Background and purpose Non-complete response after chemoradiotherapy in unresectable locally advanced head and neck squamous cell carcinoma (LA-HNSCC) has been associated with poor prognosis. We compared clinical, computed tomography (CT), and magnetic resonance imaging (MRI) radiomics models for predict non-complete response of the primary tumor, and estimated each data type's marginal contribution across all configurations. Materials and methods This retrospective study included 88 patients (67 complete and 21 non-complete responders) treated with definitive chemoradiotherapy. Radiomic features were extracted from pre-treatment contrast-enhanced CT and multi-parametric MRI. Models were optimized under nested leave-one-out cross-validation across 366 configurations spanning imbalance handling, feature selection, and classifier. Each data type's marginal contribution was estimated as the mean paired AUC difference versus a parsimonious clinical model. Early and late fusion were tested, and clinical utility was assessed by exploratory decision curve analysis. Results In a best-configuration comparison, the MRI model achieved the highest discrimination (AUC 0.91, 95% CI 0.81–0.98) but did not significantly outperform the parsimonious clinical model (AUC 0.75, DeLong p = 0.058). Across all configurations, no single-modality or feature-level fusion exceeded the clinical baseline, each performing significantly below it (all p < 0.05). Only decision-level fusion combining clinical and imaging data was positive, reaching significance for clinical plus CT (mean ΔAUC +0.059, p = 0.031). Conclusions A properly specified clinical model was a strong benchmark for radiomics. Single-best pipeline results can overstate radiomic value in small cohorts, and biascontrolled evaluation with external validation is essential before clinical use.
Background and purpose Total skin electron beam therapy (TSEBT) is an established radiation treatment modality for the management of cutaneous T-cell lymphoma (CTCL). In recent years, there has been an increasing interest in dose de-escalation strategies aimed at lowering treatment-induced side effects whilst maintaining efficacy. This evolution in clinical practice has highlighted the need for consistent technical standards to ensure treatment quality and comparability across medical institutions. A unified set of recommendations for TSEBT in the modern era was sought by the European Society for Radiotherapy and Oncology (ESTRO) Guidelines Sub-Group on Physics, including physical requirements, with the aim of harmonizing clinical practice, promoting homogeneity for clinical trials, and facilitating wider adoption of the technique. Material and methods The ESTRO Guidelines Committee developed these recommendations using a Delphi consensus methodology. An expert panel consisting of radiation oncologists and medical physicists with specific expertise in the radiation treatment of primary cutaneous lymphoma was convened. Panel members completed a 14-question survey focusing on key technical aspects of TSEBT. Results Consensus was achieved on the principal physical and technical components of TSEBT delivery, including prescribed dose levels, field dimensions, treatment techniques, and QA requirements. Areas of agreement reflect current best practices and address sources of variability that may impact treatment reproducibility. Conclusions ESTRO recommendations provide a standardised framework for the delivery of TSEBT in the management of CTCL, supporting safe implementation and consistency across medical institutions in the context of reduced-dose treatment approaches.
Background and purpose Adaptive radiotherapy for head and neck cancer is resource-intensive, and existing geometric triggers lack a quantitative dose basis. We developed a model based on dose differences derived from cone-beam computed tomography to predict adaptive radiotherapy benefit. Materials and methods Seventy-four patients with head and neck cancer treated with sequential-phase intensity-modulated radiotherapy and offline adaptive radiotherapy were analyzed. Patients were labeled as high or low adaptive radiotherapy benefit by k-means clustering of dose features derived from the resimulation computed tomography. A multivariable logistic regression classifier was trained on percent mean dose differences measured on the week 3 cone-beam computed tomography for five normal tissues (both parotid glands, both submandibular glands, and the oral cavity) and evaluated by leave-one-out cross-validation, alongside a parallel classifier using normal tissue volume changes. Results Dose differences correlated with adaptive radiotherapy benefit for all five normal tissues (Spearman ρs = −0.76 to −0.43; all p < 0.01, Jonckheere–Terpstra trend test), whereas volume changes showed no significant association (all p > 0.10). The dose model achieved an area under the receiver operating characteristic curve of 0.78 versus 0.54 for the volume model. On univariate analysis the ipsilateral parotid gland was the strongest single-organ predictor, whereas the multivariable model assigned the largest standardized coefficient to the ipsilateral submandibular gland, reflecting collinearity between the parotid glands. Conclusions Dose differences derived from cone-beam computed tomography provide an objective, quantitative basis for predicting adaptive radiotherapy benefit in head and neck cancer, offering a practical alternative to geometric triggers and supporting selective, resource-efficient adaptation.
