustness against the anatomical and setup variations during treatment course, is crucial.

Purpose

This study investigates the inter‐fraction robustness of LO plans compared to standard (STD) dose‐driven plans in pediatric and AYA patients with intracranial tumors. Both dose and LETd distributions were evaluated using re‐evaluation CTs acquired during treatment. Additionally, recalculations with variable relative biological effectiveness (RBE) models were performed to investigate plan sensitivity to biological uncertainties.

Methods

Twenty patients (prescription dose ≥ 54 Gy) were retrospectively included. For each patient, STD and LO plans were generated and clinically approved by a medical doctor. Both plan types were optimized with robust optimization parameters, including 2 mm setup and 3.5% range uncertainties. Reference plans were subsequently recalculated using variable RBE models, in addition to the fixed RBE = 1.1 assumption. To evaluate robustness, all plans were recalculated on re‐evaluation CTs acquired throughout the treatment course. Dosimetric endpoints included D98% and D1% for the clinical target volume (CTV) and D1% for the brainstem. LETd robustness was assessed by quantifying the brainstem volume receiving > 50 Gy and LETd above thresholds of 4, 3.5, and 3 keV/µm (V50@LETdx).

Results

LO and STD plans exhibited comparable dose distributions, with no statistically significant differences in CTV coverage or brainstem sparing. In terms of LET, LO plans achieved a significant reduction in V50@LETdx volumes (p < 0.01). Furthermore, inter‐fraction variation in LETd‐sensitive brainstem volumes was lower in LO plans, indicating enhanced LETd robustness across the treatment course. As expected, due to the optimized LETd values in the brainstem, plan recalculations with variable RBE models showed a smaller D1% deviation in LO compared to STD plans when referenced to the fixed RBE = 1.1 model. These results demonstrate that LO not only maintains conventional dose robustness but also reduces LETd‐related uncertainties.

Conclusions

LO and STD plans exhibited comparable dose distributions, with no statistically significant differences in CTV coverage/brainstem sparing. LO plans achieved a significant reduction in V50@LETdx (p < 0.01). Furthermore, inter‐fraction variation in LETd for brainstem was lower in LO plans, indicating enhanced LETd robustness throughout the treatment course. The plan recalculations with variable models showed a minor D~1% ~deviation in LO versus STD plans, when compared to fixed 1.1 RBE model.

INTRODUCTION

Proton therapy (PT) has become increasingly adopted in pediatric oncology due to its potential to spare healthy tissues and reduce long‐term toxicities. Given the limited access to proton therapy worldwide, eligibility is often reserved to cases where a significant dosimetric advantage is expected to translate into reduced toxicity, such as in specific tumor histologies ^1^ . In contrast, for pediatric and AYA patients, the anticipated reduction in short and long‐term toxicity is so evident that further comparative evaluations against photons are generally not required to justify access to proton therapy

(https://www.astro.org/ASTRO/media/ASTRO/Daily%20Practice/PDFs/ASTROPBTModelPolicy.pdf).

This is particularly important when critical structures such as the brainstem are in close proximity to the CTV.

However, despite its physical advantages, PT is characterized by an increase in linear energy transfer (LET) towards the end of the Spread‐out‐Bragg‐peak (SOBP), which may enhance the relative biological effectiveness (RBE) ^2^ and increase the risk of adverse effects in organs at risk (OARs).

Indeed, radiation‐induced brainstem necrosis represents a clinically significant adverse effect, particularly in pediatric and AYA patients receiving PT for central nervous system malignancies such as ependymoma, or atypical teratoid/rhabdoid tumor. ^3^ Moreover, since children exhibit greater radiosensitivity compared to adults, minimizing the sources of possible damages to critical OARs represents a challenge to be deeply pursued.

Although substantial preclinical evidence supports a correlation between LET and RBE, ^4^ a fixed RBE value of 1.1 remains the clinical standard for PT dose prescription. ^5^ This simplification persists despite recognition that RBE is higher at the distal end of the SOBP ^4^ and it is influenced by multiple factors, including LET, total dose, fractionation schedule, and tissue‐specific radiobiological parameters such as the alpha–beta (α/β) ratio. ^6^

Experimental studies, primarily in vitro, have driven phenomenological RBE models that account for LET and tissue radiosensitivity to estimate the biological effect. In vivo investigations have provided additional support for a LET‐dependent RBE, although clinical validation remains limited due to methodological heterogeneity and insufficient prospective data.

