BACKGROUND:Accurate dose-volume histogram (DVH) prediction is essential for high-quality and automated radiotherapy (RT) planning. However, existing deep learning methods primarily focus on voxel-level dose prediction followed by post-processing to extract DVHs, a workflow that can introduce cumulative errors and limit clinical interpretability. PURPOSE:In this study, we propose a novel deep learning framework that directly predicts full DVHs for multiple organs-at-risk (OARs) from computed tomography (CT) and structure contours. METHODS:Our method combines a three-dimensional convolutional neural network (CNN) for anatomical feature extraction with a graph neural network (GNN) that models each DVH as a structured graph, enabling sequential dose-volume dependencies to be learned explicitly. The model was extensively validated on nasopharyngeal cancer cases from an external institute and rectal cancer cases from our affiliated hospital, encompassing treatments with TomoTherapy and volumetric modulated arc therapy (VMAT). We additionally conducted a model-guided clinical DVH assessment on suboptimal cases to demonstrate the method's clinical utility RESULTS: The proposed method achieved low dose-wise prediction errors and outperformed previous approach. Specifically, it reduced the mean dose error for the brain stem from 1.24 Gy to 0.66 Gy and for the larynx from 1.53 Gy to 0.78 Gy, with notable improvements also observed in the parotid glands and temporal lobes. Statistical analysis of clinical indices demonstrated non-inferiority to ground-truth plans (p > 0.05). Moreover, the predicted DVHs effectively guided plan revision and improved plan quality. CONCLUSIONS:Our results suggest that direct DVH prediction using a CNN-GNN framework offers a robust and clinically interpretable solution for treatment planning and quality assurance.
BACKGROUND:Although spatially-fractionated radiation therapy (SFRT) has been receiving increasing clinical interest, its complete and optimized clinical transfer requires several technological and radiobiological issues to be addressed. PURPOSE:The aim of this work is to setup and dosimetrically characterize an experimental platform generating planar megavoltage photon minibeams to exploit quantitative radiobiological SFRT experiments. METHODS:Minibeam patterns were generated on a Varian TrueBeam™ linear accelerator equipped with a high definition (HD) MLC, which consists of 60 tungsten leaf pairs, whom the 32 inner ones project a thickness of 2.5 mm at the isocenter plane (i.e., 100 cm distance from the radiation source) and the 28 outers of 5 mm. Each pattern was obtained alternating opening one leaf after closing one, two, or three of the smallest MLC leaves. Measurements were performed with a PTW microdiamond detector in edge-on orientation in a water phantom over depths from 1 to 30 cm, varying the beam energy and the source to surface distance (SSD) in different sessions. A Python-based in-house automated fitting procedure was used to obtain from the measured profiles all the minibeam parameters: peak widths (PWs), distance between consecutive peaks (CTCs), peak to valley dose ratios (PVDRs) and depth doses. RESULTS:The best performances were obtained at SSD = 70 cm and 6 MV, where PWs and PVDRs respectively range from 1.2 to 2.4 mm and from 2 to 7. For all patterns the average dose varies in depth analogously to the reference field, PVDRs show an almost steady decrease except at shallower depths, while PWs and CTCs widen linearly with the depths. Moreover, their variation over different sessions was within the estimated uncertainty of the single measurement. CONCLUSIONS:We are claiming that a Varian TrueBeam™ equipped with an HD-MLC represents a practical, stable, reproducible and accessible platform that might be suitable for radiobiological studies on SFRT and for the investigation of the relative underlying mechanisms.
BACKGROUND:Accurate prostate segmentation in Magnetic Resonance Imaging (MRI) is crucial for clinical diagnosis and treatment. However, it remains a challenging task due to low soft-tissue contrast and structural ambiguity in the gland's appearance. While recent deep learning methods have focused on refining network structures, they often neglect the inherent anatomical consistency of the prostate. PURPOSE:To address this limitation, a novel segmentation framework integrating an Anatomy-Aware Fusion Attention (AAFA) module is proposed. METHODS:By leveraging anatomical templates, our approach establishes a multilevel feature cross-attention mechanism that enhances global contextual modeling of prostate regions. Additionally, we introduce a Phased Learning strategy that progressively trains the model to mitigate the adverse effects of invalid or noisy samples commonly found in clinical MRI data. RESULTS:Extensive ablation studies on the PROMISE12 dataset validate the contribution of each component to overall performance. Comparison experiments on both PROMISE12 and MSD Prostate datasets show that our method consistently outperforms existing approaches across key metrics, such as Dice similarity coefficient (DSC), Intersection over Union (IoU), Precision and 95% Hausdorff distance (HD95). CONCLUSIONS:These results confirm the robustness and strong generalization capability of the proposed framework in challenging clinical segmentation tasks.