Background and purpose:Accurate stopping-power ratio (SPR) estimation is essential for treatment planning of proton therapy. However, SPR accuracy is influenced by object size and position due to beam hardening. This study compared proton SPR estimation using photon-counting computed tomography (PCCT), dual-energy CT (DECT), and single-energy CT (SECT) across varying phantom sizes and insert positions. Materials and methods:Two phantoms with tissue-equivalent inserts were scanned using PCCT, DECT, and SECT scanners. Based on the PCCT and DECT scans, virtual monoenergetic images (VMIs) were created. Two SPR estimation methods were used: a DECT-based method and a SECT-based Hounsfield look-up table (HLUT). SPRs were calibrated on the Gammex Advanced Electron Density phantom and evaluated on a cylindrical phantom with four diameters. To assess size-dependency of the estimated SPR, tissue-equivalent bone inserts were placed centrally in the evaluation phantom, for each of the four diameters. To assess position-dependent SPR uncertainty, the inserts were placed at different distances from the centre of the evaluation phantom with fixed diameters. Results:For smaller phantom diameters, SPR estimation accuracy was comparable across PCCT, DECT, and SECT. For larger phantom diameters (30 and 40 cm), SPRs estimated based on SECT and DECT showed position-dependency. PCCT demonstrated the most stable SPR estimations across positions, with deviations between 0.9% and 2.1%. SPR estimations were more stable using a HLUT than the DECT-based SPR method. Conclusions:SPR estimated based on PCCT-VMI with a HLUT showed reduced impact of beam hardening effects related to size and position.
Background and purpose Differences in dose delivered during carbon radiotherapy are due largely to variations in relative biological effectiveness (RBE), which can be calculated by several models. While RBE can theoretically be calculated using Microdosimetric Kinetic Model (MKM) and measured microdosimetric quantities, clinical implementation relies on lookup tables. There exist no means of measuring RBE by MKM or other common models, including Repair Misrepair Fixation (RMF) and Local Effect Model-I (LEM). This study investigated estimating RBE using a single microdosimetric measurement for each model, allowing measurement-based validation and inter-model comparison. Materials and methods Monte Carlo simulations were performed at 260 positions across nine carbon beams, from which microdosimetric quantities (y∗) and linear (α) and quadratic (β) RBE parameters were derived using each model definition. To estimate RBE from microdosimetric input, α and β were fit as functions of y∗for two test beams. The fits were applied to microdosimetric data from the remaining simulated beams and experimental measurements to validate resulting RBE predictions against full Monte Carlo-calculated RBE. Results Modelled RBE calculated using simulated input parameters versus estimated with the measurement-based fit were compared and found to be within ±10% accuracy in 97% (MKM), 100% (LEM), and 98% (RMF) of points. Average estimation uncertainty was 2.6% (MKM), 1.7% (LEM), and 4.3% (RMF). Conclusions While modelled RBE has extensive associated uncertainty, this study demonstrated that it could be estimated with reasonable accuracy using a standardized measurement-based framework. Results suggested that microdosimetry can be used as a quality assurance tool to characterize carbon fields and verify correct implementation of RBE models.