Clinical studies evaluating the association between LET and radiographic or symptomatic toxicities in high‐dose regions have produced inconsistent findings. Some works report statistically significant correlations ^7^ , ^8^ , ^9^ while others have not demonstrated such associations. ^10^ In recent years the transition from passive scattering to active spot‐scanning delivery techniques in PT has improved the conformality of dose distributions. However, these advances have also introduced increased spatial heterogeneity in LET distributions, thereby enhancing the sensitivity of both dose and LET to uncertainties in beam range and anatomical configuration. ^11^ Multiple studies have investigated how variations in beam arrangement and target positioning influence LET distribution under different delivery techniques. ^12^ , ^13^ , ^14^ Some studies faced the LET optimization problem in proton therapy ^15^ , ^16^ , ^17^ in the past years while the clinical implementation of LET‐based optimization within Treatment Planning Systems (TPS) has only become feasible in recent years. This optimization strategy—when combined with appropriate beam arrangement, beam weighting, choice of delivery technique, e.g., Single Field Uniform Dose versus Intensity‐Modulated Proton Therapy (IMPT), and the application of distal margins to OARs—contributes significantly to ensuring the overall quality and appropriateness of the treatment plan, particularly with respect to LET distribution. Nevertheless, the LET‐based optimization planning strategy introduces an additional layer of complexity in treatment plan, potentially compromising its robustness and making it more sensitive to variations arising from setup uncertainties and anatomical changes. While the physical and dosimetric benefits of dose‐averaged LET (LETd) optimized (henceforth LO) plans have been reported in literature, ^18^ , ^19^ less is known about its performance under real‐world anatomical variations occurring during the treatment course.

Hence, beyond dose distributions, plan robustness ^20^ could play a crucial role even in terms of LET distribution that can be, as well, potentially perturbed by patient motion, setup uncertainties and anatomical changes. In this study, we evaluate the inter‐fraction robustness of standard (STD) and LO proton therapy plans in a cohort of 20 patients with intracranial tumors, with the aim of comprehensively assessing plan variations in terms of both dose and LETd distribution.

Furthermore, both STD and LO plans were recalculated with different RBE models, hence assessing the relative variation of some dosimetric parameters of the two plan types compared to the reference 1.1 fixed RBE scenario.

MATERIALS AND METHODS

Patient cohort and treatment planning optimization

A retrospective cohort of 20 pediatric/AYA patients was enrolled, with a median age of 12 years (range: 1–28 years), diagnosed with tumors of the brain and intracranial regions, and treated with a prescribed dose of ≥ 54 Gy.* The median total prescribed dose was 54 Gy (range: 54–74 Gy), delivered using the conventional daily fractionation of 1.8 Gy. One patient was treated with 2 Gy per fraction. The tumor histologies included in the cohort 6 rhabdomyosarcomas, 4 low‐grade gliomas, 3 skull base chordomas and chondrosarcomas, 3 craniopharyngiomas, 2 supratentorial ependymomas, 1 meningioma, and 1 non‐germinomatous germ cell tumor. Table 1 summarizes patient cohort's characteristics. The patient cohort was selected based on the close anatomical proximity between the brainstem and the CTV.

All patients were treated at the National Center for Oncological Hadrontherapy (CNAO), in Italy. Proton plans were optimized with the Multi‐Field‐Optimization (MFO) technique using active pencil beam scanning with beams commissioned as reported in ^21^ and by employing TPS settings routinely adopted in clinical practice. All the plans were calculated with RayStation TPS, 2024‐DTK (RaySearch Laboratories AB, Stockholm, Sweden). Scenario‐based robustness was part of the optimization the clinically adopted intracranial robustness parameters of 2 mm for setup uncertainty and 3.5% for range uncertainty were used to optimize all plans, hence generating 42 different dose distribution scenarios on the reference CT. LVH (LETd Volume Histogram) objectives were added in the cost functions for LO plans only, allowing LETd robust optimization. In contrast, STD plans were optimized without LVH objectives. Robust evaluation across the 42 scenarios was performed only for dosimetric parameters and DVHs, since LETd and LVHs can be assessed solely on the reference scenario within the TPS.

Between two and four beams, depending on the target location and its geometric proximity to the organs at risk, were used for each plan. In principle, and whenever possible, every part of the target was irradiated by at least two fields for safety and robustness reasons, both in STD and LO plans. The intervention of the medical physicist planner ensures that no single field contributes more than approximately three‐quarters of the prescribed dose. All the reference plans, for both plan types, were recalculated by using two variable‐RBE models as reported afterwards.