BACKGROUND:Four-dimensional computed tomography (4D CT) allows dynamic assessment of wrist kinematics and offers a non-invasive alternative for evaluating ligament instability. However, insufficient temporal resolution and differences in acquisition or reconstruction protocols may introduce motion-related artifacts that affect quantitative analysis. As these artifacts can mimic pathological carpal motion, measurement error must be quantified to distinguish true pathology from methodological error. Therefore, systematic evaluation of motion quantification across different CT systems and acquisition protocols is required. PURPOSE:To evaluate the errors in motion quantification across acquisition and reconstruction protocols of different CT systems in 4D CT imaging of wrist bones. METHODS:A rotating wrist phantom with three 3D-printed bones (scaphoid, lunate, capitate) was scanned on five CT systems from three manufacturers (Aquilion ONE PRISM and ONE Vision, Canon Medical; Revolution Apex, GE HealthCare; single-source and dual-source SOMATOM Force, Siemens Healthcare). One static 3D scan and multiple 4D scans were acquired at different phantom rotation speeds, each lasting 10 s. Single-source covered 0.050-0.300 phantom rotations per second (rps), and dual-source 0.100-0.600 rps. Dose dependence was evaluated on two systems (80 kV/40 mA and 120 kV/100 mA). Images were reconstructed in full and partial modes, segmented, and registered using point-to-image registration. Motion quantification error was calculated for translation and rotation of the scaphoid and capitate relative to the lunate, referenced to the static scan, and reported per rotation cycle. Motion quantification error was additionally expressed as a function of normalized motion per reconstructed frame to relate the error to the amount of motion occurring during image acquisition. Effects of CT system, bone type, reconstruction method, and rotation speed were analyzed using a linear mixed model. Statistical significance was defined as p < 0.05. RESULTS:The motion quantification error was hardly affected by dose. For 0.100 rps, the errors of the best single-source system were median 0.29 mm (interquartile range: 0.21-0.36 mm) and 1.37° (0.82°-2.18°) for full and 0.21 mm (0.16-0.25 mm) and 0.70° (0.52°-0.95°) for partial reconstructions for the capitate; the dual-source system showed the lowest errors (0.14 mm (0.12-0.17 mm) and 0.48° (0.39°-0.62°)). Motion quantification error scaled approximately linearly with normalized motion per reconstructed frame, corresponding to approximately 15% of the motion occurring during one reconstructed frame. CONCLUSIONS:Motion quantification errors were largely determined by phantom rotation speed and temporal resolution. The reported values provide a basis for defining the detection threshold for quantitative wrist kinematics, supporting differentiation between true kinematics and apparent displacement caused by motion artifacts.
BACKGROUND:Gliomas are the most common primary tumors of the central nervous system. Their treatment remains highly challenging, with high rates of associated disability and mortality. Conventional prognostic indicators no longer adequately satisfy the clinical demands of precision medicine. Therefore, it is essential to further explore novel prognostic biomarkers to enable accurate risk stratification and to provide new reference indicators for personalized precision therapy. PURPOSES:This study aimed to investigate the prognostic significance of vascular endothelial growth factor A (VEGFA) in patients diag nosed with lower-grade gliomas (LGGs) using an MRI based radiomics model. METHODS:Data regarding VEGFA expression and clinical records of LGG patients were retrieved from The Cancer Genome Atlas (TCGA). Corresponding preoperative MRI data were obtained from The Cancer Imaging Archive (TCIA) for radiomic feature extraction. Patients were stratified into high- and low- VEGFA expression groups based on survival information from the current cohort using the survminer package. The overall survival (OS) was assessed using Kaplan-Meier analysis and Cox proportional hazards regression. Predictive models were developed using logistic regression (LR), and model performance was evaluated via receiver operating characteristic (ROC) curve analysis, with area under the curve (AUC) values reported. An optimized model incorporating the Akaike information criterion (AIC) was also constructed (AIC-LR). RESULTS:VEGFA expression was significantly associated with OS (P = 0.002). Multivariate Cox regression confirmed VEGFA as an independent prognostic factor (hazard ratio [HR] = 2.545, 95% confidence interval: 1.422-4.555). Furthermore, VEGFA expression correlated with immune infiltration levels, particularly of M1 and M2 macrophages and T follicular helper cells, and was associated with enrichment in Wnt signaling and B cell receptor signaling pathways. The LR and AIC-LR models demonstrated acceptable predictive performance, with AUCs of 0.728 (95% CI: 0.612-0.843) and 0.725(95% CI: 0.612-0.839) in the training cohort, and 0.704 (95% CI: 0.562-0.847) and 0.718(95% CI: 0.576-0.861) in the validation cohort, respectively. CONCLUSIONS:The MRI based radiomics model showed potential for noninvasive assessment of VEGFA expression and may provide auxiliary information for prognostic evaluation in LGG. Further validation in larger samples and independent external cohorts is required before clinical application.
BACKGROUND:X-ray pulsatility index (XPI) quantifies cardiac-synchronous attenuation changes as a surrogate for regional lung perfusion, but current implementations are limited by projectional overlap and the lack of reference values in healthy individuals. PURPOSE:To characterize regional XPI patterns and establish preliminary reference values in healthy participants using a novel multi-angle fluoroscopic imaging system. METHODS:Nineteen healthy participants underwent an 8 s breath-hold during fluoroscopic imaging using 15 fps at four simultaneous projections (LPO, AP caudal, RPO, AP cranial) using a novel scanner. Frames were cropped to exclude initial x-ray tube stabilization and bulk motion. Spectral analysis exploited the periodic signal attenuation in the lungs to create XPI maps, as previously described. To focus on the clinically important regions of lung perfusion, the peripheral 3 cm of lung was manually segmented and divided into upper, middle, and lower lung zones. XPI values from these regions were compared to evaluate laterality and cranial-caudal differences using paired t-tests corrected with the Benjamini-Hochberg procedure to control the False Discovery Rate (FDR). Peripheral XPI contrast-to-noise (pCNR) was calculated in the AP caudal projection to quantify signal-to-background separation. Data were retrospectively resampled to assess the effect of reduced frame rate (7.5, 5 fps) and shorter acquisition time using multiple regression analysis. Radiation dose was estimated using phantom measurements and simulation. RESULTS:All participants performed the breath-hold without difficulty. XPI maps demonstrated bilateral lung perfusion across all four views, enabling multi-projection assessment of regional perfusion. No focal defects were observed; however, one participant demonstrated globally reduced XPI. Significant cranial-caudal gradients in XPI were observed across projection angles, consistent with known gravity-dependent physiology. Following FDR adjustment (Q = 0.05), the lower zones continued to demonstrate significantly higher values than the upper zones in 3 of 4 projections (all q < 0.05). There were no significant laterality differences (q > 0.05). XPI values remained stable across acquisition lengths and frame rates, indicating robustness of the metric. CNR increased with longer acquisitions and higher frame rates, reflecting reduced noise and improved signal reliability. CONCLUSION:Multi-angle XPI fluoroscopy enables non-invasive regional lung perfusion assessment with low radiation exposure and provides preliminary reference values for future clinical investigation.
