Background and purpose:Radiotherapy for pancreatic cancer requires effective respiratory motion management and precise image guidance due to the proximity of multiple dose-limiting organs. Respiratory gating is commonly used for patients who cannot tolerate breath hold, with gated cone-beam computed tomography (gCBCT) routinely employed to verify gating thresholds and patient positioning. However, the gCBCT scans are lengthy (2-8 min). This study investigates the feasibility of a novel nonstop gated CBCT (ngCBCT) technique as a replacement for current clinical gCBCT, reducing scan time to ∼1 min and decreasing imaging dose by >40%. Materials and methods:Clinical gCBCT scans from 15 pancreatic cancer patients (58 half-fan and 98 full-fan scans) were retrospectively collected. The ngCBCT projections were emulated by alternate-cycle down-sampling gCBCT projection data. A dual-domain convolutional neural network (DDCNN) was optimized for pancreatic cancer imaging and evaluated for ngCBCT reconstruction. Results:DDCNN-based ngCBCT reconstructions achieved image quality comparable to gCBCT, preserving soft-tissue contrast and anatomical details. The projection-domain CNN played a vital role in yielding high image quality, while the image-domain CNN provided additional modest improvements to the reconstruction. Incorporating a 15% dropout rate to DDCNN further enhanced model robustness and generalizability. Conclusions:This work demonstrates the first successful application of ngCBCT to pancreatic cancer imaging. By achieving high-quality reconstructions with a one-minute scan, ngCBCT has the potential to transform respiratory gating pancreas radiotherapy, offering improved patient comfort and enhanced treatment efficiency. The approach paves the way for extending ngCBCT to other abdominal sites impacted by respiratory motion.

PURPOSE:This study aimed to commission the AlignRT inBore system in Ethos/Halcyon linear accelerator (Linac) systems for surface-guided radiotherapy (SGRT), including developing a soft collision quality assurance (QA) to pass daily AlignRT QA without daily calibration, ensuring compatibility between SGRT and image-guided radiotherapy (IGRT) setups, establishing a baseline performance in AlignRT motion monitoring, and enabling breast DIBH SGRT using manual beam gating with timing accuracy assessment. METHODS:The AlignRT inBore system contains three ceiling-mounted camera pods for SGRT setup at the simulation isocenter (sim-ISO), and two ring-mounted camera sets on the spring-supported bore cover for motion monitoring at treatment ISO (txt-ISO), after iterative cone-beam computed tomography (iCBCT) setup. The commissioning includes confirming the ring cameras are outside the radiation fields, verifying IEC coordinate consistency between SGRT and IGRT, and ensuring the compatibility of AlignRT daily QA with collision QA. A dot-array plate was used for SGRT calibration and daily QA at both ISO positions. A soft collision check method was developed to minimize the bore-cover displacement by (1) triggering the collision alarm from the rear side, distant from the front ring cameras, and (2) reproducing the rear bore-cover position by marking before the collision check. Clinical staff were trained for the new daily QA procedure, and the average period between calibrations was calculated over the past 12 months. Additionally, the application-specific breast DIBH SGRT procedure was commissioned through an end-to-end test with manual gating, which was retrospectively evaluated for timing accuracy. RESULTS:The AlignRT inBore camera ring was moved 2 cm away from the iCBCT field toward the bore front, and the IEC 61217 coordinate system yields consistent SGRT and IGRT setups. AlignRT inBore daily QA can pass using the soft collision procedure without daily calibration, with an average calibration period of 2 weeks, as ad-hoc calibrations are required after open-bore services. The manual gating accuracy is found to be 0.6s ± 0.2s, which can be further reduced if the beam can be proactively turned off before coaching patients to relax from a DIBH. CONCLUSION:The commissioning of the AlignRT inBore systems was successfully implemented, including developing and implementing the soft collision QA without daily calibration for 2 weeks, updating inBore camera position (2 cm to the bore front), ensuring SGRT-IGRT couch-shift consistency, establishing a baseline performance (< 0.2 mm drift), and allowing SGRT applications on Ethos/Halcyon systems. The application-specific breast DIBH SGRT procedure with manual beam gating is also commissioned with a new retrospective timing analysis for assessing gating accuracy (0.6s ± 0.2s).

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.

