Functional lung avoidance stereotactic body radiation therapy for central lung cancer patients.

Introduction
Over the past two decades, stereotactic body radiation therapy (SBRT) has emerged as the standard treatment for patients with medically inoperable early-staged non-small cell lung cancer (NSCLC). Overall survival (OS) for stage I NSCLC after SBRT was 91% (95% confidence interval: 85–98%) at 3 years and 87% (95% confidence interval: 79–95%) at 5 years (1). With the prolongation of patients’ OS, more attention is paid to quality of life (2). SBRT delivers high biologically effective dose to focal tumor and allows for a sharp dose gradient around the target, thus preserving the surrounding organs-at-risk (OARs). For peripherally located lung cancer, the incidence of grade ≥2 radiation pneumonitis (RP) after SBRT has been reported to be 5.2–21% (3-6).
Despite the highly conformal dose distribution, patients with centrally located tumors are still at a higher risk of radiation-induced toxicity from high-dose ablative radiotherapy (RT) (7). A phase II study of SBRT for central lung cancer reported an 11-fold risks of toxicity, i.e., tumors within 2 cm of the mediastinum or the central airways, compared to peripheral tumors (8). Therefore, patients with central lung cancer require more careful patient-counseling and shared decision-making. Fine-tuned planning strategies may be effective in toxicity reduction and improving the treatment outcome, as well as the quality of life.
In our clinic, dose constraints of SBRT for total lung are V5 Gy <30% and V20 Gy <10%. However, lungs are non-homogeneous organs, especially for patients with comorbidities (9). Functional lung avoidance (FLA) plan allows for selectively avoiding relatively high-function areas, which generally results in the dose being preferentially funneled through defected regions due to their weak-ability to re-ventilate after RT (10). For long-term survivors, the preservation of high-function lung has remarkable clinical benefits, including reduced pulmonary toxicity, and less decline in pulmonary function tests (PFTs) results (11-13). On the other hand, the application of low-dose helical computed tomography (CT) has altered the landscape of pulmonary screening. The early detection of tumors has led to a 20% decline in mortality for lung cancer patients (14). The above-mentioned reasons greatly increase the interest in preserving high-function lung during RT (15,16). Yamamoto et al. (17) and Faught et al. (18) demonstrated that CT ventilation-based FLA planning can effectively reduce pulmonary toxicity. Our team also investigated avoidance plans based on ventilation and density in conventional fractionated RT (19).
Previous researchers have generated FLA plans for NSCLC, confirming that functional avoidance is a promising direction for RT (20-22). In addition, the dose-function metrics has shown notable improvements over dose-volume metrics for toxicity assessment (23-25). Nevertheless, studies on FLA plans for SBRT are still not sufficiently investigated, particularly for central lung cancer. To the best of our knowledge, there are no studies on the FLA SBRT designs for central lung cancer. Furthermore, how much improvement in ventilation preservation can be achieved in central lung cancer patients after FLA stays unknown.
The purpose of our study was to evaluate function-based and anatomy-based SBRT plans (conventional clinical plans) qualitatively and quantitatively for patients with central lung cancer. We hypothesized that FLA plans based on 4-dimensional computed tomography (4DCT) ventilation would improve dosimetric sparing of high-function lung and reduce lung toxicity in SBRT for central lung cancers. Limiting dose to functional lung can mitigate decline in pulmonary function, and our study could then be used in image-guided treatment paradigm and represent an important step towards reducing pulmonary toxicity for central lung cancer patients.

Methods

Patients selection
To generate FLA plans, 20 central lung cancer patients treated with SBRT in Shanghai Chest Hospital were retrospectively investigated. Inclusion criteria were defined as follows: (I) pathologically confirmed lung cancer; (II) age ≥18 years; (III) planned receipt of stereotactic body radiation therapy (SBRT); (IV) availability of pre-RT PFT data; (V) a pre-RT 4DCT scan to estimate the ventilation map. Exclusion criteria: a history of prior chest radiation therapy. This criterion excludes the interference of radiation-related lung tissue changes on CT ventilation mapping results and the risk assessment of RP. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Institutional Review Board of Shanghai Chest Hospital (No. KS23043), and the requirement for written informed consent was waived due to institutional policy and the retrospective study design.
The prescription dose of 60 Gy in 8 fractions was prescribed to cover at least 95% of planning target volume (PTV). Each patient underwent a 4DCT scan for treatment planning using the SOMATOM Definition AS (Siemens Healthcare Gmbh, Erlangen, Germany) 3 to 7 days before RT. The following scanning parameters were used: 120 kVp, 3 mm slices, and variable mAs/rotation. For reconstruction, 4DCTs were sorted into 10 breathing phase volumes representing various levels of inspiration or expiration using a Real-time Position Respiratory Gating System (version 1.7.5, Varian Medical Systems, Varian, Palo Alto, CA, USA).
All patients were treated with SBRT in an Edge linear accelerator (Varian) equipped with 120 multi-leaf collimator (MLC) leaves (Millennium MLC) and a 6 megavoltage (MV) photon beam. All patients received cone beam computed tomography (CBCT) by using online IGRT (Varian) before each fraction. The characteristics (age, gender, histology, etc.) of the patients are shown in Table 1.