BACKGROUND:Dedicated breast CT is an emerging breast X-ray imaging modality. While current commercial breast CT systems use prone-patient, pendant-breast geometry, the system described here uses upright patient geometry with the uncompressed breast supported by a cup. PURPOSE:The purpose of this work is to describe the development of a newly designed, upright geometry, dedicated cone-beam breast CT system and to evaluate its imaging performance using objective metrics. METHODS:The prototype system uses a tungsten-target, mammography-format, X-ray tube operating at 60 kV with 0.25 mm Cu and 1 mm Al added filtration, and a complementary metal-oxide semiconductor (CMOS) detector with 0.152 mm pixel pitch coupled to 500 microns thick CsI:Tl scintillator. During short scan acquisition, the X-ray source moves inferior to the breast, and 210 projections are acquired over an angular range of 210 degrees. The projections are reconstructed to an isotropic voxel pitch of 0.22 mm using Feldkamp-Davis-Kress (FDK) algorithm with Parker weights. Quantitative performance measures including linearity, modulation transfer function (MTF), and noise power spectrum (NPS) were evaluated. Phantom studies were conducted at various X-ray tube current (mA) and pulse-width (ms) combinations with the objective of determining the minimum detectable size of low-contrast targets and calcium carbonate spheres representing soft tissue lesions and microcalcification clusters, respectively. RESULTS:The measured 1st HVL was 4.23 ± 0.01 mm of Al. The limiting resolution (10% MTF) was 2.18 mm-1 in the coronal (cross-sectional) plane near the axis of rotation. In the coronal plane, the peak of the NPS occurred at 0.5 mm-1. Phantom studies at a mean glandular dose of 3-5.7 mGy showed the ability to visualize 2-3 mm low-contrast targets and 0.27-0.29 mm calcium carbonate spheres. CONCLUSIONS:The developed upright breast CT system showed the ability to achieve high spatial resolution and low contrast resolution. The excellent technical performance of the breast CT system reported here suggests that further investigations using patient imaging are warranted.
BACKGROUND:Technological developments in computed tomography (CT) have increased the diversity of acquisition and reconstruction strategies available for clinical imaging. Task-based image quality metrics that account for patient-size variability are needed to characterize reconstructed image performance and support protocol optimization across clinically relevant imaging conditions. PURPOSE:To propose and evaluate a novel size-specific weighted detectability index (SSW-d'), a task-based image performance metric (d') that incorporates lesion contrast weighting and water-equivalent diameter variability for the characterization of CT images acquired at low dose. METHODS:First, a multi-sized image quality phantom was scanned on two CT systems: one photon-counting CT (PCCT, CT1) and one CT equipped with energy integrating detectors (EICT, CT2). Acquisitions were performed with a volume CT dose index (CTDIvol) of 3.2 mGy at 120kV without a tin filter (Sn), at Sn100 and Sn140kV. Second, acquisitions at 3.2 mGy were performed on CT1 using three reconstruction algorithms (FBP, QIR-2, QIR-4). Subsequently, five reconstruction kernels (Br32, Br44, Br56, Br68, and Br76) were evaluated at 1.1 mGy. A higher CTDIvol value of 6 mGy, combined with Br68 and QIR-4, was also tested. The detectability index values (d') were determined for two 10 mm diameter lesions according to two different levels of contrast, based on iodine and bone equivalent tissue. The SSW-d' was calculated for each CT system and for each individual task. Then, SSW-d' was expanded to include both tasks, applying contrast weights determined for each of them. RESULTS:On CT1, the highest SSW-d' values were 19.3 ± 0.3 and 6.9 ± 0.0, at 120 kV for bone and iodine rods, respectively. On CT2, the highest SSW-d' results were 19.0 ± 0.3 and 4.9 ± 0.0 at Sn100 kV for bone and iodine rods, respectively. The lowest values were at Sn140kV for both rods and CT systems. When encompassing both tasks, the maximum SSW-d' values were 10.05 ± 0.04 and 8.25 ± 0.03, respectively obtained at 120 kV for CT1 and at Sn100 kV for CT2. The results showed that the lowest SSW-d' values were at Sn140 kV for both CT systems. Comparison of acquisition and reconstruction parameters showed that the highest SSW-d' value was obtained using the Br32 kernel (17.6 ± 0.7), approximately twofold higher than that obtained at 6 mGy (8.6 ± 0.3). The lowest SSW-d' value (1.9 ± 0.1) was obtained with FBP, while the highest performance was achieved with QIR-4 reconstruction. CONCLUSIONS:A novel metric based on a size-specific weighted detectability index (SSW-d') was developed and evaluated for the characterization of reconstructed CT image performance across a range of patient-equivalent diameters and clinically relevant detection tasks. The proposed SSW-d' consolidates multiple detectability index measurements into a single descriptor that accounts for lesion contrast weighting and patient-size variability. The metric was sensitive to acquisition and reconstruction parameters and proved useful for identifying imaging conditions that maximize task-based performance in ultra-low-dose CT. These findings suggest that SSW-d' may serve as a practical tool for image-quality characterization and protocol optimization across clinically relevant imaging scenarios.
