BACKGROUND:In intensity modulated proton therapy (IMPT) accurate patient positioning prior to treatment delivery is important to ensure that the prescribed dose is delivered as intended. Typically, in-room 3D imaging is utilized to position the patient based on anatomical features, which can be challenging due to interfractional anatomical changes. PURPOSE:This study aims to introduce a fast automated in-room dose-guided patient positioning (DGPP) algorithm and to evaluate its performance for obtaining accurate target coverage in head-and-neck cancer (HNC) patients. METHODS:The DGPP algorithm employs Moqui, an open-source Monte Carlo (MC) dose engine, for fast proton dose computations. A beam model was configured specific to our institute's treatment machine using measured beam data and it was compared to the clinical treatment planning system (TPS) by means of a 2%/2 mm gamma analysis of 45 HNC treatment plan dose computations. The same cohort of 45 HNC patients, each with five to seven weekly repeat CTs (rCTs) acquired for longitudinal treatment quality assurance, was used to evaluate DGPP performance. For each rCT, the patient position relative to the plan isocenter was optimized using gradient descent to maximize the V 95 % of the primary CTV, 70 Gy(RBE), and elective CTV, 54.25 Gy(RBE), with a 1 mm margin applied to the CTVs during optimization. Six initial positions per rCT were considered, randomly generated within ±7 mm in each translational directions relative to the anatomy-based alignment. The resulting V 95 % values were compared with those obtained using anatomy-based patient positioning (ABPP) in the clinical TPS by evaluating the difference in V 95 % ( Δ V 95 % ). RESULTS:The 3D 2%/2 mm gamma pass rate of Moqui dose computations, with as reference the doses computed in the clinical TPS, was at least 99.1%. By employing DGPP to optimize the rCT positions we found Δ V 95 % = ( 0.1 ± 0.3 ) % and Δ V 95 % = ( - 0.1 ± 0.4 ) % for the CTV 70 Gy(RBE) and CTV 54.24 Gy(RBE), respectively. Considering only the cases that did not reach the clinical goal V 95 % ≥ 98 % with ABPP, the V 95 % CTV 70 Gy(RBE) improved with (1.0 ± 0.8)%. The average optimization time was (149 ± 51) seconds. CONCLUSIONS:We demonstrated that with a gradient descent algorithm for DGPP using a GPU MC proton dose engine, accurate automated HNC patient alignment to the CTV is possible at near-clinically acceptable timescale for pre-treatment patient positioning. Further work is warranted to facilitate clinical adoption.

Background and purpose Mediastinal lymphoma patients are typically young and are at risk of long-term radiation-induced adverse events, making organ of interest (OOI) sparing a key objective of radiotherapy. This study investigated whether reducing setup uncertainty during robust intensity-modulated proton therapy (IMPT) planning is feasible without compromising target coverage and delivered dose robustness. Materials and methods Ten consecutive mediastinal lymphoma patients treated with IMPT were retrospectively analyzed. Treatment plans were generated on the four-dimensional average computed tomography (4DCT) using a 5 mm setup uncertainty (clinical standard). Additional plans with 4 and 3 mm setup uncertainties were created using a dose-mimicking optimization approach while maintaining identical clinical priorities. Robustness was evaluated using a three-dimensional robustness evaluation method (3DREM) and a longitudinal four-dimensional robustness evaluation method (4DREM), incorporating setup and range variations, machine uncertainties, respiratory motion, and weekly anatomical changes. Results All plans achieved adequate target coverage in the 3D-nominal and 4DREM voxel-wise mean dose distributions (V-95 % > 99.6 %). The voxel-wise minimum dose distributions from 4DREM confirmed robust target coverage (V-95 % > 98 %, D-98 % > 95 % of prescription dose) for all setup uncertainties, with one clinically accepted exception. Reducing setup uncertainty resulted in statistically significant reductions in organs of interest doses (p < 0.01), corresponding to median reductions in predicted lifetime acute coronary event risk of 0.17 % and 0.32 % for 4 mm and 3 mm setup uncertainties, respectively (p < 0.01). Conclusion Reducing setup uncertainty during robust IMPT planning for mediastinal lymphoma patients was feasible and improved OOI sparing without compromising delivered dose robustness.

