VMAT-Based Planning Allows Sparing of a Spatial Dose Pattern Associated with Radiation Pneumonitis in Patients Treated with Radiotherapy for a Locally Advanced Lung Cancer

Simple Summary A sub-region localized in the posterior right lung was significantly associated with the risk of grade ≥ 2 acute pulmonary toxicity in patients with locally advanced lung cancer treated with radiotherapy. Avoiding this sub-region with volumetric-arctherapy-based planification leads to a significant reduction of the predicted APT risk by reclassifying 43.2% (19/44) of the patients. Abstract Introduction: In patients treated with radiotherapy for locally advanced lung cancer, respect for dose constraints to organs at risk (OAR) insufficiently protects patients from acute pulmonary toxicity (APT), such toxicities being associated with a potential impact on the treatment’s completion and the patient’s quality of life. Dosimetric planning does not take into account regional lung functionality. An APT prediction model combining usual dosimetry features with the mean dose (DMeanPmap) received by a voxel-based volume (Pmap) localized in the posterior right lung has been previously developed. A DMeanPmap of ≥30.3 Gy or a predicted APT probability (ProbAPT) of ≥8% were associated with a higher risk of APT. In the present study, the authors aim to demonstrate the possibility of decreasing the DMeanPmap via a volumetric arctherapy (VMAT)-based adapted planning and evaluate the impact on the risk of APT. Methods: Among the 207 patients included in the initial study, only patients who presented with APT of ≥grade 2 and with a probability of APT ≥ 8% based on the prediction model were included. Dosimetry planning was optimized with a new constraint (DMeanPmap < 30.3 Gy) added to the usual constraints. The initial and optimized treatment plans were compared using the t-test for the independent variables and the non-parametric Mann–Whitney U test otherwise, regarding both doses to the OARs and PTV (Planning Target Volume) coverage. Conformity and heterogeneity indexes were also compared. The risk of APT was recalculated using the new dosimetric features and the APT prediction model. Results: Dosimetric optimization was considered successful for 27 out of the 44 included patients (61.4%), meaning the dosimetric constraint on the Pmap region was achieved without compromising the PTV coverage (p = 0.61). The optimization significantly decreased the median DMeanPmap from 28.8 Gy (CI95% 24.2–33.4) to 22.1 Gy (CI95% 18.3–26.0). When recomputing the risk of APT using the new dosimetric features, the optimization significantly reduced the risk of APT (p < 0.0001) by reclassifying 43.2% (19/44) of the patients. Conclusion: Our approach appears to be both easily implementable on a daily basis and efficient at reducing the risk of APT. Regional radiosensitivity should be considered in usual lung dose constraints, opening the possibility of new treatment strategies, such as dose escalation or innovative treatment associations.