BACKGROUND:Consistently generating clinically acceptable plans without human intervention remains a challenge in radiotherapy. Rule-based automation provides deterministic execution, and knowledge-based planning (KBP) provides statistical dose estimation, but both often require manual refinement. Large language models (LLMs) offer clinical reasoning capability, but effective autonomous planning also requires a mechanism to execute complex planning actions within the treatment planning system (TPS). PURPOSE:To develop and evaluate PlanningCopilot, an agentic system that utilizes the reasoning capability of LLM and a validated Eclipse Scripting API (ESAPI) optimization module integrating KBP initialization ("PlanAct") to autonomously generate treatment plans. This study evaluates the system's ability to produce clinically acceptable plans for locally advanced non-small cell lung cancer (LA-NSCLC) and assesses its potential to refine performance by self-learning. METHODS:PlanningCopilot was implemented as a multi-agent framework linked to the TPS through PlanAct API. It comprises four specialized GPT-4.1 agents that iteratively interact with the TPS: (1) an Evaluator agent that accesses the plan and generates plan quality reports, (2) a Supervisor agent that validates these reports before passing them to a Planner agent, (3) the Planner agent that executes initialization and optimization tasks through PlanAct API and planning guidelines, and (4) an optional Learner agent that synthesizes optimization history into Planner-facing prompt addendums. We retrospectively analyzed 62 patients with conventionally fractionated LA-NSCLC and compared original clinical plans with autonomous plans with and without the Learner agent. Measurement-based patient-specific quality assurance (PSQA) was performed on the first 21 autonomous IMRT plans in planning order. RESULTS:All autonomous plans met clinical dosimetric requirements, including those not achieved in the clinical plans and KBP (RapidPlan) plans. Paired Wilcoxon signed-rank tests showed no significant differences between autonomous and clinical plans for Lungs Dmean (p = 0.371), Lungs V20Gy (p = 0.449), Lungs V5Gy (p = 0.309), Heart D50% (p = 0.175), Esophagus Dmean (p = 0.750), Spinal Cord D0.03cc (p = 0.422), and Plan D0.03cc (p = 0.941). Furthermore, autonomous plans achieved significantly lower Esophagus D0.03cc (p = 0.027). Compared with RapidPlan initialization, PlanningCopilot improved multiple dosimetric endpoints, including Lungs Dmean (p < 0.001), Lungs V20Gy (p < 0.001), Lungs V5Gy (p = 0.004), Heart D50% (p = 0.037), and Plan D0.03cc (p < 0.001), with the cost of higher Esophagus Dmean (p < 0.001) and Esophagus D0.03cc (p < 0.001). In a subset of 18 cases requiring at least two iterations, applying Learner-derived knowledge reduced required iterations by an average of 11.8% while maintaining comparable plan quality (p > 0.05). All 21 autonomous IMRT plans passed measurement-based PSQA. CONCLUSION:PlanningCopilot enables autonomous generation of clinically acceptable and deliverable treatment plans for LA-NSCLC. It consistently satisfies clinical dosimetric requirements across varying anatomical complexities and improves optimization efficiency through self-learning from prior optimization history.
BACKGROUND:Coincidence timing resolution (CTR) is a key performance factor in PET systems. Cherenkov radiation, due to its ultrafast emission on the femtosecond scale, offers potential for achieving improved timing resolution compared to traditional scintillation-based detectors. However, the inherently low photon yield in pure Cherenkov radiators makes depth-of-interaction (DOI) estimation and its associated timing uncertainty a significant challenge, especially when trying to achieve sub-30 ps CTR. PURPOSE:This study aims to reduce DOI-induced timing uncertainty in Cherenkov-based PET detectors by proposing a novel wedge-shaped light guide design. The goal is to improve the CTR performance by compensating for optical path differences resulting from varying interaction depths. METHODS:A Monte Carlo simulation framework was developed to model gamma-ray interactions within a Cherenkov radiator, incorporating electron scattering and Cherenkov photon generation. Two detector geometries were compared: one with a conventional planar light guide and another with the proposed wedge-shaped light guide. The wedge design was tailored to delay early-arriving photons and synchronize photon transit times, thereby reducing time spread. The simulation tracked photon paths through the wedge and evaluated performance using metrics such as spatial resolution, timing spread, and sensitivity. RESULTS:Simulations demonstrated that the wedge-shaped light guide significantly reduced DOI-induced timing uncertainty-by 26.3% in a 3 mm-thick Cherenkov radiator-compared to the conventional configuration. The wedge geometry also enabled improved spatial resolution in multi-layer detector configurations. However, a trade-off with reduced sensitivity was observed. To address this, strategies such as utilizing discarded photons via refractive index manipulation and cascade detector structures were proposed. CONCLUSIONS:The wedge-shaped light guide effectively compensates for DOI-related path length differences and enhances CTR in Cherenkov-based PET detectors. This approach opens avenues for more accurate event localization without increasing photon yield, and could serve as a building block for advanced PET systems requiring ultrafast timing performance.