BACKGROUND AND PURPOSE:Synthetic computed tomography (sCT) from magnetic resonance imaging (MRI) enables computed tomography (CT)-free cranial radiotherapy planning, yet no consensus exists on which head phantoms are usable for sCT quality assurance (QA). This study benchmarked head phantoms to identify the most promising candidates for cranial sCT QA and the design characteristics associated with favorable sCT-to-conventional CT agreement. MATERIALS AND METHODS:Ten head phantoms underwent clinical MRI (1.5 T) and CT imaging. Two vendor-provided sCT algorithms (2D slice-based, 3D volume-based) were applied to identical input. sCT and CT were compared using Dice similarity coefficient (DSC), 95th-percentile Hausdorff distance (HD95), CT number mean error (ME) within both a self-mask (sCT-derived) and a common-mask (CT-derived), and percentage dose deviations (%ΔD2%, %ΔD98%, %ΔDmean) across six beam configurations. Metrics were interpreted against a priori three-tier acceptance thresholds. RESULTS:Phantoms without internal skull-equivalent structures showed bone DSC ≤ 0.001 and common-mask bone ME≤ -1100 HU. Phantoms with skull-equivalent anatomy achieved bone DSC of 0.24-0.71, bone HD95 of 4.2-24.4 mm, and self-mask bone ME between -95 and +470 HU across both algorithms. Common-mask bone ME in skull-containing phantoms ranged from -198 to -1115 HU across both algorithms. Soft-tissue ME stayed within ±120 HU across skull-containing phantoms for both masks. Average absolute |%ΔDmean| was 3.2% (2D) and 3.7% (3D) without significant inter-algorithm difference (p = 0.49). CONCLUSIONS:Internal skull-equivalent anatomy is a necessary but not sufficient design characteristic; low-susceptibility and MR-compatible materials proved essential. Joint evaluation of geometric, CT number, and dose agreement was required to rank phantom performance. No phantom met all criteria simultaneously.
Background and purpose Ultra-high dose rate (UHDR) irradiation has been demonstrated to reduce normal tissue damage compared to conventional dose rates, while maintaining tumor response (FLASH-effect). UHDR could potentially freeze intra-fraction breathing motion, enabling margin reduction for moving tumors when accurately timed. Targeting the optimal breathing-phase could reduce organ-at-risk (OAR) dose and side effects. In this treatment planning study, the optimal phase(s) for UHDR proton therapy were identified and potential benefits were evaluated. Materials and methods Twenty lung cancer patients, previously treated with 66 GyE/24 or 60 GyE/30 fractions, were included. Four-dimensional computed tomography (4D-CT) scans with clinical target and OAR delineations were used to create new treatment plans for individual 4D-CT phases, one-phase plans (OPP) and multiple-phase plans (MPP). Clinically relevant dose-volume parameters and normal tissue complication probabilities (NTCP) were evaluated. Results Phase-targeted proton therapy (PTPT) significantly reduced OAR dose. The largest reductions were achieved with OPP, while MPP showed smaller reductions. With OPP, mean lung dose (mean: −0.7 GyE, range: −1.7 to 0.3 GyE), mean heart dose (mean: −0.4 GyE, range: −1.4 to 0.5 GyE), and mean esophagus dose (mean: −0.9 GyE, range: −6.0 to 0 GyE) were reduced, with most reductions in the 0%, 40%, and 70% phases, respectively. NTCP values indicated reduced complication probabilities across all phases, with additional gains for optimal phases. Conclusions With PTPT, OAR dose may be reduced with potential clinical benefit across all phases. The optimal phase depended on the endpoint, suggesting patient-specific phase targeting could further improve outcomes. Future research should address phase targetability, residual variation and required robustness.