Introduction
(Chemo)-radiotherapy is the treatment of reference in locally advanced lung cancer (LALC) not amenable for surgery [1]. Modern radiation techniques, such as intensitymodulated radiation therapy (IMRT) and volumetric arctherapy (VMAT), allow a higher sparing of organs at risk (OAR) with lower doses without compromising the coverage of the planning tumour volume (PTV) [2,3]. Indeed, the VMAT dose sculpting proved to be very useful in the treatment of challenging tumor sites, for which the sparing of strictly close or even inner OARs was the goal in addition to achieving an excellent target coverage [4,5]. Despite the implementation of advanced RT techniques and of stricter dose-volume constraints, acute pulmonary toxicity (APT) remains frequent, with an approximate rate of 5-25% for grade ≥ 2 APT [6][7][8][9]. In a population treated with adjuvant immunotherapy, the APT rate rose to 33.9% for all grades and 3.3% for grades 3-4 APT [10,11].
The high frequency of APT could be explained by the functional heterogeneity of the lungs, these functional regions being mainly localized in the lower lungs [12][13][14][15]. On a daily basis, avoiding these regions could further reduce acute and late pulmonary toxicities [15]. However, although advanced RT techniques, such as VMAT, allow optimization, dosimetric planning does not currently take into account this regional lung heterogeneity. Indeed, the functional mapping of the lungs requires performing costly nuclear imaging, such as pulmonary ventilation and perfusion planar scintigraphy, or positron emission tomography/computed tomography (PET/CT) [15,16].
Based on a voxel-based analysis, as presented by Palma et al. [12,13], we previously identified and prospectively validated the possible role of a cluster of voxels localized in the posterior right lung. Patients with a high mean dose to this sub-region (DMean Pmap ≥ 30.3Gy) were 4.7 times more likely to present with APT ≥ grade 2 when compared to the low-risk patients. Furthermore, when combining the Dmean Pmap with ten other clinical and dosimetric features, a model was previously developed for the prediction of APT [17].
In patients with a predicted probability of APT ≥ grade 2 (Prob APT ) ≥ 8%, the relative risk ratio rose to 12.6. Decreasing the Dmean Pmap could, thus, lower the risk of APT. In this study, we aim to demonstrate the possibility of decreasing the Dmean Pmap via VMAT-based adapted planning and evaluate its impact on the predicted risk of APT.

Population
Data from two separate cohorts were analyzed. The first cohort consisted of all patients treated with VMAT with curative intent for a histologically proven, locally advanced lung cancer (non-small cell or small-cell lung cancer) between 2015 and 2018. The second cohort consisted of the patients treated within the prospective trial TEFRARC (NCT03931356), evaluating the functional impact of lung radiotherapy. Characteristics of these 207 patients (165 from the retrospective and 42 from the prospective cohorts, respectively) were previously presented [9,17].

Radiation Sensitive Sub-Region
All computed tomographies (CTs) were registered to a thoracic phantom using a segmentation-based elastic registration via MIM Maestro (MIM v7.0.0, Cleveland, OH, USA). The segmentation used for registration was a hybrid volume of interest (VOI) consisting of the union of the lungs and the heart. The previously identified Pmap sub-region [17] was then transferred to the patient's CT using the same elastic registration matrix but in reverse. The mean DICE coefficient between the phantom's segmentation and each patient's segmentation was evaluated. For the analysis, the doses were converted to biologically equivalent doses (BED) using the dose map conversion tool in MIM Maestro (MIM v7.0.0, Cleveland, OH, USA). Alpha/beta ratios of 3 for non-tumour volumes and 10 for tumour volumes were used. An explanatory flowchart is provided in Supplementary Figure S1.

Dosimetric Planning
Initial RT was delivered with a prescription dose of 60 to 66 Gy to the PTV, with 95% of the dose covering 95% of the prescription volume while respecting the usual dose constraints to the lungs, heart, and spine [18][19][20][21][22] (Supplementary Table S1).
Planning was optimized with the addition of specific constraints to the Pmap region, namely, a Dmean Pmap < 30.3Gy (BED), while maintaining the PTV coverage and usual dose constraints. The initial Dmean Pmap will now be defined as the Dmean Pmap-Ini , while the optimized Dmean Pmap will be named Dmean Pmap-Opti .
Among the overall cohort, only patients that actually presented APT were considered (n = 45). Three cohorts were then defined, and differentiated on their APT risk, the Dmean Pmap-Ini , and the Dmean Pmap-Opti : - The 8% threshold was based on a previous study [17], defining this cut-off as the most effective for the stratification of patients based on their APT risk. Dose constraints for the initial and optimized plannings are presented in Supplementary Table S1. The DICE score between the Pmap region and the PTV (DICE PTV-Pmap ) was computed for each patient of Cohorts 1 and 2. The association between the DICE PTV-Pmap and the success of optimization was analyzed using the Receiver Operative Characteristics (Area Under the Curve: AUC). An optimal cut-off was defined by maximizing the Youden index.