BACKGROUND:Knowledge-based planning (KBP) has improved the quality and efficiency of radiotherapy treatment planning. However, its broader clinical adoption remains limited because effective deployment often requires institution-specific model training and tuning. Publicly available KBP models provide a convenient starting point but may not consistently meet local clinical objectives across institutions. PURPOSE:We developed and evaluated the Planning Copilot, a large language model (LLM)-guided plan refinement framework designed to operate as a model-agnostic post-processing layer for KBP. METHODS:The Planning Copilot is a closed-loop, multi-agent system that iteratively refines KBP-generated plans through structured dosimetric feedback and the selection of clinically validated optimization actions within a treatment planning system. For each case, an initial step-and-shoot IMRT plan was generated with each of three RapidPlan models, including a publicly available model and two institutional models with different optimization constraints. To assess whether the refinement depends on KBP, we additionally evaluated PlanningCopilot starting from a non-KBP fixed objective template applied identically to all cases. The PlanningCopilot was applied without model-specific tuning to 62 retrospective locally advanced NSCLC cases. Clinical goal achievement rates and clinically relevant dose-volume metrics were compared between the initial KBP plans and the refined plans. RESULTS:Across all three KBP models, the PlanningCopilot substantially improved plan quality. Clinical goal achievement increased from 79% to 98% for the UCSD model, from 73% to 97% for Institutional T1, and from 69% to 98% for Institutional T2. Starting from the non-KBP fixed template, the achievement rate increased from 68% to 97%, comparable to the KBP initializations. Significant reductions were observed in key lung dose metrics, including lung Dmean across all models and lung V20 in the Institutional T1 model and template initializations, while target coverage and doses to critical structures were maintained. Notably, the KBP model that prioritized OAR sparing, which exhibited the lowest initial pass rate, showed the highest rescue rate after refinement. CONCLUSIONS:An LLM-guided refinement layer can improve the success rate and portability of KBP across heterogeneous models without retraining the underlying KBP system. This approach provides a practical strategy to enhance the reliability of KBP and supports the use of off-the-shelf models through automated, model-agnostic post-processing.
BACKGROUND:Spot scanning proton arc therapy (SPArc) combines the dosimetric advantages of proton therapy with the beam-angle freedom of arc delivery. However, current planning algorithms rely on static delivery assumptions that do not account for the temporal characteristics of pulsed-beam synchrocyclotron systems during continuous gantry rotation. This mismatch between nominal plans and actual treatment delivery may lead to clinically meaningful dose deviations. PURPOSE:To develop and evaluate a dynamic arc delivery sequencing optimization framework that incorporates machine-specific delivery characteristics to minimize deviations between planned and delivered dose in SPArc. METHODS:A five-step dynamic arc delivery sequencing optimization framework was developed. The framework includes: (1) static and dynamic delivery time calculation, (2) spot and energy-layer disassembling, (3) incorporation of dynamic delivery timing into control points, (4) spot-weight fine-tuning, and (5) reconstruction of energy-layer sequences. Five multi-metastatic brain stereotactic radiosurgery cases were retrospectively evaluated. Delivery accuracy, efficiency and plan quality were assessed using virtual machine logfiles. RESULTS:The sequencing optimization framework substantially improved delivery accuracy while preserving plan quality and efficiency. For the total gross tumor volume, the mean absolute D98 deviation between planned and virtual logfile reconstructed doses decreased from 77.4 ± 81.0 cGyE (4.2 ± 4.5%) with static SPArc plans to 9.6 ± 4.0 cGyE (0.5 ± 0.2%) after sequencing optimization. For the worst metastasis in each case, D98 deviation decreased from 184.4 ± 145.2 cGyE (9.8 ± 7.9%) to 19.0 ± 16.4 cGyE (1.0 ± 0.8%), and D2 deviation decreased from 148.4 ± 114.4 cGyE (6.8 ± 5.6%) to 13.2 ± 8.6 cGyE (0.6 ± 0.4%). Target coverage and normal brain sparing remained statistically unchanged (p > 0.05), and total delivery times differed by < 1 s. CONCLUSIONS:The proposed sequencing optimization framework addresses the temporal mismatch between static SPArc planning and dynamic delivery in synchrocyclotron-based systems. By improving delivery accuracy without compromising plan quality or delivery efficiency, the framework demonstrates the feasibility of incorporating machine-specific delivery timing into dynamic proton arc therapy.