Introduction.This study aims to generate synthetic-magnetic resonance imaging (synMRI) for abdominal patients undergoing radiotherapy using deep inspiration breath hold (DIBH) on a conventional Linac.Methods.Extended cardiac-torso (XCAT) phantom and patient data were used to develop and test a patient-specific model to generatesynMRI. During patient simulation, 6 DIBH-MRIs and 1 DIBH-CT were acquired. One DIBH-MRI was set as the reference MRI (refMRI) and principal component analysis (PCA) was performed using five deformation field maps (DFMs) by applying deformable image registration betweenrefMRI and the other DIBH-MRIs. PCA eigenvalues were sampled 1000 times to generate new deformations applied to a synthetic-CT with the same anatomic conditions asrefMRI. A convolutional neural network was trained to predict the eigenvalues corresponding to on-board conditions from a cone-beam CT (CBCT) to generate a finalsynMRI with the new DFM. Four XCAT scenarios simulated changes from simulation to treatment. The model was evaluated using mean-absolute-error (MAE) and root-mean-square-error (RMSE), and the image quality was evaluated by structure-similarity-index metric (SSIM) and normalized RMSE (nRMSE). The accuracy of the predicted target volume for XCAT and fiducial clips for each patient was analyzed using center-of-mass-shift (COMS) between on-board conditions (ground-truthsynMRI for XCAT and CBCT for patients) and the predictedsynMRI. The liver dome difference was also evaluated.Results.Model performance yielded an MAE of 0.11 ± 0.02 for XCAT, and 0.15 ± 0.01, 0.15 ± 0.02, 0.10 ± 0.03, 0.12 ± 0.09 for four patients, respectively. RMSE values were 0.11 ± 0.07, 0.16 ± 0.03, 0.19 ± 0.01, 0.13 ± 0.07, and 0.14 ± 0.04. For image quality, SSIM values were 0.999 ± 0.001 for XCAT, and 0.998 ± 0.001, 0.995 ± 0.002, 0.996 ± 0.001, 0.998 ± 0.001 for the patients. The nRMSE were 0.10 ± 0.06, 0.58 ± 0.03, 0.87 ± 0.06, 0.39 ± 0.04, and 0.35 ± 0.02, respectively. For XCAT, the liver dome difference and tumor COMS were <0.5 mm in all scenarios. For all patients, liver dome differences were <1 mm, and fiducial COMSs were <1.2 mm.Conclusions. The novel method generatessynMRI using on-board conditions in DIBH abdominal treatments performed on a conventional LINAC.

BACKGROUND:The nonstop gated CBCT (ngCBCT) technique has been proposed as a next-generation replacement to the current inefficient clinical gated CBCT (gCBCT), reducing half-fan scan time from 2-8 min to 1 min on C-arm linear accelerators while substantially lowering imaging dose. However, ngCBCT yields highly non-uniform and under-sampled projections, posing a major challenge for conventional reconstruction methods to achieve high-quality images. PURPOSE:To develop a powerful and efficient Dual-Domain Convolutional Neural Network (DDCNN) tailored to the unique characteristics of ngCBCT projections to achieve high-quality ngCBCT imaging. METHODS:Clinical raw gCBCT projections from 31 free-breathing respiratory-gated lung SBRT patients (77 half-fan and 65 full-fan scans) were retrospectively retrieved and down-sampled based on respiratory signals to emulate ngCBCT acquisitions. The proposed DDCNN integrates a projection-domain network, which completes missing projection data, with an image-domain network that further reduces artifacts. The two domains are linked by the fast FDK algorithm, enabling accurate and efficient reconstruction suitable for real-time clinical use. RESULTS:DDCNN outperformed conventional reconstruction methods and image-domain-only CNN approaches both qualitatively and quantitatively. It achieved high image quality and rapid reconstruction (<1 min) for ngCBCT acquisitions, supporting clinical adoption in pretreatment setup for respiratory gating lung SBRT. CONCLUSION:Pairing the innovative ngCBCT acquisition strategy with the DDCNN framework enables substantial reductions in scan time and imaging dose while maintaining high image quality. This advancement has the potential to improve patients' treatment experience and overall efficiency of respiratory gating lung SBRT, and also paves the way for broader adoption of respiratory gating techniques in other motion-affected tumor sites.

Background and purpose:Significant anatomical changes during head and neck cancer (HNC) radiotherapy challenge accurate dose delivery. Deformable image registration (DIR) is essential for adaptive radiotherapy (ART), yet conventional methods are too slow for online clinical use. This study proposed a novel deep learning-based DIR algorithm for longitudinal HNC imaging. Materials & methods:We used sixty HNC patient datasets, each containing a planning CT (pCT) and six weekly cone-beam CTs (CBCTs). Fifty datasets were used for training with cross-validation, and the remaining ten were reserved for testing. The proposed DIR algorithm is a patch-based model that integrates 3D convolutional neural networks, self-attention, and a convolutional Long Short Term Memory to model temporal deformations. The model predicted bidirectional deformation vector fields and was trained with a composite loss function combining image similarity, DVF smoothness, and inverse consistency. Performance was benchmarked against the large deformation diffeomorphic metric mapping (LDDMM) algorithm using Dice similarity coefficient (DSC), Hausdorff distance, and Jacobian analysis. Results:The proposed method achieved significantly faster inference, performing bidirectional DIR between the pCT and all six weekly CBCTs in under 3 min and averaging about 30 s per patient, while matching or exceeding LDDMM's accuracy. DSC remained above 0.8 for all key structures, and the method demonstrated improved DVF consistency with lower mean and 95th percentile Hausdorff distances. Unlike LDDMM, it required no manual parameter tuning, providing consistent results. Conclusion:The proposed DIR algorithm enabled rapid, accurate, and consistent image registration, supporting real-time ART workflows and retrospective dose accumulation in personalized radiotherapy.