In LO plan optimizations, additional LVH robust objectives to the brainstem were adopted, while other OARs and CTV objectives and optimization weights remained unaltered. Beam arrangements and clinical settings were kept unchanged for both plan types, for consistency. Both plans were then approved by an experienced pediatric radiation oncologist following analogous acceptance criteria. In this analysis D1% and D98% for CTV and D1% for brainstem were selected as dosimetric indices for plan comparison in terms of robustness.

In particular the volumes of brainstem receiving more than 50 Gy and LETd higher than a certain threshold (V50@LETdx) were extracted by a custom in‐house script. LETd of 4, 3.5 and 3 keV/µm were investigated in agreement with values reported in. ^8^ , ^22^

Dose and LETd robustness assessment on the re‐evaluation CTs

Inter‐fraction robustness was evaluated in a separate phase, entirely following the optimization process described above, to assess potential real changes occurring during the course of treatment and to reflect actual variations. Each patient underwent multiple re‐evaluation CTs (median: 2, 1–4), routinely acquired to assess the appropriateness and robustness of the reference treatment plan. The timing of these re‐evaluation CTs could be scheduled at regular intervals (e.g., weekly) throughout the treatment course or adjusted according to specific patient‐related clinical needs. The inter‐fraction dose robustness was quantified, for both plan types, by evaluating the normalized relative differences of the investigated dosimetric parameters (D98% and D1% for CTV, D1% for brainstem) between the j‐th evaluation CT and the reference CT: ΔDx%/Dx%=Dx%,j−Dx%,ref/Dx%,ref

Since the involved brainstem volumes were very small, the inter‐fraction LETd robustness was assessed by evaluating the absolute differences of the investigated quantities as ΔV50@LETdx=V50@LETdx,j−V50@LETdx,ref

For both dosimetric and LET values, normalized relative differences were tested for significance with a Wilcoxon signed‐rank test with a significance threshold of p < 0.01 on paired samples. Moreover, in order to determine the sample size, a power analysis test was also conducted with alpha level (α = 0.05) and power of 0.8.

Variable RBE weighted dose evaluation

For both plan types, all dosimetric indices were evaluated by recalculating each plan with the Unkelbach model ^23^ , ^24^ which linearly relates RBE to LET and defines an RBE of 1.1 at a LETd of around 2–2.5 keV/µm, typical in the center of a spread‐out Bragg peak. This choice allowed a more direct interpretation of LET effects, avoiding the complexities introduced by alpha–beta ratio dependence in phenomenological models. For comparison, the McNamara model was also applied, assuming an alpha–beta of 2 Gy. ^25^

RESULTS

Figure 1 shows the CTV dose metrics over the patient cohort. In particular, CTV D98% and D1% were found not significantly different (p > 0.2) between STD and LO optimization techniques. All plans (100%) respected the D1% dose constraint to the brainstem, with no statistically significant differences (Figure 2) between plan types (p > 0.2).

Figure 3 shows an example of dose distributions for a STD and LO plan (fixed RBE = 1.1).

Concerning LETd evaluation, the difference in average V50@LETdx was always statistically significant (p < 0.01) between STD and LO plans, with corresponding mean volumes of 0.36 cc versus 0.11 cc at 4 keV/µm, 0.86 cc versus 0.31 cc at 3.5 keV/µm, and 1.76 cc versus 1.04 cc at 3 keV/µm, respectively. V50@LETd4 reduced to zero cc in LO plans for 7 out of 20 patients with max ΔV50@LETd4 = 0.52 cc.

In addition, to assess the significance of the LETd‐related results, a power analysis (α = 0.05) was performed on the sample for V50@LETd4, V50@LETd3.5, and V50@LETd3, yielding statistical power values of 0.55, 0.83, and 0.41, respectively. Sample sizes of 35, 19, and 51 patients would be required to reach 0.8 power.

Figure 4 shows an example of LETd distribution for a STD and a LO as it can be observed, LVH alone does not adequately capture the differences in LETd between the two plan types. Therefore, we developed a custom in‐house script that extracts dose and LVH data showing them in an integrated graphical form, as illustrated in Figure 5.