BACKGROUND:Pancreatic cancer Volumetric Modulated Arc Therapy (VMAT) planning presents a significant dosimetric challenge due to the high-dose gradients required to spare adjacent, radiosensitive organs at risk (OARs) like the stomach and duodenum. This anatomical complexity has limited the scope of automated planning for this site. Broadly, while deep learning (DL) has been introduced to streamline treatment planning, most existing models only predict intermediate outputs, such as dose distributions or fluence maps, which still necessitate a subsequent, computationally expensive inverse optimization step on a treatment planning system. PURPOSE:To address these challenges, we aim to develop an optimization-free fully automated VMAT planning framework for pancreatic cancer. As a key component of this system, this study introduces a DL model designed to directly generate machine parameters from dose distributions and anatomical contours. METHODS:A total of 200 vmat plans for pancreatic cancer (prescription: 42 Gy in 15 fractions) were retrospectively collected. The dataset was randomly split into training (n = 170), validation (n = 10), and testing (n = 20) sets. The proposed Multi-modal, Attention & Transformer-Enhanced U-Net (MATE-UNet) utilizes beam's-eye-view (BEV) projections of the reference dose distribution and anatomical contours to directly predict machine-executable multi-leaf collimator (MLC) apertures and Monitor Units (MUs). The proposed model was benchmarked against baseline U-Nets (using contour-only, dose-only, and combined inputs) on the testing set. Model accuracy was assessed using the Dice Similarity Coefficient (DSC) for MLC and Mean Absolute Error (MAE) for MU, while plan quality was evaluated using clinical dose-volume histogram (DVH) metrics and Conformity Index (CI), Homogeneity Index (HI), and Gradient Index (GI). RESULTS:MATE-UNet achieved a clinical acceptance rate of 100% (20/20), compared to 70% for the best-performing baseline model. In terms of prediction accuracy, the proposed model achieved a DSC of 0.9553 ± 0.0042 for MLC apertures and an MAE of 1.960 ± 0.396 for MU. Dosimetric evaluation demonstrated that, relative to the reference plans, MATE-UNet maintained comparable target coverage, CI (0.773), and HI (0.100), while achieving a significantly improved GI (3.732, p < 0.05). Furthermore, MATE-UNet significantly reduced the V39Gy of the stomach and duodenum, as well as the global maximum dose, compared with the baseline models (p < 0.05). CONCLUSIONS:This study demonstrates the feasibility of MATE-UNet for the direct prediction of VMAT machine parameters without iterative optimization. By leveraging multi-modal BEV inputs and a Transformer-enhanced architecture, the proposed framework represents a valuable step toward bridging the gap between dose distributions and clinically usable treatment plans. Although an additional normalization step is required to determine the absolute MUs, MATE-UNet has the potential to serve as a downstream component of a future fully automated anatomy-to-plan pipeline.
To establish a public repository of computed tomography (CT) texture phantom images paired with objective 3D image quality measurements from multiple scanner models using a diverse range of imaging protocols, facilitating investigations into relationships between image quality and quantitative imaging features. Three specialized CT phantoms were scanned: (1) the Corgi® phantom for image quality assessment, (2) a radiomics liver phantom, and (3) an open-source 3D-printed texture phantom. Image quality assessment included measurement of contrast-to-noise ratio, 3D modulation transfer function, and 3D noise power spectrum. Data were acquired on four CT scanner models from two manufacturers at five CTDI v o l levels (2.05-17.11 mGy) and reconstructed with eight different kernels, yielding 160 total conditions (combinations of scanner, dose, kernel). Data were validated for integrity and completeness, resulting in the exclusion of six conditions. The paired texture phantom and image quality (PTP-IQ) dataset includes: (1) DICOM image series of two texture phantoms acquired across the 154 conditions, as well as (2) image quality metrics derived from each corresponding set of scanner, acquisition and reconstruction settings provided in HDF5 format. This dataset enables the development and validation of harmonization methods for multi-center quantitative imaging studies, investigation of protocol-dependent QIF variability, and optimization of acquisition protocols for radiomics applications. The controlled and systematic study design facilitates isolation of individual protocol effects on quantitative measurements.
BACKGROUND:Superficial brachytherapy enables localized dose delivery for superficial cutaneous lesions, but conventional applicators may be limited in treating extended, irregular, or curved surfaces. Geometric constraints, gaps or overlaps between adjacent applicators, and placement-related uncertainties can reduce dose uniformity, underscoring the need to optimize applicator geometry and deployment strategy. PURPOSE:To enhance the efficacy of superficial brachytherapy for irregular or curved cutaneous surfaces, we conducted a dosimetric analysis of beta and photon sources utilizing modular circular and hexagonal applicator geometries. The study emphasizes dose uniformity, the impact of applicator dimensional variations, and the evaluation of alternative deployment strategies. METHODS:Monte Carlo simulations were performed for beta sources (P-32, Sr/Y-90, and Ho-166) and photon sources (I-125, Pd-103, and Yb-169). Applicator models incorporating a zirconia absorber with ceramic and tungsten shielding were constructed, and dose distributions were scored in a water-equivalent medium. Percent depth-dose (PDD) curves were generated for beta sources, and photon sources. Multi-applicator configurations were evaluated on planar and curved surfaces under both simultaneous and sequential delivery. Applicator-size effects were examined using two active radii (5.50 mm and 2.75 mm). Dose uniformity was quantified using the mean dose and the peak-to-valley ratio (PVR) at 1 mm depth. RESULTS:Beta sources were strongly affected by applicator size and geometry, producing larger variations in near-surface dose uniformity than photon sources. Sequential delivery reduced peak-to-valley modulation for all radioisotopes. Reducing the applicator radius increased the number of deliverable positions on curved surfaces by 60%-83%, increased mean dose by 28%-71%, and decreased PVR by 33%-86%, indicating improved uniformity. Photon sources showed more stable behavior, with lower sensitivity to applicator geometry and size. CONCLUSION:Radioisotope type, applicator radius, and delivery mode significantly influence superficial dose distributions. Smaller applicators and sequential deployment improve dose uniformity, particularly for beta sources. These results provide practical guidance for radioisotope selection and modular applicator design in superficial brachytherapy.