Target and OARs delineation
An internal target volume (ITV) was generated based on the maximum intensity projection (CTMIP) of 4DCT image, and PTV was created by a 5 mm uniform expansion of the ITV. Also, OARs were delineated in CTMIP manually by experienced radiation oncologists.
4DCT scans were used to calculate ventilation maps. The lungs were segmented on the end-inhale and end-exhale phases. Lung voxel elements were then mapped from inhale to exhale phase using deformable image registration (DIR) (26) in MIM (MIM Maestro, version 7.0.4, MIM Software Inc., Cleveland, OH, USA). The DIR module adopted a free-form intensity-based algorithm, and the deformable transformation was created starting from a rigid fusion of the initial image sets, through the minimization of a cost function that takes into account the image similarity and the physical likelihood of the transformation.
The three-dimensional (3D) deformation on the voxel base was recorded, and the Jacobian determinant (27) was calculated based on Eq. [1] by using MIM toolbar. A Jacobian determinant of 1 represents no volume change during the breathing motion. The voxelized ventilation was defined as Jacobian determinant minus 1 {Eq. [2]}.
In this study, the high-ventilation (H-V) regions were defined as the highest 30% of the intensity value in the ventilation map. We have adopted this definition from our previous research (19). O’Reilly et al. (28) also showed that thresholds in the range of 45% to 60% of the ventilation values provided the optimal H-V zone definition in their receiver operating characteristic curves analysis to estimate prediction of RP. To summarize, total lung was separated into HV, medium-ventilation (M-V) and low-ventilation (L-V) regions: the highest 30% of the ventilation voxels were deemed the H-V region, the highest 60% of the ventilation voxels minus the H-V regions was the M-V region, and the rest lung was defined as an L-V region.

Treatment planning
For each patient, an anatomy-based and a function-based plans were generated in Pinnacle3TM treatment planning system (TPS, v9.10, Philips Medical Systems, Cleveland, OH, USA) using intensity modulated radiation therapy (IMRT) delivery with the number, direction of beams, and beam energy at the discretion of the planner. Both plans have the same beam angles. Pulmonary heterogeneity was taken into account by using the collapsed cone convolution algorithm. For the optimization process, anatomy-based plans were optimized using no sparing other than clinical dose-volume constraints (29). For the function-based plans, the reduction in dose to the functional-avoid structure was accomplished by inserting objective parameters for the functional structure in the optimization that minimizing dose in H-V lung. Higher dose to other OARs were allowed but still within the clinical dose constraints. Detailed optimization objectives and priorities were listed in Table 2.

Plan evaluation
To compare the performance of the two plans, we calculated dose-volume histograms (DVHs) for the standard OARs and PTV, along with Dmean and V15.4 Gy of the H-V lung. Since our patient cohorts were treated with SBRT, we used V15.40 Gy instead of the commonly used V20 Gy in conventional fractionated RT; 15.40 Gy in 8 fractions is the equivalent of 20 Gy in 30 fractions given by the universal survival curve model (30) with α =0.133, β =0.037, D0 =1.30, Dq =3.56 Gy, and Dt =8.6 Gy. Therefore, V4.5 Gy, V8.4 Gy, V12 Gy, V15.4 Gy, and V21.8 Gy in our study correspond to the V5 Gy, V10 Gy, V15 Gy, V20 Gy, and V30 Gy, respectively. It should be noted that although we evaluated the plan using the equivalent dose in 2-Gy fractions (EQD2), the optimization objectives still adhere to the clinical dose-volume constraints.
For plan quality assessment, the conformity index (CI), homogeneity index (HI), and gradient index (GI) were calculated. For each plan, the monitor units (MUs) were collected to quantify the deliverability and the accuracy of plans, since it reflects the overall modulation level of the plan (31,32). The definitions of above metrics are given by:

RP prediction and pulmonary function assessment
Normal tissue complication probabilities (NTCPs) of grade 2+ and grade 3+ RP for the dose-function metrics (fVX Gy) were calculated using the parameter from Faught’s model (23).
For the ventilation preservation assessment of anatomy-based and function-based plan, , the ventilation loss caused by radiation were calculated. And the of three time points: namely during, three months and one year after SBRT were involved. The population-level dose-function response model of Owen et al. (33) was adopted:
where α is the maximum possible reduction (loss), µ is the dose at which point a maximum reduction of 50% occurs, γ is the overall mean reciprocal of the dose effect slope at the midpoint, EQD2i is the dose corrected to the nominal dose equivalent of 2-Gy per fraction of lung voxel i (minus PTV), and α+wk represents the patient-specific asymptote for voxels in baseline intensity group k. For a specific dose, the function loss is different depending on the baseline ventilation values, while the baseline ventilation value is obtained from the pre-RT ventilation map, . This model was stratified into four groups: , , , and ; with the pre-RT ventilation in lung voxel i. Population-level values (α, wk, µ, γ) were obtained from Table 3 in Owen (33) for SBRT/Mid, SBRT/3-month Post-RT, and SBRT/1-year Post-RT.
By accumulating the ventilation loss value caused by radiation induced lung parenchymal injury in each lung voxel (i) , the total ventilation loss was calculated:
The ventilation preservation was defined by:

Statistical analysis
A paired sample t-test (normally distributed data) or related samples Wilcoxon Signed Rank Test (non-normally distributed data) were utilized for statistically significant difference of the dose metrics between the clinical and FLA plans. SPSS for Windows (version 12.0; SPSS Inc., Chicago, IL, USA) was used for statistical analyses.

Results

Ventilation map and DVH of an example patient
The red arrows in the patient’s dose distribution indicate that the FLA plan can protect H-V lung, depositing doses to medium- or low-ventilation areas. A noticeable decrease in dose within the H-V region is illustrated in Figure 1. Furthermore, it is evident that the dose distribution of the FLA plan is less spherical, particularly regarding the 36 Gy dose line.

Dose-function metrics of clinical and FLA plans
Compared to clinical plans, FLA plans significantly decreased the V4.5 Gy, V8.4 Gy, V12 Gy, V15.4 Gy, V21.8 Gy, as well as the mean dose to the H-V lung (Table 3). For instance, V4.5 Gy decreased by 2.17%, V15.4 Gy by 1.06%, and mean lung dose (MLD)_high has dropped from 398.51 to 361.33 cGy. The dose metrics of M-V lung region also decreased, with V4.5 Gy decreasing from 26.16% to 24.52%, V15.4 Gy from 9.42% to 8.58%, and MLD_medium dropped by 22.33 cGy. The above-mentioned metrics were statistically different (Figure 2). In addition, there was no statistically difference of the dose to the L-V lung region between clinical plans and FLA plans. The V4.5 Gy, V15.4 Gy, and MLD for the L-V lung regions were slightly lower than the clinical plan, while the V21.8 Gy of the FLA plan was slightly higher by 0.28%.

Dose metrics of the target and OARs
As shown in Table 4, the CI was lower in the FLA plans (0.82±0.04) compared to the clinical plans (0.86±0.03) (P<0.05), with the GI showing a similar trend to CI. The HI of the FLA plans was worse than that of the clinical plans, although this difference was not statistically significant. For the total lung, the FLA plans demonstrated a statistically significant reduction in V4.5 Gy, V8.4 Gy, V12 Gy, V15.4 Gy, and MLD by 1.18%, 1.04%, 0.73%, 0.51%, and 13.86 cGy, respectively. Meanwhile, the decrease in V21.8 Gy for the total lung was 0.11%, with no statistically significant difference observed. The maximum dose to the spinal cord and the mean heart dose (MHD) in the FLA plans were slightly higher than those in the clinical plans but remained within dose constraints (29). Additionally, the maximum dose (Dmax) and dose to 0.5 cc (D0.5 cc) of the proximal bronchus, V30 Gy (cc) and V40 Gy (cc) of the chest wall, as well as Dmax and dose to 10 cc (D10 cc) of the great vessels, were slightly higher in the FLA plans compared to the clinical plans (P<0.05). Meanwhile, dose metrics for the esophagus and trachea were comparable between both plans. Finally, there was no statistically significant difference in MU between the two plans, indicating comparable complexity between the FLA and clinical plans.