Abstract FLASH-RT is an irradiation modality using Ultra-High-Dose-Rates, where a healthy tissue sparing effect is observed. Different particle types have distinct characteristics in lateral and distal dose deposition, which can affect the amount of healthy tissue sparing by the FLASH effect. However, few comparative studies investigate this impact. In this Monte Carlo-based study, we compare different irradiation modalities in a simple geometrical model and demonstrate how dose, dose-rate, and particle type influence the volume of healthy tissue benefiting from FLASH. The results indicate that less conformal modalities, such as X-rays and VHEE, benefit more from the effect than hadrons. Despite this, FLASH X-rays and FLASH VHEE still deposit higher dose to healthy tissue than conventional hadrons. Assuming a FLASH-triggering dose threshold and dose-rate threshold from literature, we found that hadrons require a target dose of 20 Gy and target dose-rate of 75 Gy/s to achieve healthy tissue sparing, about twice as high as for VHEE and X-rays. The results imply that highly fractionated treatments cannot be applied under current assumptions of FLASH-triggering dose and dose-rate thresholds. Future studies using clinical dose distributions could explore the use of multiple angles, fractionation possibilities, and relative biological effectiveness (RBE) for more specific clinical cases.

Quality assurance is required to ensure safe and reliable use of deep learning (DL)–based intensity modulated proton therapy (IMPT) planning for oropharyngeal cancer patients. This study presents a range of quality assurance measures applied during routine clinical use and include manual adjustments to DL-based plans, independent organ-of-interest dose guidance and multidisciplinary plan review. DL-based plans were clinically acceptable for all 78 patients and achieved 1.1 Gy (RBE) lower parotid dose than manual plans. The implemented quality assurance measures enabled safe routine clinical use of DL-based IMPT planning over 28-months.

BACKGROUND AND PURPOSE:Clinical implementation of online adaptive proton therapy requires rigorous testing of all facility-specific components. This work aimed to develop a multi-purpose phantom-based framework capable of testing the efficacy of diverse adaptive proton therapy workflows. MATERIALS AND METHODS:A custom-built, 3D-printed phantom mimicking different cavity fillings and swelling in the head-and-neck region was designed. The baseline scenario was compared with the anatomically altered scenario employing an initial and an adapted treatment plan. Across six institutes, dose deterioration and recovery were measured using an ionisation chamber and radiochromic films. Tools supporting adaptive proton therapy, i.e. log-file-based dose reconstruction, cone-beam computed tomography (CBCT)-based synthetic CTs, prompt- gamma imaging and proton radiography were included depending on their availability to demonstrate the framework's versatility. RESULTS:Median dose deviation between ionisation chamber measurements and calculation was -0.7%[IQR=1.2%] averaged over all institutes and scenarios without significant differences. Not adapting to the anatomical change led to a 13.9%[IQR=1.5%] median dose decrease inside the target and 169.2%[IQR=11.7%] in the buildup region. Adapting to the changed anatomy, a median 0.2%[IQR=0.8 %] difference in the target and maximum 3.3% in the buildup could be achieved. Gamma-pass-rate (2%, 2mm) analysis of the radiochromic film measurements revealed a comparable trend. Institute-specific tests showed good compatibility with the various adaptive workflows tools. CONCLUSIONS:The combination of the developed multi-purpose phantom and the end-to-end testing framework could reliably detect dosimetric effects of anatomical changes and provided comparable results across six institutes.