BACKGROUND:Boron Neutron Capture Therapy (BNCT) utilizes high linear energy transfer (LET) charged particles from the 10B(n,α)7Li reaction to selectively destroy tumor cells. Unlike conventional radiation therapy, BNCT facilities rely on fixed beam ports, making patient positioning critical in treatment planning. Current treatment planning, however, depends on manual forward planning using computationally intensive Monte Carlo (MC) simulations. PURPOSE:We propose an automated patient position optimization framework for BNCT to address these limitations. This method overcomes the computational bottleneck of MC simulations by integrating a rapid deep learning (DL) dose prediction model with a high-fidelity GPU-accelerated MC engine, enabling efficient inverse planning. METHODS:We developed a hybrid coarse-to-fine optimization workflow driven by the Trust Region Bayesian Optimization (TuRBO) algorithm. In the coarse stage, a 3D U-Net architecture rapidly identifies promising patient positions. In the fine stage, the solution is refined within a restricted parameter bound using the GPU-accelerated MC engine. We evaluated the framework on the GLIS-RT open dataset (229 glioblastoma patients, split into train, validation, and test sets at a 70:20:10 ratio). RESULTS:Optimizing the patient position with TuRBO improved target dose and homogeneity over a non-optimized baseline, reaching equivalent plan quality about 5.65× faster than an exhaustive grid search. In the coarse stage, the DL dose predictor ran about 37× faster than the high-fidelity MC. The coarse-to-fine workflow then matched high-fidelity MC plan quality while cutting optimization time by a factor of 2.7, outperforming optimization driven by the DL predictor alone. Two-port plans required no change to the method, where the added geometric freedom raised the mean target dose from 59.96% to 74.91% and lowered the homogeneity index (HI) from 1.226 to 0.647. CONCLUSIONS:We successfully developed a Bayesian optimization (BO) framework for BNCT patient positioning. The proposed coarse-to-fine approach effectively balances computational speed with accuracy, offering a practical way toward automated inverse planning in BNCT.
BACKGROUND:As MRI continues to advance toward higher field strengths, RF safety assessment has become more complex. Accurate estimation of specific absorption rate (SAR) is essential for evaluating both hardware and implant safety, yet thermometry and simulation-only approaches face limitations. In addition, these evaluations are typically performed within the MR scanner, where failures or excessive loading can damage the transmit chain and result in costly repairs and scanner downtime. To overcome these challenges, alternative experimental approaches that enable direct and controlled SAR validation outside the MR environment are needed. PURPOSE:This study demonstrates direct SAR validation of RF hardware and implants in a dedicated B0-free RF safety laboratory. The work aims to establish direct SAR measurement as a practical and accurate alternative for quantitative safety verification while expanding the range of tools available for MR hardware evaluation beyond the constraints of the scanner environment. METHODS:Hardware assessments were conducted in the safety lab equipped with a 16-channel broadband RF amplifier system (45-450 MHz), a 100 dB shielded Faraday cage, automated 3D and 1D motion systems, and a high-precision SAR probe. For coil validation, direct SAR mapping was performed on a custom-built parallel transmit (pTx) coil and a commercial single-channel RF coil at 447 and 297 MHz, respectively, and results were compared with electromagnetic simulations. Transfer function (TF) validation for a directional DBS electrode at 128 MHz was performed in a rectangular electric-field generator, whose field distribution was first validated using the SAR probe. The same probe was then used to compare measured SAR with TF-based SAR predictions, providing a direct quantitative validation of the TF method. RESULTS:Measured SAR distributions showed strong quantitative agreement with simulations across all experiments. For the 8-channel pTx head coil at 447 MHz, the normalized root-mean-square error (NRMSE) across four excitation patterns was below 13%. For the commercial single-channel head coil at 297 MHz, plane-wise comparisons across axial, coronal, and sagittal slices yielded an average NRMSE value of 7.18%, demonstrating spatial consistency between measured and simulated fields. The electric-field generator used for DBS transfer function (TF) validation exhibited 3.68% NRMSE relative to simulation, confirming accurate field reproduction. SAR measurements using the same probe further showed close agreement with TF-based SAR predictions across all seven lead trajectories, confirming both the accuracy and broadband applicability of the direct SAR method. CONCLUSION:Direct SAR measurement provides a reliable and quantitative approach for validating RF coil and implant safety, matching the accuracy of conventional scanner-based thermometry and simulation methods. Conducting these measurements in a B0-free RF safety laboratory further simplifies and accelerates the process by eliminating magnetic field constraints and the waiting periods required for the setup to return to thermal equilibrium in thermometric techniques. This configuration provides an efficient and controlled platform for systematic SAR validation across a wide range of MR frequencies.