Dose and LETd Robustness on the nominal plans

Table 2 shows the median (with range) and interquartile range (IQR) of the worst‐case scenario for key dosimetric parameters, with values normalized to the prescription dose for the CTV (D98% and D1%), while reported in absolute dose for the brainstem. Data are collected both in STD and LO plans, calculated across the entire patient cohort. These values were derived from the robust DVHs evaluated across the 42 uncertainty scenarios generated during the optimization phase. The averaged normalized deviation of the investigated dosimetric parameters in the worst‐case scenario, with respect to the nominal STD plan, was 4.43% and 3.25% for the CTV |ΔD98%|/D98% and |ΔD1%|/D1%, respectively. For the brainstem, the averaged |ΔD1%|/D1% was 3.78%.

For LO plans, similar results were |ΔD98%|/D98% and |ΔD1%|/D1% for the CTV were 4.63% and 3.38%, respectively, and the averaged brainstem |ΔD1%|/D1% was 4.79%.

Inter‐fraction dose and LETd Robustness for STD optimized plans

The STD dose distributions to the CTV and brainstem are the normalized relative differences are within 2% for more than 95% of the STD reference plans recalculated on the evaluation CTs. For the CTV, the inter‐patients averaged normalized deviations |ΔD98%|/D98% and |ΔD1%|/D1% were 0.53% and 0.55%, respectively. Regarding the brainstem, the averaged |ΔD1%|/D1% was 0.73%.

The quantities V50@LETd4, V50@LETd3.5, V50@LETd3 for reference plans and for the corresponding recalculation on the j‐th evaluation CT are robust as well. The corresponding worst cases, hence the absolute maximum deviations from the reference plan and CT (max‐ΔV50@LETdx) across the patient cohort, have been chosen as indicators of LETd robustness. Data are summarized in Table 3.

As reported, 0.24 cc (0.97 cc for the reference CT versus 0.73 cc for the evaluation CT2), 0.32 cc (2.35 cc for the reference CT versus 2.67cc of the evaluation CT1) and 0.79 cc (0.11 cc of the reference CT versus 0.90 cc of the evaluation CT1) were the maximum ΔV50@LETd4, ΔV50@LETd3.5 and ΔV50@LETd3, respectively. Regarding inter‐scan variability, for patients with the highest number of evaluation CT scans (4), the maximum deviation between the i‐th CT and the j‐th CT was 0.1, 0.21, and 0.76 cc for ΔV50@LETd4, ΔV50@LETd3.5, and ΔV50@LETd3, respectively. These values are generally lower, or at most consistent, with the maximum deviations reported in Table 3.

Inter‐fraction dose and LETd Robustness for LO optimized plans

The normalized relative difference is within 2% for more than 95% of the recalculation of the LO reference plans on the evaluation CTs. For the CTV, the inter‐patients averaged normalized deviations |ΔD98%|/D98% and |ΔD1%|/D1% were 0.60% and 0.53%, respectively. Regarding the brainstem, the averaged |ΔD1%|/D1% was 0.80%.

Although smaller brainstem volumes at high LETd than STD plans were involved, the intrinsic LETd robustness was well maintained. As reported in Table 4, 0.21 cc (0.76 cc for the reference CT versus 0.55 for the evaluation CT2), 0.35 cc (1.42 cc for the reference CT versus 1.77 for the evaluation CT2) and 0.40 cc (2.45 cc of the reference CT versus 2.05 cc of the evaluation CT1) were the maximum ΔV50@LETd4, ΔV50@LETd3.5 and ΔV50@LETd3, respectively.

Regarding inter‐scan variability, for patients with the highest number of CT scans (4), the maximum deviation between the i‐th CT and the j‐th CT was 0.04, 0.10, and 0.17 cc for ΔV50@LETd4, ΔV50@LETd3.5, and ΔV50@LETd3, respectively. These values are generally lower, or at most consistent, with the maximum deviations reported in Table 4.

No statistical difference between STD and LO optimization plans were found in ΔV50@LETdx for none of the LETd levels here analyzed. Figure 5 shows the LVHs of five representative patients for a brainstem dose level of 50 Gy. A single exemplary DVH is presented as representative of the entire cohort, since all DVHs exhibit comparable dose profiles. All patient's data are reported in Supplemental Materials (Figures S1–S20).

As expected both nominal and recalculated LO plans show lower LETd values at the established dose threshold when compared with STD plans. High LETd robustness is observed for both plan types.

Figure 6 provides an overview of the variability in dose and LETd across the evaluation CTs.