Comparison of RP probability
Figure 3 presented the risk of RP as a function of dose-function metrics. We chose the models of Faught (23) with the highest AUC. Decreasing fV8.4 Gy from 12.70% to 11.03% resulted in a reduction in the probability of grade 2+ RP from 12.69% to 11.18%, an absolute reduction of 1.51%. fV15.4 Gy dropped from 7.43% to 6.37%, thus leading to a decrease of grade 3+ RP from 3.81% to 3.42%, a relative reduction of 10.24%. The toxicity reductions of FLA plans showed statistical significance (P<0.05).

Preservation of ventilation function
The mean value of ventilation preservations was reported for clinical plans and FLA plans in Table 5. The radiation-induced pulmonary parenchyma injury was taken into account. The Wilcoxon rank test indicated that all these improvements were statistically significant (P<0.05). We also presented the lung injury of the patients during the treatment, 3 months and one year after SBRT, which were 0.02%, 0.16%, and 0.17%, respectively. A trend of increase in the preservation of pulmonary ventilation along with time was observed.

Discussion
4DCT ventilation has been developed to improve clinicians’ knowledge of patients’ pulmonary condition prior to treatment. Our study employed functional imaging to tailor radiation treatment plans to optimize patient outcome and mitigate pulmonary complications. We previously performed the FLA plans in the conventional fractioned RT and a promising result was obtained (19). Due to the higher risk of RP in patients with central lung cancer after SBRT, further work is needed to clearly explore whether FLA approach is clinically applicable. Furthermore, the dose-response incidence model makes it possible for us to theoretically predict the accumulative lung injury in patients and supports the function-guided RT program by weighting voxels based on their risk of injury.
Radiation-induced pulmonary toxicity is important for patients and can be a limiting factor for RT prescription dose, health care, quality of life and applicability of newly introduced adjuvant immunotherapy. Current techniques usually ignore the regional variances of lung function. Yet many patients with lung cancer have significant regional differences in lung function due to smoking or chronic lung complications (34). Therefore, measuring and predicting ventilation values are of clinical importance. FLA plans are evaluated based on the estimated preservation of lung ventilation after treatment, and this prediction raises a direct measurement of clinical endpoints such as RP, radiation fibrosis, and even airway injury, all of which can impair the ventilation. A recently concluded multi-center prospective study of 4DCT ventilation with functional avoidance radiation therapy showed a 14.9% incidence of grade ≥2 RP, which compared favorably with a 25% historical rate (35).
Our data showed that though the conformability and heterogeneity index of the target could be deteriorated, the H-V lung had been successfully spared. Both V4.5 Gy and V15.4 Gy obtained significant decreases, and the ventilation preservation of FLA plans was improved (P<0.05). Although there was slight increase in maximum dose to the spinal cord in functional plans compared with the clinical plans, this was below dose constraint and not clinically significant. Our result fully demonstrated that the FLA planning strategy is clinically feasible, and valuable to provide clinical benefits for patients with central lung cancer.
To achieve the functional lung preservation in this study, some compromises may be inevitable. In terms of the maximum dose of spinal cord and the CI of PTV, clinical plans were superior to FLA plans. This finding is consistent with our previous studies (19). For patients who underwent conventional fractionated RT, the clinical MHD was 18.36±9.18 Gy, and the MHD of FLA plan was 18.58±9.69 Gy. Although statistical differences have been observed for some OARs, the absolute differences in these metrics were minimal (161 cGy for the spinal cord). In the FLA plans, all metrics of OARs adhered to the clinically relevant planning criteria. In addition, we also compared the MU of the two plans and no statistical difference was found, which justified that no extra accelerator burden would be made by the FLA planning strategies.
As can be seen in Table 5, the preservation values of FLA plans are increasing over time, indicating that the benefits of FLA plans are long-term and valid. The ventilation data in this study are statistically significant, yet the absolute difference between the two plans is similar. SBRT has limited damage to the pulmonary parenchyma, due to the relatively small target and the sharp dose gradient. Nevertheless, we evaluated the NTCP probabilities of RP2+ and 3+ of patients (Figure 3), and confirmed that there was significant difference between the two plans (P<0.001). The significant reduction of the probability of RP (RP 2+: 12.69% vs. 11.18%; RP 3+: 3.81% vs. 3.42%) was observed in FLA plans.