BACKGROUND:Respiratory motion is known to cause blurring in Single-Photon Emission Computed Tomography (SPECT) images which can mask or mimic disease. Pediatric imaging with 99mTc-labeled dimercaptosuccinic acid (DMSA) is used to assess cortical defects in kidneys and may be especially susceptible to artifacts introduced by respiratory motion due to the thin kidney cortices and small kidney volumes seen in many pediatric patients. PURPOSE:The purpose of this study was to assess a data-driven method to estimate respiratory motion signals in pediatric 99mTc-DMSA renal SPECT and to evaluate the impact of respiratory motion correction on image quality. METHODS:Listmode 99mTc-DMSA SPECT data were acquired for 77 pediatric patients aged from 6 weeks to 20 years. The data were binned into 100 ms temporal frames and forward-projected kidney masks were used in determining the axial center-of-mass (aCOM) of counts in each temporal frame as a surrogate renal respiratory motion signal for each patient. Amplitudes of respiratory motion in the lateral, anterior-posterior and axial (superior-inferior) axes were determined using a rigid-body six-degree-of-freedom intensity-based registration method and evaluated as a function of patient weight. Welch t-tests were performed to compare the respiratory motion amplitudes of male and female patients. Subsequently, a rigid-body 6-degree-of-freedom respiratory motion correction was applied during reconstruction and the images were quantitatively assessed for improvement in contrast and sharpness. RESULTS:Renal respiratory motion surrogate signals were estimated for 69 patients, after eight were removed due to gross body motion. Axial, lateral and anterior-posterior translational renal respiratory motion amplitudes were all found to positively correlate with patient weight, with lateral motion showing the strongest correlation. Respiratory motion was largest in the axial direction where it ranged from (2.19 ± 0.48) mm for the patients under 7 kg to (6.96 ± 3.07) mm for patients over 60 kg. No significant differences were found in axial or anterior-posterior renal respiratory motion between male and female patients, but lateral motion was slightly lower for females. Quantitative assessment of reconstructed images showed that respiratory motion correction improved contrast and sharpness as a function of estimated amplitude of axial respiratory motion, suggesting a threshold where motion correction becomes beneficial. CONCLUSIONS:A data-driven respiratory motion estimation method found significant positive correlation in renal motion amplitudes and weight for pediatric DMSA SPECT studies. The derived respiratory signals were used to perform rigid-body respiratory motion correction during reconstruction. The renal respiratory motion correction reduced cortical blurring and improved contrast in some patients, generally those with the largest estimated motion, highlighting potential for this method to improve diagnostic accuracy.
BACKGROUND:The degree of freedom has been recognized as a key factor in particle beam therapy, such as improving the dose conformity and linear energy transfer distributions. Therefore, proton arc therapy has drawn significant interest in the Society of Radiation Oncology as a potentially efficient and optimal treatment option for cancer patients. However, state-of-the-art spot-scanning proton arc therapy (SPArc) still relies on single or dual- co-planar arc trajectories, which limits its capacity to advance the treatment outcome further. There is an urgent need to explore particle beam therapy's full potential via 4pi. PURPOSE:This study aims to develop the first 4pi arc optimization algorithm that searches for efficient arc trajectories in 4pi space, explores the potential dosimetric improvements, and demonstrates its feasibility through simulation. METHOD:Dynamic Programming, originally from control theory, was translated into this new concept of the 4pi SPArc optimization algorithm (SPArc-4pi) for the trajectory search and route decision-making. It breaks down the complicated and high-computational-demand main problem into a series of small sub-problems and searches for delivery-efficient 4pi arc trajectories through an iterative approach. Five different disease sites, e.g., head & neck, partial brain, clival brain chordoma, lung and pancreatic cancer, were used for testing purposes. Conventional Intensity Modulated Proton Therapy (IMPT) and 2D co-planar arc (SPArc-2d) plans were generated as a benchmark. Treatment delivery efficiency was evaluated through a published and validated dynamic arc system controller. Plan quality was assessed through target coverage and organ at risk (OAR) sparing. RESULT:The new SPArc-4pi demonstrated superior dosimetric performance across various evaluation metrics for all five disease sites. The simulation result shows that SPArc-4pi is able to be delivered within a reasonable time of around 5-11 mins, which is comparable to the SPArc-2d plans CONCLUSION: The study introduced the first optimization algorithm for SPArc-4pi technique. It not only showed SPArc-4pi 's potential to improve the plan quality via a greater degree of freedom compared to the conventional IMPT and state-of-art 2D co-planar SPArc-2d technique but also demonstrated its feasibility for future clinical implementation based on the current proton beam therapy system's machine characteristics.
BACKGROUND:Accurate attenuation correction (AC) is essential for quantitative positron emission tomography (PET). Conventional CT-based AC provides reliable attenuation (μ-) maps but adds radiation, introduces PET/CT misalignment artifacts, and is unavailable on stand-alone PET systems. Existing deep learning (DL) methods use non-attenuation corrected (NAC) PET data to predict CT-like or directly AC-PET images, but their dependence on emission characteristics limits generalizability across tracers and anatomical coverage, especially for long axial field-of-view PET. PURPOSE:We propose incorporating tissue-density information from dual-view scout radiographs as supplementary input to NAC PET, acquired with a clinically significant radiation reduction relative to standard low-dose CT. METHODS:Two DL methods were evaluated for generating transmission (Tr) images as surrogate μ-maps for PET AC: (1) (NAC)-to-(Tr), using coronal and sagittal NAC PET slices, and (2) (NAC + Scout)-to-(Tr), combining NAC PET with anterior-posterior and lateral scout views. 53 research scans across four tracers from the PennPET Explorer were used for training and testing, with three additional tracers included for testing. RESULTS:(NAC + Scout)-to-(Tr) improved quantitative accuracy, reducing NRMSE% to < 10% (versus up to 18% for (NAC)-to-(Tr)) and SUV biases to within -10%, with significant reductions in brain, liver, and muscle compared to (NAC)-to-(Tr). Out-of-distribution evaluation confirmed generalizability, maintaining SUV biases below 5%-8% versus 10%-13% for (NAC)-to-(Tr). In a longitudinal biodistribution study, the scout-guided model showed consistent performance across repeat scans. CONCLUSIONS:By leveraging routinely acquired scout scans, this approach enables accurate quantitative PET reconstruction without a full CT, substantially reducing radiation dose and misalignment artifacts.