Dose Map Comparisons
Dose-volume histograms (DVHs) were calculated from the delivered RTPlan using the Pinnacle treatment planning system. Vx y (Gy) will further be defined as the percentage of the volume of interest (VOI = y) receiving x dose (Gy). Dmean y and Dmax y correspond to the mean and maximal dose received by the VOI (y), respectively. On the homolateral (LungH) and contralateral (LungC) lungs, V5, V10, V13, V20, and V30, the Dmax and Dmean were collected. Regarding both lungs (2Lungs), VOI, V13, V20, V30, and the Dmean were considered. Finally, regarding the heart, V30, V40, and the Dmean were extracted. Dmean Pmap and the PTV 95 (percentage of the PTV receiving 95% of the prescription dose) were also analyzed. The significance tests used to estimate the inference between the two dose maps were the t-test for the independent variables and the non-parametric Mann-Whitney U test, otherwise.
The plan quality was assessed through the evaluation of conformity indices and heterogeneity indices. The conformity index (IC) is defined by IC = PIV PTV , where the PIV (Prescription Isodose Volume) represents the volume receiving the prescription dose. The heterogeneity index (HIV) is defined by H IV = D 95% D 5% , where D x% represents the dose received by x% of the PTV.

Prediction Based on the Pmap Model
A Pmap model was previously developed combining 11 features, among which the Dmean Pmap achieved the highest importance (35.8%) [17]. The remaining features were ranked by order of importance: Dmean 2Lungs , V30 2Lungs , Smoking Status, Mean Expiratory Volume/Second (MEVS), Chronic Obstructive Pulmonary Disease (COPD), V10 LungH , AJCC Stage (American Joint Cancer Committee), V5 LungH , Dmean LungH , andV40 Heart . We applied the same Pmap model to the included patients using the optimized parameters. A new probability of APT was thus established. The pre-defined cut-off of 8% was used for patients' classification.

Population
Among the 207 initial patients, 25 and 20 patients were, respectively, considered eligible for Cohorts 1 + 2 and Cohort 3. Among the 25 patients from Cohorts 1 + 2 and before optimization, the mean Dmean Pmap and median Dmean Pmap were 40.4Gy (the standard deviation was 4.4Gy) and 39.3Gy (with a range of 30.6-60.2), respectively. The Pmap model correctly classified 96% of the patients (24/25), with one patient classified at low risk of APT (5.4%) despite further presenting with an APT ≥ grade 2 and a Dmean Pmap of 34.7Gy. This patient was not considered for the rest of the analysis. Among the 20 patients eligible for Cohort 3, no patient was excluded, resulting in a mean Dmean Pmap of 14.6Gy. A flowchart for the patients' selection is presented in Figure 1.

Prediction Based on the Pmap Model
A Pmap model was previously developed combining 11 features, among which the DmeanPmap achieved the highest importance (35.8%) [17]. The remaining features were ranked by order of importance: Dmean2Lungs, V302Lungs, Smoking Status, Mean Expiratory Volume/Second (MEVS), Chronic Obstructive Pulmonary Disease (COPD), V10LungH, AJCC Stage (American Joint Cancer Committee), V5LungH, DmeanLungH, andV40Heart. We applied the same Pmap model to the included patients using the optimized parameters. A new probability of APT was thus established. The pre-defined cut-off of 8% was used for patients' classification.