Background and purpose:Robust optimization mitigates the effects of range and setup uncertainties in intensity-modulated proton therapy (IMPT) through robustness settings. However, it increases the dose to organs-at-risk (OARs). This study investigated how reducing range robustness (RR) and setup robustness (SR) settings impacts plan adaptation frequency in IMPT of head-and-neck cancer (HNC) patients. Material and methods:A cohort of twenty-six HNC patients treated with IMPT were retrospectively analyzed. Original plans used RR/SR settings of 3%/3 mm. For each patient, four plans were generated with reduced robustness settings. All plans were robustly evaluated on weekly repeat computed tomography scans (rCTs) across 28 error scenarios, applying 2.7-3.2% range and 1 mm setup uncertainty. The failure rate of the voxel-wise minimum D98%≥94% criterion was used as a surrogate for plan adaptation frequency. Differences and equivalence in failure rates between settings were assessed using both two-proportion z-test and Two-One Side Test. Results:Fixing SR at 3 mm, each 1% reduction in RR setting reduced the mean OAR dose by 0.45  Gy (RBE) and normal tissue complication probability (NTCP) by 1.1%-points at a 3.5%-points increase in adaptation triggering rate. Implementing a RR of 3%, each 1 mm reduction in SR settings reduced the OAR dose by 1.0 Gy (RBE) and NTCP by 2.2%-points at a 7.0%-points increase in adaptation triggering rate. Conclusions:Reducing robustness margins decreased OAR dose and NTCP with a manageable increase in plan adaptation; however, larger reductions would require adaptive workflows and implementation of range uncertainty mitigation strategies before safe clinical implementation.

Background and Purpose: Due to superior soft tissue contrast, MRI may provide more prognostic information than CT/PET for outcome prediction. This study aims to compare the prognostic value of MRI with CT and PET in deep learning models for local control, regional control and overall survival in oropharyngeal cancer patients compared to a clinical benchmark model.Materials and Methods: A dataset comprising 266 oropharyngeal cancer patients was assembled. Each patient’s data includes pretreatment axial T1 and a coronal T2 MRI scan, CT and PET scans, gross tumor volume of the primary tumor, clinical parameters and information on local control, regional control and overall survival. Various 2D and 3D convolutional neural networks were trained using images of contoured gross tumor volume with and without a margin for outcome prediction.Results: The 2D models using T2 images within a bounding box region determined by the gross tumor volume achieved concordance index of 0.88 and 0.75 for local control and overall survival prediction, respectively. Additionally, MRI-based models achieved higher concordance indexes than CT- or PET-based models for local control prediction. In comparison to a clinical benchmark model, the T2-based 2D model showed improved local control prediction (concordance index: 0.88 vs. 0.80), and combining the T2 prediction with clinical model improved overall survival prediction (concordance index 0.81 vs. 0.78). Furthermore, the clinical + MRI models showed enhanced performance compared to a clinical routine survival risk stratification system.Conclusions: MRI-based deep learning models can improve prediction of local control and overall survival in oropharyngeal cancer.

BACKGROUND:Synthetic computed tomography (sCT) images generated using deep learning (DL) methods enable the use of on-board CBCT imaging systems for online adaptive proton therapy workflows. However, DL models are susceptible to data drifts, such as changes in the quality of the CBCT images due to software upgrades. PURPOSE:This study aims to assess the effectiveness of transfer learning strategies in addressing changes in the input image quality and to evaluate the sustainability of potential sCT-dependent workflows following CBCT software upgrades. METHOD:Transfer learning strategies were utilized to re-train two existing DL-based sCT models (DCNN and cycleGAN). A dataset comprised 69 head and neck (HNC) patients with paired CBCT-CT images acquired after an image reconstruction software upgrade were used for this study. 60 patients were used for training and validation, and the remaining 9 were reserved for testing. To assess the efficacy of transfer learning strategies, several transfer learning models (TL-models) were trained using various subsets of data, ranging from 5 to 40 image pairs. Additionally, a post-upgrade sCT (New(PU)) model was trained utilizing the complete set of 60 patients to benchmark the TL-models to a post-upgrade-trained model. The synthetic CTs generated from the test set were evaluated using established image quality metrics. Furthermore, dosimetric accuracy was assessed using the patient's clinical treatment plan and our existing clinical NTCP models. RESULTS:Comparison of the average mean absolute error (MAE) between the baseline pre- and post-upgrade (PU) models shows no significant difference. The baseline model exhibited an average MAE of 81.46 ± 49.0 HU and 86.25 ± 14.49 HU for DCNN and cycleGAN, respectively. The TL-05 model demonstrated an average MAE of 69.85 ± 5.9 HU and 95.0 ± 10.95 HU, while the post-upgrade new model had an average MAE of 74.4 ± 12.42 HU and 65.32 ± 10.36 HU for DCNN and cycleGAN, respectively. Additionally, dosimetric quantities showed no significant differences, with mean dose differences ranging from -0.98 ± 3.74% to 2.99 ± 4.74% for DCNN and -0.34 ± 5.45% to 3.15 ± 6.68% for cycleGAN, compared to the post-upgrade new model. Evaluation of the difference between the normal tissue complication probability (∆NTCP) values between the verification CT (rCT) and post-upgrade models showed minimal deviations ranging from -0.001% to -0.03% and 0.0006% to 0.0027% for Grade 2 or higher dysphagia, for DCNN and cycleGAN, respectively. CONCLUSION:Transfer learning strategies, including fine-tuning or freezing feature extraction layers, can minimize disruptions in sCT-dependent workflows. Moreover, the small number of patients required to implement these methods can mitigate extensive downtime due to the limited availability of new data from post-upgrade sources.