Background and purpose:Cone-Beam Computed Tomography (CBCT) based synthetic CT is being increasingly used for post-delivery dose computation or adaptive workflows in radiotherapy. However, the limited Field-of-View of the CBCT can cause inaccuracies when tissues fall outside the Field of View. This study evaluates the integration of surface-guided radiotherapy information to enhance synthetic CT generation for breast cases with missing tissues on CBCT. Materials and Methods:A retrospective analysis was performed on 20 breast patients. CBCT volumes were acquired on a linac, with simultaneous surface scans from three cameras. A full Field of View CBCT was used to generate a reference synthetic CT. A smaller Field of View CBCT was reconstructed from the same raw data to create two synthetic CTs: a standard synthetic CT and a surface-guided synthetic CT, for which surface scan information was incorporated into the algorithm to guide reconstruction outside the Field of View. We compared the image similarity, external contour geometry, and dose distribution. Results:The surface-guided synthetic CT showed superior agreement with the reference for nine tested metrics. Specifically, it showed an average Mean Squared Error reduction of 1963 HU 2 ( p = 0.002 ), revealing hot spots in some cases. Conclusion:Integrating surface scan information into deformable registration improves synthetic CT generation for CBCT with limited Field of View, yielding more accurate images and enhanced dose distribution precision for breast cases. Future work will explore other sites, and potential applications for adaptive radiotherapy.
Background and purpose:Artificial intelligence (AI)-based contouring reduces delineation workload, yet manual corrections remain required. Reliable, scalable quantification of structure-level editing time is needed to evaluate clinical usability beyond geometric similarity metrics as geometric metrics may not reflect how tooling impacts edit time. Materials and methods:This retrospective observational study analysed 3083 AI-generated, edited structures created in routine practice. Structure-modifying actions were extracted from the treatment planning system audit log database and converted to per-structure editing time by summing inter-event intervals. Automatic times were benchmarked against manual time recordings for 54 structures across 8 patient cases using Lin's concordance correlation coefficient (CCC). Added path length (APL) between AI and corrected contours was computed, and the associations between editing time, APL and tooling (interpolation use) were investigated using Spearman's correlation coefficient. Results:Audit log-derived editing time agreed well with manual timing (CCC = 0.93). In the full cohort, 3083/24970 structures were edited (12% correction rate). Brush tools were used in 80% and contour interpolation in 42% of edited structures. Median editing time and APL were higher with interpolation than without (81 s vs 58 s; 232 mm vs 18 mm; both p < 0.05), and correlation between editing time and APL was fair (r = 0.38 vs 0.51). Conclusions:Vendor audit logs could be translated into accurate structure-level editing time estimates, enabling low-burden monitoring of manual workload. Editing strategy influenced time-geometry relationships, supporting log-based time endpoints alongside geometric metrics for meaningful auto-contouring evaluation.
Background and purpose:Triggered kilovoltage images acquired during radiotherapy delivery remain underexploited for quantitative intrafraction motion analysis. This study aimed to develop and evaluate a retrospective workflow for intrafraction motion verification using triggered kilovoltage images during bone metastases stereotactic body radiotherapy. Materials and methods:Triggered kilovoltage images and corresponding treatment-planning data were retrospectively extracted from 49 patients treated in two radiotherapy centers. For each triggered image, a digitally reconstructed radiograph was generated from planning computed tomography dataset. After preprocessing, rigid in-plane registration between triggered images and digitally reconstructed radiographs was performed. The workflow was evaluated using phantom experiments with known physical displacements and patient datasets with simulated shifts, then applied to 8717 triggered images acquired during 183 treatment fractions. Results:Correct registration was obtained for 7105 images (81.5%), while 1612 images (18.5%) were classified as aberrant and excluded from motion analysis. After exclusion, the mean displacement vector magnitude was 1.2 ± 0.8 mm. Overall, 85% of measurements showed displacements ≤2 mm and 98% showed displacements ≤3 mm. Sustained displacement events exceeding 2 mm over three consecutive images were observed in 405 of 7105 images (5.7%) and in 46 of 159 analyzable fractions (28.9%). The mean computation time per image registration was 1699 ± 504 ms. Conclusions:Triggered kilovoltage images contain quantitative information that can support retrospective intrafraction motion assessment during bone metastases stereotactic body radiotherapy. This workflow may help identify fractions requiring further review and support offline quality assurance, although prospective validation is required before clinical implementation.