Population
Among the 207 initial patients, 25 and 20 patients were, respectively, considered eligible for Cohorts 1 + 2 and Cohort 3. Among the 25 patients from Cohorts 1 + 2 and before optimization, the mean DmeanPmap and median DmeanPmap were 40.4Gy (the standard deviation was 4.4Gy) and 39.3Gy (with a range of 30.6-60.2), respectively. The Pmap model correctly classified 96% of the patients (24/25), with one patient classified at low risk of APT (5.4%) despite further presenting with an APT ≥ grade 2 and a DmeanPmap of 34.7Gy. This patient was not considered for the rest of the analysis. Among the 20 patients eligible for Cohort 3, no patient was excluded, resulting in a mean DmeanPmap of 14.6Gy. A flowchart for the patients' selection is presented in Figure 1.
When focusing on Cohort 1, a significant decrease in the median Dmean Pmap was also observed (p < 0.0001), from 36.6Gy (CI95% 32.9-40.5) to 27.7Gy (CI95% 26.7-29.8). The PTV coverage was conserved with no significant differences before (94.4%) and after (96.2%) the optimization (p = 0.65). Graphical comparisons between the initial planning and the optimized planning in each cohort are presented in Figure 2a,c (the Dmean Pmap ) and Figure 2b,d (the PTV 95 coverage). An example of the comparison between the two dose maps for both a successful patient and an unsuccessful patient is presented in Figure 3a-d, respectively. No significant differences were observed for the main dose constraints to the lungs, heart, and spinal cord. For example, a non-significant increase for the V20 2Lungs and V30 2Lungs appeared for the optimized treatment planning, with the V20 2Lungs reaching 26.6% (vs. 22.2%, p = 0.22) and the V30 2Lungs reaching 18.4% (vs. 15.0%, p = 0.43) when compared to the initial dose maps. Regarding the Dmean Heart , a non-significant increase (p = 0.40) was also observed from 9.8Gy (CI95% 5.5-12.3) to 10.9Gy (CI95% 7.6-16.4). Similar results were observed for the sub-set of patients in which the optimization was successful. The detailed results for the dose constraints and PTV coverage are presented in Table 1. Regarding the evaluation of the treatment plans, no significant differences between the initial and the optimized plans were found for both the mean IC  (Table 1).
Similar results were observed for the patients included in Cohort 3, with a reduction of the Dmean Pmap from 14.6Gy to 10.3Gy (p = 0.08) without compromising the PTV coverage or other dosimetric features (Table 1, Figure 2e,f). Regarding the evaluation metrics, no significant differences were found (Table 1).
In Cohorts 1 and 2, with an AUC of 0.83 (p = 0.0001), the DICE PTV-Pmap was significantly correlated with the success of the optimization (Supplementary Figure S2). The optimization was more likely to be successful in patients with a DICE PTV-Pmap ≤ 0.15.

Model Prediction
The previously developed and validated Pmap model was applied to the 44 included patients with the new Dmean Pmap-Opti and DVHs parameters. In Cohort 1, for the 14 patients in which the optimization was successful, 7/14 (50.0%) patients were reclassified at low risk of APT whereas initially predicted by the Pmap model at high risk. For the seven remaining patients (29.2%) from Cohort 1 and the 10 patients (41.6%) from Cohort 2 in which the optimization was unsuccessful, the prediction did not change with the exception of a single patient (Patient #3) from Cohort 2. The optimization, thus, significantly reduced the risk of APT (p = 0.0007) by reclassifying 33.3% (8/24) of the Cohort 1 + 2 patients.
Similarly, the risk of APT was significantly reduced in Cohort 3 by reclassifying 55.0% of the patients (p = 0.004), with 11 patients newly classified at low risk of grade ≥ 2 APT. The detailed results of the prediction model for each plan are presented in Table 2, while the individualized results are presented in Supplementary Table S2. A flowchart explaining the results is presented in Supplementary Figure S3.

Discussion
After an external validation of a prediction model of APT, we aimed to optimize a treatment plan in order to reduce the DmeanPmap and subsequently lower the patients' risk of APT. From the initial cohort of 207 patients, 24 patients had a DmeanPmap above the