BACKGROUND:In the HECKTOR 2022 challenge set [1], several state-of-the-art (SOTA, achieving best performance) deep learning models were introduced for predicting recurrence-free period (RFP) in head and neck cancer patients using PET and CT images. PURPOSE:This study investigates whether a conventional DenseNet architecture, with optimized numbers of layers and image-fusion strategies, could achieve comparable performance as SOTA models. METHODS:The HECKTOR 2022 dataset comprises 489 oropharyngeal cancer (OPC) patients from seven distinct centers. It was randomly divided into a training set (n = 369) and an independent test set (n = 120). Furthermore, an additional dataset of 400 OPC patients, who underwent chemo(radiotherapy) at our center, was employed for external testing. Each patients' data included pre-treatment CT- and PET-scans, manually generated GTV (Gross tumour volume) contours for primary tumors and lymph nodes, and RFP information. The present study compared the performance of DenseNet against three SOTA models developed on the HECKTOR 2022 dataset. RESULTS:When inputting CT, PET and GTV using the early fusion (considering them as different channels of input) approach, DenseNet81 (with 81 layers) obtained an internal test C-index of 0.69, a performance metric comparable with SOTA models. Notably, the removal of GTV from the input data yielded the same internal test C-index of 0.69 while improving the external test C-index from 0.59 to 0.63. Furthermore, compared to PET-only models, when utilizing the late fusion (concatenation of extracted features) with CT and PET, DenseNet81 demonstrated superior C-index values of 0.68 and 0.66 in both internal and external test sets, while using early fusion was better in only the internal test set. CONCLUSIONS:The basic DenseNet architecture with 81 layers demonstrated a predictive performance on par with SOTA models featuring more intricate architectures in the internal test set, and better performance in the external test. The late fusion of CT and PET imaging data yielded superior performance in the external test.

Background and purpose: Deep learning (DL) models can extract prognostic image features from pre-treatment PET/CT scans. The study objective was to explore the potential benefits of incorporating pathologic lymph node (PL) spatial information in addition to that of the primary tumor (PT) in DL-based models for predicting local control (LC), regional control (RC), distant-metastasis-free survival (DMFS), and overall survival (OS) in oropharyngeal cancer (OPC) patients. Materials and methods: The study included 409 OPC patients treated with definitive (chemo)radiotherapy between 2010 and 2022. Patient data, including PET/CT scans, manually contoured PT (GTVp) and PL (GTVln) structures, clinical variables, and endpoints, were collected. Firstly, a DL-based method was employed to segment tumours in PET/CT, resulting in predicted probability maps for PT (TPMp) and PL (TPMln). Secondly, different combinations of CT, PET, manual contours and probability maps from 300 patients were used to train DL-based outcome prediction models for each endpoint through 5-fold cross validation. Model performance, assessed by concordance index (C-index), was evaluated using a test set of 100 patients. Results: Including PL improved the C-index results for all endpoints except LC. For LC, comparable C-indices (around 0.66) were observed between models trained using only PT and those incorporating PL as additional structure. Models trained using PT and PL combined into a single structure achieved the highest C-index of 0.65 and 0.80 for RC and DMFS prediction, respectively. Models trained using these target structures as separate entities achieved the highest C-index of 0.70 for OS. Conclusion: Incorporating lymph node spatial information improved the prediction performance for RC, DMFS, and OS.