Background and Purpose: Accurate tumour delineation is key in radiotherapy workflows. Concurrently, diffuse tumour types are inherently difficult to delineate, rendering contour uncertainty information highly valuable for downstream treatment planning. In this study, we investigated whether uncertainty in manually delineated tumour contours can be inferred from clinician behaviour (eye and mouse movements) and image-derived indicators. Materials and Methods: In a two-stage controlled experiment, 36 clinical imaging experts described and manually delineated brain tumours on T2-FLAIR (fluid-attenuated inversion recovery) MRI scans, while their eye and mouse movements were recorded. Inter-observer contour variability was used as a proxy for contour uncertainty. In addition, we extracted image-derived features, including U-Net saliency maps, pixel-wise U-Net segmentation probabilities, and image entropy. To assess and predict contour uncertainty, we applied mixed-effects models and machine learning algorithms. Results: Higher contour uncertainty was associated with higher saccade velocities and greater fixation density. Uncertainty was also significantly correlated with segmentation error and image-derived features (p < 0.05). A random forest regressor, combining behavioural and image-derived features, explained 39% of the variance in contour uncertainty. Notably, features derived from downsampling layers were the strongest predictors, despite displaying the lowest saliency overlap with human attention. Conclusions: Our results suggest that uncertainty in manually delineated tumour contours can be estimated using behavioural and image-derived features, without explicit uncertainty annotations. While clinical deployment will require validation under realistic acquisition conditions and with specialty experts, these findings established the feasibility of passively inferred uncertainty as a foundation for future uncertainty-aware delineation tools.
Background and purpose To determine whether longitudinal changes on routine thoracic computed tomography (CT) predict overall survival (OS) in non-small cell lung cancer (NSCLC) after curative radiotherapy and identify dose predictors of adverse tissue changes. Materials and methods We performed a retrospective, single-centre study of 231 stage I-IV NSCLC patients who had at least two follow-up scans. In total, 2708 CT scans were analysed (median follow-up, 22 months; range, 1–97). Automated segmentation quantified left ventricular (LV) myocardium, L1 skeletal muscle (SKM) and fat volumes. For each tissue, baseline-normalised trajectories were used to compute monthly rates of change (“velocity”), and non-linear associations with OS were evaluated. Logistic regression identified dose metrics associated with adverse tissue change. Results SKM velocity stratified OS (C-index, 0.70): SKM loss < −0.4%/month vs ≥ −0.4%/month, HR 4.41 (95% CI 2.46–7.91, p < 0.005). LV-myocardial mass velocity showed a U-shaped relationship with OS (C-index, 0.75): atrophy < −0.3%/month vs stable, HR 4.12 (95% CI 1.65–10.27, p < 0.005); hypertrophy >0.3%/month vs stable, HR 7.90 (95% CI 2.82–22.15, p < 0.005). Dose predictors included oesophagus V10Gy for SKM loss (OR, 1.40; p = 0.03), Aorta V5Gy for myocardial atrophy (OR, 1.71; p = 0.02), and right-atrium V10Gy for myocardial hypertrophy (OR, 1.63; p = 0.02). Conclusion Longitudinal CT biomarkers, particularly SKM loss rate and deviation in LV-myocardial mass, are associated with OS after curative NSCLC irradiation. These findings require validation in multicentre studies with more complete clinical information.