Variable RBE weighted dose

All plan types were recalculated with the Unkelbach (UKL) and McNamara (MCN) variable RBE models, as shown in Figure 7. The resulting differences in the previously evaluated dose metrics, comparing the STD and LO plans with those obtained using the fixed RBE 1.1 model, are presented in Figure 8. As expected, the most significant differences between the fixed RBE model and the variable RBE models are LETd‐related. At the end of proton range, for these cases near the brainstem, LETd is higher (generally > 3 keV/µm) than in the target core (1.5‐2 keV/µm), resulting in larger dose differences between models (as for brainstem D1%). While no significant differences were observed for the CTV (p > 0.01), STD plans showed significantly greater differences for brainstem parameters (p < 0.01). Similarly, for the Unkelbach model, CTV differences remained comparable between STD and LO plans, whereas brainstem dose differences were again significantly higher in STD plans (p < 0.01).

DISCUSSION

This study assessed the inter‐fraction robustness of LETd‐optimized proton therapy plans in comparison to purely dose‐based plans in a cohort of 20 pediatric/AYA patients with intracranial tumors. The results indicate that LO plans exhibit similar dose robustness as STD plans, preserving both target coverage and brainstem sparing. Additionally, LETd‐optimized plans significantly reduce LETd hotspots in critical structures, and this reduction is robust, being consistent across multiple re‐evaluation CT scans.

In particular, no statistically significant differences were found between LO and STD plans for CTV D98%, CTV D1%, and brainstem D1%. These results confirm that the addition of LETd optimization objectives in the cost function during the optimization phase, does not negatively impact the plan dosimetric quality. Furthermore, both plan types demonstrated an excellent inter‐fraction dose robustness, assessed by comparing the re‐evaluation CTs with the reference plan CT. As detailed in the previous sections, the inter‐fraction averaged normalized relative differences of the investigated dosimetric parameters (|ΔD98%|/D98% and |ΔD1%|/D1%) remain below 1% with respect to the reference plans. Therefore, the dose deviations observed on the re‐evaluation CTs fall within the robustness envelope defined during the optimization phase on the reference CT. The corresponding worst‐case scenario values (across the 42 simulated scenarios) of the same parameters, in fact, are approximately 4%–5%, hence significantly higher. These observations proved that the robustness settings used during optimization effectively accounted for inter‐fraction anatomical variations in the evaluated cohort of patients.

These findings are valid for the vast majority (> 95%) of the analyzed cases. The most pronounced differences were observed in patients undergoing radiotherapy shortly after surgery, where substantial postoperative anatomical changes such as surgical cavity reduction or edema occurred. Those plans definitely necessitated timely replanning.

On the other hand, although the use of LETd as a surrogate for RBE for clinical endpoints is still controversial ^22^ , ^26^ we found of interest, beyond dose robustness, to assess the LETd robustness across the evaluation CTs as a useful parameter for establishing the appropriateness of a treatment plan.

It is well known, in fact, that LETd distributions in the OARs were characterized by higher heterogeneity and exhibited higher LETd values, compared to those within target volumes. Furthermore, LET distributions in OARs showed a pronounced sensitivity to range uncertainties ^14^ highly enhanced in tissue heterogeneities that proton beams could traverse along their path.

LO plans significantly reduced brainstem volumes exposed to high LETd levels (V50@LETd4, V50@LETd3.5, and V50@LETd3) compared to STD plans. The average reduction in the most clinically relevant dose‐LET level here investigated, V50@LETd4, exceeded 60%, with p‐values well below 0.01. Regarding V50@LETd4 and V50@LETd3 values, which both remain significantly lower for LO plans as confirmed by the Wilcoxon test, a larger sample size would be necessary to reach a high statistical power (0.8). This may be due to the fact that brainstem volumes receiving LETd ≥ 4 keV/µm at doses ≥ 50 Gy are very small—or even zero—in STD plans without LET optimization, since consciously planned to minimize the distal dose delivered to the brainstem. Moreover, LETd values ≥ 3 keV/µm at doses ≥ 50 Gy likely represent a particularly challenging cutoff, difficult to reduce in these patients. Lower LETd values are generally observed in the entrance region, or at least outside the distal part of the SOBP. In the clinical setting here investigated, however, the prescribed dose is ≥ 54 Gy, the number of fields is intentionally limited to reduce the dose bath, and the targets are located in close proximity to the brainstem. These factors constrain the possibility of effectively exploiting entrance regions, thereby making it inherently difficult to achieve LETd values below 3 keV/µm in the brainstem, as both geometry and prescription dose impose limitations.