Objectives. Breast cancer patients treated with proton therapy receive weekly repeat CTs (rCT) for treatment evaluation. However, plan adaptations are rarely necessary in our department. Therefore, the goal of this study was to develop an online method using surface guided radiotherapy (SGRT) to predict the dosimetric impact of changes in breast anatomy to identify patients that require plan adaptations, in order to reduce the number of rCTs. Approach. Seven breast cancer patients, treated with proton therapy with daily 3D surface-images in treatment position used for SGRT, were included. Clinical proton and backup photon plans were available. A novel in-house developed method, Dose Impact Prediction from Surface Imaging (DIPSI), was used to quantify the deformation between the planning CT (pCT) BODY contour, and the daily 3D SGRT surface images. To validate DIPSI, known isotropic deformations of 2, 4, 6, 8, 10 and 15 mm were introduced in the pCT BODY contour surrounding the clinical target volume (CTV). Clinical proton and photon treatment plans were recalculated including this deformation to evaluate the dosimetric impact of known deformations on the CTV D98%, D2% and average dose (Dmean). Deformation thresholds were determined to detect unacceptable dose deviations, from which the corresponding sensitivity, specificity and false negative rate were determined. Main results. DIPSI showed an accuracy of 0.2 mm and perfect correlation with known deformations (R2 = 1.00). Optimal thresholds to trigger treatment plan adaptation were 6.0 mm for the proton plan and 10.0 mm for the photon plan, resulting in sensitivities >95.5%, specificities >93.9% and false negative rates <1.6% in identifying breast deformations that require plan adaptations. Significance. Breast surface deformations could accurately be determined from daily SGRT images, making them suitable in identifying patients that require plan adaptations in photon and proton therapy potentially reducing the need for rCTs.

Background One of the main challenges of utilizing spot-scanning proton arc therapy (SPArc) is treatment delivery efficiency. Previous studies focus on reducing the number of energy layers by ascending switching to shorten the beam delivery time. However, this is not true of all proton therapy systems. The new energy layer switching system was recently upgraded in the University Medical Center Groningen (UMCG), which enables a fast energy layer ascending switching (ELAS).Purpose We introduce a novel adaptive energy switching SPArc optimization algorithm (SPArc-AES) based on the machine-specific delivery characteristics of proton therapy systems.Methods The SPArc-AES optimization algorithm is based on the polynomial increasing feature of energy layer ascending switching. K-Medoids clustering analysis and simulated annealing algorithm were used to optimize the energy delivery sequence. Ten cases were selected to evaluate the plan quality, plan robustness, and the delivery efficiency compared with the previously SPArc energy sequence optimization algorithm, SPArc_seq.Results Without extra constraints in the energy ascending constraints, the SPArc-AES offers a better plan quality and robustness, while the treatment delivery efficiency was significantly improved compared to the SPArc_seq. More specifically, SPArc-AES effectively shortened the energy layer switching time and the beam delivery time by 34.03% and 31.10%, respectively, while offering better target dose conformality and generally lower dose to organs-at-risk.Conclusions Based on the machine-specific delivery characteristics, we introduced a novel adaptive energy switching algorithm for efficient SPArc optimization, which could significantly improve delivery efficiency while enhancing the plan quality by eliminating no longer necessary constraints on the total number of energy layer ascending switching.

The tissue sparing of ultra-high dose rate irradiation has been partially observed in vitro while controlling oxygen levels. Here, we developed a hypoxia system for high-throughput irradiations that accommodates embedded 3D cultures, such as organoids and spheroids, enabling the study of the effect of oxygen on radiation responses.