Background and Purpose:This study aimed to identify dose-volume parameters and anatomical features associated with the implementation of online dose re-optimization in magnetic resonance-guided adaptive radiotherapy (ART) for prostate cancer and to evaluate its impact on target coverage and organs-at-risk dose-volume parameters. Materials and Methods:Treatment plans from 150 fractions (with and without dose re-optimization) in 30 patients receiving five-fraction stereotactic body radiotherapy were retrospectively analyzed. Evaluated parameters included planning target volume (PTV) V 100%, bladder and rectum dose-volume parameters, volumetric changes, and relationship between changes in target coverage. Results:Online dose re-optimization was performed in 60% of fractions and became more frequent in later fractions (p = 0.02). Compared with reference plans, non-ART plans showed a 3.7% lower median PTV V 100% (p < 0.01), and 65.3% of fractions failed to achieve the 95% coverage threshold. The target volume increased significantly after the third fraction (p < 0.05). Target volume expansion was associated with reductions in PTV V 100% in non-ART plans, while this relationship was substantially weakened after dose re-optimization. Although bladder and rectum dose-volume parameters increased significantly in non-ART plans, all median values remained within predefined dose constraints. Conclusions:Progressive target-volume expansion appears to be the primary anatomical factor associated with deterioration in target coverage. A reduction in PTV V 100%, particularly below 95%, represented a clinically meaningful indicator for online dose re-optimization. These findings suggested that online dose re-optimization remained important for maintaining target coverage and limiting gradual deterioration of organs-at-risk dose-volume parameters, even when predicted plans satisfied clinical constraints.
Background and purpose:Proton and photon plan comparison is often the gateway to proton therapy for head and neck cancer (HNC). Manual planning is resource demanding, whereas automated planning may efficiently create consistent treatment plans. An automated rule-based intensity-modulated proton therapy (IMPT) plan optimisation script was developed within a commercial treatment planning system for comparative planning. This study benchmarked automated against clinical plans and evaluated whether a standard or patient-specific field set-up was necessary. Materials and methods:Twenty HNC patients previously treated with IMPT using manual plans were included. For each patient, two automated IMPT plans were generated, differing in field set-up: one with a standard five-field set-up and one matching the clinical field set-up. Robustness, dose distribution, and normal tissue complication probability (NTCP) for dysphagia and xerostomia were compared. Results:On average, automated IMPT plans were created in 64 min. Compared with clinical plans, both automated plan types, generally, reduced organ of interest mean doses, with median differences of up to 3.3 Gy (glottis). Automated plans also had lower median NTCP for dysphagia of 0.1%-points (interquartile range (IQR): -0.8-0.9) and 0.3%-points (IQR: 0.0-0.6) for standard and clinical field set-ups, respectively. For xerostomia, the corresponding results were 0.3%-points (IQR: 0.0-1.5) and 0.3%-points (IQR: 0.1-0.8), respectively. Conclusions:The IMPT plan optimisation script created plans with comparable robustness, dose distribution, and NTCP to clinical plans and is applicable for selecting HNC patients for IMPT. Additionally, a standard field set-up was adequate for comparative planning for most patients.
Background and Purpose: For thoracic and upper abdominal tumours, respiratory motion management is essential; dynamic tumour-tracking radiotherapy (DTT-RT) is one such approach, but it typically requires fiducial marker (FM) insertion. We evaluated intra-respiratory variation in the tumour–diaphragm geometric association and its potential as a surrogate for DTT-RT.Materials and Methods: Data from 22 patients (11 lung cancer [LC]; 11 locally advanced pancreatic cancer [LAPC]), treated with FM-based DTT-RT, were analysed. Ten-phase, four-dimensional computed tomography was used to quantify the 3D gross tumour volume (GTV)–FM displacement, 3D GTV–diaphragm displacement, and 3D GTV centroid displacement under free-breathing (FB). Phase-wise relative displacements for GTV–FM, GTV–diaphragm, and the GTV centroid under FB were calculated along the left–right (LR), anterior–posterior (AP), and superior–inferior (SI) axes as differences from the end-expiratory phase.Results: In LC, the mean 3D GTV–FM displacement, 3D GTV–diaphragm displacement, and 3D GTV centroid displacement under FB across all phases were 1.0 (range, 0–6.9), 4.6 (0.1–19.4), and 5.8 (0.1–31.8) mm, respectively (p < 0.001). In LAPC, corresponding values were 1.6 (0.2–4.7), 4.5 (0.3–21.8), and 3.2 (0.4–17.0) mm (p < 0.001). Phase-wise relative GTV–diaphragm displacement was largest in the SI direction in LC and in the AP direction in LAPC, with patterns primarily driven by inspiratory-phase variations.Conclusions: As a surrogate, the diaphragm exhibited larger intra-respiratory variation than FMs, primarily driven by inspiratory displacements. Incorporating phase-dependent tumour–diaphragm geometry may support markerless, diaphragm-based DTT for LC and LAPC.