Importantly, the inter‐fraction variability of high‐LET brainstem volumes was also reduced in LO plans, indicating enhanced plan stability during the treatment course. To date, no established threshold for either dose or LETd has been defined by the scientific community to indicate a specific risk of severe brainstem toxicity in pediatric patients undergoing proton therapy ^27^ The dose and LETd values analyzed in this study lie within the critical range potentially associated with brainstem necrosis. ^8^ , ^22^ It is also important to highlight that none of the patients in this cohort, treated with the standard (STD) plan, exhibited any form of treatment‐related toxicity, demonstrating that even the standard plans were carefully optimized using all previously mentioned strategies to mitigate potential LETd‐related toxicities. Moreover, the selected dose and LETd levels aimed to identify a non‐point‐like region within the brainstem potentially associated with severe toxicities.

Our findings could be further substantiated by a larger patient cohort, when available. Nevertheless, the results obtained provide a reliable trend. It would certainly be of interest to validate our findings by analyzing other anatomical sites, where setup errors and uncertainties related to cavities filling/emptying, or respiratory motion, may pose even greater challenges in LET‐robustness. Recalculation of approved treatment plans using real clinical datasets of re‐evaluation CT scans offers meaningful insights into the overall robustness of the plan. However, robustness evaluation can be alternatively performed on CBCT datasets, which can be converted into high‐quality synthetic CT images suitable for adaptive proton therapy and plan recalculations. This strategy enables a more frequent assessment of plan robustness without subjecting very young patients to the additional radiation exposure associated with repeated acquisition of conventional evaluation CT scans.

As an additional finding, the recalculation of both STD and LO plan types with variable RBE models ^23^ , ^28^ showed no significant dose differences in CTV, where the LETd is relatively low, approximately 1.5–2 keV/µm for both plan types. In contrast, deviations in brainstem dose were significantly smaller for LO plans, where LETd values remain lower than in STD plans, further supporting the robustness of LET optimization against uncertainties in biological modeling. LET‐based optimization, recently made feasible within TPS and investigated here in terms of both plan quality and robustness, does not, by itself, ensure overall plan quality and robustness. The conventional optimization criteria routinely employed in particle therapy planning remain fundamental and indispensable. Appropriate beam arrangement, such as avoiding the placement of OARs distally to beam paths, and minimizing beam traversal through cavities or regions affected by imaging artifacts, continue to be an essential first‐line strategy in treatment planning. More advanced techniques, such as LET minimization in OARs through the use of transmission beams ^29^ in critical regions, may also be considered to further improve plan safety and efficacy. Moreover, it is important to consider the incidence of brainstem toxicity during irradiation of the posterior cranial fossa. Tumors in this region are among the most frequent and prevalent in young children ^30^ who often require multiple surgical interventions followed by adjuvant radiotherapy. These treatment modalities are significant risk factors for the development of radiation‐induced toxicity—as commonly seen in ependymomas—but also represent the only prognostic factors associated with local disease control.

A critical point of discussion concerns the definition of brainstem dose constraints, an area of ongoing debate between European and American groups ^31^ , ^32^ Discrepancies remain regarding the incidence of brainstem toxicity with the use of proton therapy compared to photon techniques ^33^ , ^34^ As a result, a more cautious approach in using proton therapy is often adopted in this anatomical region, due to the close proximity of the tumor or tumor bed to a critical structure such as the brainstem.

It is possible that LET could play a pivotal role in addressing this issue, offering additional insights and helping to define safe and effective strategies for clinical practice. If validated by further evidence the safe implementation of proton therapy could unlock its full potential in minimizing radiation exposure to healthy tissues, thereby enhancing therapeutic outcomes in pediatric patients with posterior fossa tumors.

CONCLUSIONS

In conclusion, our findings indicate that LET optimization strategies can be safely integrated into clinical practice for these specific intracranial tumors, ensuring treatment robustness while potentially reducing LET‐related toxicities, although conclusive scientific evidence to firmly establish this correlation is still lacking.

CONFLICT OF INTEREST STATEMENT

The authors have no relevant conflicts of interest to disclose.

DECLARATION OF GENERATIVE AI AND AI‐ASSISTED TECHNOLOGIES IN THE WRITING PROCESS

During the preparation of this work, the author(s) used ChatGPT (OpenAI) in order to improve the linguistic clarity and readability of some sections of the manuscript. After using this tool, the author(s) reviewed and edited the content as needed and take full responsibility for the content of the publication.

Supporting information