Twelve out of 20 patients experienced a decrease of more than 2.0% in HV4.5 Gy, with the most significant reduction being 6.02% (from 33.49% to 27.47%). In contrast, three patients showed an increase in HV4.5 Gy, with the highest increase being 1.10%. For HV8.4 Gy, seven patients had a decrease greater than 2%, with one patient achieving a reduction of 3.52%, corresponding to a 5.96% reduction in the probability of RP 2+. Regarding HV15.4 Gy, eight patients experienced a decrease of more than 1%, with the most significant reduction being 2.73%, which corresponded to a 5.95% decrease in the probability of RP3+. Only one patient experienced a 0.43% increase in HV15.4 Gy. For the small subset of patients who do not benefit from photon lung avoidance RT, alternative modalities such as proton therapy may offer greater clinical benefits. In summary, the functional avoidance protocol proposed in this study may not be suitable for all patients; therefore, it is essential to accumulate sufficient cases through clinical trials to enable effective screening of the target patient population.
There are several strengths of the study, including translating the reduction in dose to functional lung to a decline in the risk of radiation toxicities to central lung cancer patients, and quantifying the preservation of lung function. Our study provides a theoretical basis for exploring ventilation FLA in patients with SBRT for central lung cancer. Our team selected Faught’s prediction model of NTCP for central lung cancer patients following SBRT, as it was the only study to develop NTCP models based on dose-function metrics. Although this model was originally developed using conventional fractionated RT regimens (1.8–2.0 Gy per fraction), comprehensive research investigating its applicability to central lung cancer patients treated with SBRT remains limited. To clarify the differences between various treatment techniques, this study first calculated the 2 Gy equivalent dose (EQD2) using a standard formula in the field of RT, with the specific formula as follows: . On this basis, the optimal parameter combination of the Faught model was further used to predict the probability of RP moving forward, our primary research focus will be on developing NTCP models integrated with dose-function metrics to predict radiation-induced toxicity in patients with central lung cancer.
There are several limitations in our study. First, plan CT was generated by averaging 10 phases of 4DCT, thus implicitly accounting for the respiratory motion. And the uncertainty for registering small H-V regions to the planning CT also led to inaccuracies (36). However, as far as we know, no effective methods for compensating the methodological uncertainties have been proposed by other research teams. Second, in the optimization process, there remains uncertainty about the ideal combination of functional dose metrics and cutoffs of FLA plans, which needs prospective clinical trials. Third, some regions where the determinant of the Jacobian transformation is high might not imply high ventilation but simply high deformation (i.e., sliding of pleura, fissures between lobes). Nevertheless, the Jacobian method remains one of the standard methods for ventilation calculation (37,38). Jacobian determinant-based CT-ventilation has performed well in validation studies with other pulmonary function measures, enabling potential dose reduction and demonstrating an advantage over Hounsfield unit (HU)-based methods (39). Another limitation is the potential for substantial differences in RP predictions when different radiobiological models are applied. The Faught model used in this study is based on traditional fractionated RT. Notably, the single-fraction dose delivered in SBRT is significantly higher than that used in conventional fractionation. Consequently, the original dose-response parameters of the model may not fully capture the mechanisms of lung injury induced by high-dose radiation, and the accuracy of its predictions requires further validation. Additionally, SBRT demands more stringent target volume conformity. While sparing H-V lung regions may cause slight deviations in dose distribution from the ideal plan, there is currently no available data to assess its potential impact on clinical safety. Finally, this study did not account for lung perfusion or indirect radiation damage effect. Irradiated airways and/or vasculature may result in the posttreatment functional decline to lung regions (40-42). Advances in imaging techniques to derive perfusion data directly from 4DCTs and include individual airways as OARs (43) to reduce post-treatment toxicity suggest feasibility of including these aspects in future studies. Owen’s model is based on the SPECT perfusion model, which differs slightly from our application scenario. However, Forghani’s research indicates that the ventilation dose response can diffuse from perfusion dose response for 20–30% of patients. Therefore, we continue to use Owen’s model to assess lung function injury. The functional relationship between 4DCT-based lung function injury and dose is also a topic we are currently investigating. Despite these current limitations, the prospective of FLA planning strategy is promising and this study is valuable in providing data for the RP prediction and ventilation preservation in patients with SBRT for central lung cancer.

Conclusions
Our results suggest that FLA plans can significantly reduce the dose to functional lung regions and minimize damage to the pulmonary parenchyma in patients with central lung cancer, while maintaining target coverage and adhering to dose constraints for most OARs. This approach thereby reduces the probability of RP following SBRT. Additionally, the ventilation function prediction results validate the FLA plans’ ability to mitigate the loss of pulmonary function (with 0.17% ventilation preservation at 1 year) and offer valuable guidance for future clinical trials.

Supplementary
The article’s supplementary files as