Background and purpose Daily auto-contouring remains a workflow bottleneck in magnetic resonance-guided adaptive prostate radiotherapy (MRgRT). This study proposes and clinically validates a novel deep learning-enhanced deformable image registration (DIR) solution to accelerate this critical step. Materials and Methods A hybrid framework combining a 3D nnU-Net segmenting bladder/rectum on planning/daily MRI with an in-house DIR algorithm was implemented for 5-fraction prostate MRgRT (5×7.25 Gy) on an MR-Linac. The DIR uses nnU-Net contours to propagate target and organs-of-interest structures. The solution was clinically deployed and evaluated in 275 patients/1375 fractions. Results: Evaluation following clinical introduction, has shown a median contouring time of ≈190 s, halving the time required by the previously-employed vendor-provided solution. Quantitative evaluation showed high agreement with clinically approved contours: Dice similarity coefficients >0.9 and 95th percentile Hausdorff distances <2.0 mm for most structures. Conclusions: The implemented solution demonstrated reliable, high-accuracy daily auto-contouring, significantly accelerating MRgRT workflows. It has become our institutional standard for prostate treatments. Future work will extend this approach to additional treatment sites and modalities.
Background and Purpose:Synthetic computed tomography (sCT) enables magnetic resonance imaging (MRI)-only radiotherapy by providing electron density information for dose calculation. While sCT-based planning has been proven sufficiently accurate for convolution-based dose calculation algorithms, linear Boltzmann transport equation (LBTE)-based approaches exhibit higher sensitivity to tissue heterogeneities. This study evaluated the accuracy of both dose calculation algorithm types for MRI-only in prostate and glioma radiotherapy. Materials and methods:Clinical treatment plans for thirty-nine prostate cancer patients and seventeen glioma patients were recalculated on atlas- and deep learning (DL)-based sCTs and conventional CT using convolution- and LBTE-based algorithms. Target dose metrics were analyzed as relative differences, while organs of interest (OOI) dose criteria were evaluated as absolute differences between sCT- and CT-based calculations. Statistical significance was assessed using paired t-tests (α = 0.05). Results:For target volumes, statistically significant differences between sCT- and CT-based calculations were observed for both dose calculation algorithms (p < 0.001). In prostate patients, mean differences in target dose metrics were ≤ 1.4% for the LBTE-based algorithm and ≤ 0.5% for the convolution-based algorithm. Slightly smaller dose differences were observed for glioma patients, ≤ 0.8% and ≤ 0.5%, respectively. OOI dose differences were small (≤ 0.6%). The LBTE-based algorithm consistently yielded the largest dose differences between CT and sCT. Conclusions:Although the LBTE-based approach demonstrated higher sensitivity to sCT-related CT number discrepancies than the convolution-based dose calculation algorithm, the resulting dose differences remained within clinically acceptable limits, supporting the robustness of MRI-only radiotherapy for both atlas- and DL-based sCTs.