Discussion
After an external validation of a prediction model of APT, we aimed to optimize a treatment plan in order to reduce the Dmean Pmap and subsequently lower the patients' risk of APT. From the initial cohort of 207 patients, 24 patients had a Dmean Pmap above the predefined threshold and effectively presented an APT ≥ grade 2. Starting from the initial planning, the optimization was successful for 58.3% of the patients, meaning that the Dmean Pmap significantly decreased under 30.3Gy and that the PTV 95 coverage was maintained. For a subset of these patients (50.0%), the decrease in the Dmean Pmap was associated with a change in the APT risk classification, with these patients being no longer classified at high risk of APT. For the 10 remaining patients (41.7%), no compromise between the Dmean Pmap and the PTV 95 was found. Based on the association of the DICE PTV-Pmap and the success of the optimization (AUC 0.83), these patients appear to have a larger intersection between the PTV and the Pmap region. Interestingly, one patient from Cohort 2 was reclassified at low risk of APT thanks to the decrease of the Dmean Pmap . Despite a threshold having been previously set at 30.3Gy, the Dmean Pmap appears to be predictive of APT even in patients with a lower Dmean Pmap .
In Cohort 3 (20 patients), optimizing the dosimetric planning resulted in a significant reduction of the APT risk, with 55.0% of these patients being reclassified as low risk patients. With a negative predictive value of 96.3% on the testing set [17], these results come with a certain robustness.
Other APT prediction tools were developed, such as radiomics-based tools based on the dose maps. A radiomics-based approach, while very efficient (a Bacc of 0.92 for the risk of APT ≥ grade 2) [23], remains limited to post-dosimetry evaluation. If a patient is categorized at high risk of APT based on the radiomics model, dosimetry planning must be renewed. With the voxel-based sub-region and the prediction model, constraint based on the Dmean can be defined a priori, with a risk deducted by combining the Dmean Pmap and other DVH parameters. It appears as easily implementable in all centers treating lung cancer patients with VMAT.
Previous efforts were made regarding dosimetry planning with an adaptation to the functional sub-regions of the lungs. Higher doses to the perfused functional lung were a stronger predictor of toxicity than the dose to the conventionally measured lung [24]. Based on a cohort of 14 patients treated with 3D conformational radiotherapy, Siva et al. demonstrated proof of the principle that 3D conformational radiotherapy enables functional lung avoidance 12 . Similarly, the same team, using a 68 Ga-ventilation/perfusion PET/CT, was able to optimize treatment planning with an avoidance of the perfused (Q) but not ventilated (V) lung sub-regions [15]. V/Q PET/CT functional volumes were strongly associated with pulmonary function tests [25]. While based on the physiological anatomy of the lungs, this approach requires performing high-resolution PET/CTs and has several limitations, among which are multi-modality registration and a smaller resolution. Furthermore, the clinical benefit of this optimization for a VMAT-based treatment remains yet to be proven [16]. Our approach relies on the retrospective analysis of a model developed on a large cohort of 165 patients and further prospectively validated on a cohort of 42 patients. In the subset of 24 eligible patients for this study, 33.3% of the patients were reclassified as low risks, thanks to the optimization of dosimetry planning.
Apart from the retrospective study and the relatively small cohort, the main limitation of our work is the inability to optimize the treatment for all included patients. For 41.7% of the patients in Cohorts 1 and 2, the optimization was unsuccessful at respecting both the dose constraints to the OARs and the PTV coverage. The APT risk classification only changed for a sub-set of these patients (33.3%), while a substantial number of patients from Cohort 3 benefited from a lower Dmean Pmap . Our approach appears to be both easily implementable on a daily basis and efficient at reducing the risk of APT without performing supplementary imaging. Regional radiosensitivity should be considered in usual lung dose constraints, opening the possibility of an easily implementable adaptive dosimetry plan. Lastly, the findings reported here might be important for an APT risk assessment also and especially in LALC cases amenable to limited re-irradiation to eradicate any residual cancer [26].

Conclusions
The proposed approach appears to be both easily implementable on a daily basis and efficient at reducing the predicted risk of APT. Regional radiosensitivity should be considered in usual lung dose constraints, opening the possibility of new treatment strategies, such as dose escalation or innovative treatment associations.