BACKGROUND:Intensity-modulated proton therapy (IMPT) holds promise for improving outcomes in head-and-neck cancer (HNC) patients by enhancing organ-at-risk (OAR) sparing. A key challenge in IMPT is ensuring an accurate dose delivery at the distal edge of the tumor, where the steep dose gradients make treatment precision highly sensitive to uncertainties in both proton range and patient setup. Thus, IMPT conformality is increased by incorporating robust margins in the treatment optimization. However, an increment in the plan robustness could lead to an OAR overdosing. Therefore, an accurate distal edge verification during dose delivery is crucial to increase IMPT conformality by reducing optimization settings in treatment planning. PURPOSE:This work aims to evaluate, in a quasi-clinical setting, a novel approach for accurate instantaneous proton beam distal edge verification in IMPT by means of spot-by-spot positron emission tomography (PET) imaging. METHODS:An anthropomorphic head and neck phantom CIRS-731 HN was irradiated at the head and neck region. The targets were defined as 4 cm diameter spheres. A 60-ms delay was introduced between the proton beam spots in order to enable the spot-by-spot coincidence detection of the 511-keV photons resulting from positron annihilation following the positron emission from very short-lived positron-emitting, mainly 12N (T1/2  = 11.0 ms). Additionally, modified irradiations were carried out using solid water slabs of 2 and 5 mm thickness in the beam path to assess the precision of the approach for detecting range deviations. The positron activity range (PAR) was determined from the 50% distal fall-off position of the 1D longitudinal positron activity profile derived from the 2D image reconstructions. Furthermore, Monte Carlo (MC) simulations were performed using an in-house RayStation/GATE MC framework to predict the positron activity images and verify the PAR measurements. RESULTS:PAR measurements achieved a precision between 1.5 and 3.6 mm (at 1.5σ clinical level) at the beam spot level within sub-second time scales. Measured PAR shifts of 1.6-2.1  and 4.2--.7 mm were observed with the 2- and 5-mm thickness range shifters, respectively, aligning with the corresponding proton dose range (PDR) shifts of 1.3-1.8 and 3.9-4.3 mm. The simulated PAR agrees with the measured PARs, showing an average range difference of ∼0.4 mm. CONCLUSION:This study demonstrated the feasibility of instantaneous distal edge verification using PET imaging by introducing beam spot delays during dose delivery. The findings represent a first step toward the clinical implementation of instantaneous in vivo distal edge verification. The approach contributes to the development of real-time range verification aimed at improving IMPT treatments by mitigating range and setup uncertainties, thereby reducing dose to organs-at-risk and ultimately enhancing patient outcomes.

BACKGROUND AND PURPOSE:Deformable image registration (DIR) is widely utilized for dose accumulation, but errors in image registration can compromise its accuracy. This study evaluated the precision of DIR in dose accumulation for parotid gland and proposed a method to correct errors induced by DIR. MATERIALS AND METHODS:This retrospective study included 123 patients with head and neck cancer. For each patient, the accumulated mean dose (Dmean) to the parotid gland was obtained by manual segmentation (ground truth) and two DIR strategies: contour propagation and dose mapping. The normal tissue complication probability (NTCP) model predicting xerostomia was employed to translate accumulated Dmean into NTCP values. Comparisons were made between ground truth and DIR for accumulated Dmean and NTCP. RESULTS:The maximum discrepancy of accumulated Dmean and NTCP between DIR and ground truth was 4.87 Gy and 3.94 %, respectively. The discrepancies of accumulated Dmean between DIR and ground truth were significantly correlated with the discrepancies between accumulated Dmean of ground truth and nominal Dmean. Mid-treatment weekly Dmean discrepancies between contour propagation and manual segmentation showed the capability to correct the accumulated Dmean of DIR and decrease the error of NTCP prediction from 2.89 % and 3.81 % to 1.26 % and 2.04 % for baseline xerostomia Grade 0 and Grade 1-2, respectively. CONCLUSIONS:Significant discrepancies were observed in accumulated Dmean of parotid glands between DIR and manual segmentation in candidates for adaptive radiotherapy. Utilizing mid-treatment CT scans offers a practical solution to correct DIR-induced errors, improving the accuracy of dose accumulation.