Background and purpose:Proton minibeam radiotherapy (pMBRT) uses a 1D array of narrow beams to widen the therapeutic window of difficult-to-treat tumors. With the aim of identifying tumor locations that could benefit most from pMBRT, we evaluated how irradiation parameters shape 3D dose distributions. Materials and methods:Monte Carlo simulations were used to compute dose distributions in water for different proton energies, beam widths (bws) and center-to-center distances (ctcs). Optimal parameter combinations were selected according to three criteria: (i) minimization of the bw in normal tissue; (ii) maximization of the valley dose in the target; and (iii) minimization of the peak dose in normal tissue. Results:For shallow tumors (≤ 2 cm), 0.5 mm beams with ctc = 3bw kept normal-tissue widths < 1 mm with Bragg-peak-to-entrance dose ratio (BEDR) > 1. For intermediate and deep-seated tumors (8-20 cm), 1.0-1.5 mm beams with ctc = 4-5bw kept normal-tissue widths < 7 mm with peak-to-valley dose ratio (PVDR) > 3 and achieved lateral dose homogeneity in the target. For very deep-seated tumors (> 20 cm), 2 mm beams with ctc = 4bw maintained normal-tissue widths < 10 mm with PVDR > 3 at the cost of BEDR ∼ 0.5. Conclusion:pMBRT may offer advantages over conventional proton therapy and GRID therapy for treating shallow and deep-seated tumors. For very deep-seated tumors (> 20 cm), feasibility will depend on tumor size and proximity of organs at risk.
Background and Purpose:Stereotactic Radiosurgery with robotic image-guided systems is promising for recurrent glioblastoma (rGBM), but long delivery times may increase morbidity and reduce accuracy. Proton single-energy Bragg-peak FLASH-RT (SEBP-FLASH) may address these limits by delivering each field in <1 s, with potential for normal-tissue sparing through the FLASH effect. Materials and Methods:The SEBP-FLASH technique was used to reoptimize treatment for rGBM patients treated with robotic image-guided systems. Patient-specific aperture devices were incorporated into the SEBP-FLASH technique for sharper penumbra. The gross tumor volume (GTV) coverage, conformity index (CI), dose gradient index (GI50 and GI30), and dose-volume histogram (DVHs), were evaluated. Additionally, ultra-high dose-rate ratios in critical organs-at-risk were assessed. Results:SEBP-FLASH successfully met all clinical objectives. Both SEBP-FLASH and robotic techniques showed comparable performance, with GI50 and GI30 values of 2.9 ± 0.5 and 6.2 ± 1.8 for SEBP-FLASH, versus 2.7 ± 0.2 and 5.5 ± 0.5 for robotic (p-values at 0.085 and 0.189). SEBP-FLASH demonstrated superior conformality, achieving a CI of 1.4 ± 0.2 compared to CK's 1.9 ± 0.3 (p-value 0.001). All other metrics were similar across both techniques. For the brain, the FLASH dose rate coverage (V40 Gy/s) exceeded 80% under a 5 Gy dose threshold and minimum monitor unit of 400. Conclusions:This study evaluated the dose-volume characteristics of SEBP-FLASH and robotic image-guided systems and found broadly comparable dose distributions, suggesting that SEBP-FLASH may be a feasible option for rGBM reirradiation. SEBP-FLASH also showed potential advantages in conformity, normal tissue sparing, and reduced on-table time.
We present a multimodal deep learning model for segmenting 25 organs defined in the European Particle Therapy Network (EPTN) international neurological contouring atlas. Multiple input configurations were evaluated on 74 patients using 5-fold cross-validation (59 training, 14–15 per fold for evaluation), with each patient assessed once on unseen data. The dual-input model combining contrast-enhanced T1-weighted magnetic resonance (MR) and computed tomography (CT) achieved the best overall results (median Dice of 0.80, median surface Dice of 0.84), with no added benefit from T2 FLAIR.