Purpose:To investigate the relationship between geometric and dosimetric characteristics of salivary glands and acute xerostomia during or shortly after radiation therapy in patients with head and neck cancer (HNC). Methods and Materials:We used an Automated Watchdog in Adaptive Radiotherapy Environment (AWARE) program to monitor changes in tumor and organ-at-risk using weekly cone beam computed tomography (CT) images for patients with HNC. Gross tumor volume and organ-at-risk contours were propagated from the planning CT to the weekly cone beam CT using deformable registration and reviewed by clinical users. This study analyzed 165 evaluable patients with AWARE-HNC treated between 2021 and 2024. The National Cancer Institute Common Terminology Criteria for Adverse Events version 5.0 was employed to grade toxicities. Acute xerostomia was defined as grade ≥2, occurring during or within 3 months after radiation therapy completion. We compared weekly salivary (parotid and submandibular) gland shrinkage and radiation doses between patients with and without acute xerostomia. Results:Of the 165 patients with HNC, 66 (40.0%) developed acute xerostomia. Patients with acute xerostomia exhibited significantly larger parotid gland shrinkage in the second and third weeks and submandibular gland shrinkage in the second week (15.4%, 21.0%, and 13.1%, respectively; t test P = .047, .032, and .007) during radiation therapy, compared with those without acute xerostomia (11.4%, 17.2%, and 7.7%). Compared with the nonxerostomia group, patients who developed acute xerostomia received significantly higher mean doses to the ipsilateral parotid (24.7 Gy vs 19.8 Gy; P = .003) and ipsilateral submandibular glands (60.7 Gy vs 57.4 Gy; P = .045). Lower parotid dose in the cone-down boost and subsequent adaptation mitigated the risk of acute xerostomia (P = .015). Conclusions:Increased mean doses to the ipsilateral parotid and submandibular glands were associated with larger salivary gland shrinkage during radiation therapy and a significantly higher risk of acute xerostomia. This early signal could trigger adaptive replanning promptly. Cone-down and adaptive techniques can reduce the salivary gland dose, helping mitigate the risk of acute xerostomia.

PURPOSE:To develop, clinically implement, and use automated treatment planning for lung cancer radiotherapy (RT) via an in-house treatment planning optimization system, the expedited constrained hierarchical optimization (ECHO). METHODS:The ECHO system accepts segmented tumor and normal tissue contours and clinical dose/volume criteria as inputs and generates optimized fluence maps for intensity-modulated RT (IMRT) and leaf trajectories for volumetric-modulated arc therapy (VMAT). Dose/volume criteria for our clinically used seven lung cancer fractionation schedules were implemented in ECHO. For each schedule, ECHO internal optimization parameters were tuned using 5-7 previously treated patients and validated with additional 20-25 patients. RESULTS:Since May 2021, a total of 431 lung cancer patients have been treated with ECHO IMRT. Additionally, from April 2023, 93 lung SBRT patients were planned with ECHO VMAT. Treatment plans optimized with ECHO IMRT and VMAT provide more consistent target coverage and similar organ sparing compared with manually optimized plans. The average planning target volume was 390 cm3 (range: 12-3441) for IMRT and 17 cm3 (range: 2.5-56) for VMAT. The average optimization time was 33 min (range: 7-126) for IMRT and 56 min (range: 21-178) for VMAT. On average, ECHO-optimized plans save 1.5 h per plan compared to manual planning. CONCLUSIONS:We have implemented an automated system for lung RT treatment planning at our institution for seven fractionation schedules. Our ECHO approach is robust and adapts to changes in clinical criteria without requiring algorithm modifications or retraining. The proposed treatment planning automation framework saves resources and improves treatment plan quality and consistency.

PURPOSE:We report our experience with the implementation of a same-day simulation and treatment C-arm linear accelerator (linac)-based stereotactic program for patients with intracranial and extracranial metastatic disease. METHODS:Between May 2021 and October 2023, patients were treated in our same-day program with linac-based SRS/SBRT. Two slots per week were offered. Patients with expedited clinical needs, able to undergo SRS/SBRT simulation and treatment, were considered. Extracranial treatments were required to meet standards for automated intensity modulated radiation therapy (IMRT) optimization. Intracranial treatments were limited to 1-3 lesions and 1-2 isocenters. The day before treatment, the patient needed to be identified, and any diagnostic imaging had to be available for the physician and dosimetrist to discuss the plan. On the day of treatment, simulation was scheduled for 8 AM and treatment at 4 PM by default, with the goal to complete treatment by 6 PM. We analyzed information about each patient's treatment plan and time spent on each step of the workflow. RESULTS:Ninety-seven patients followed our same-day workflow and were included in the analysis. Seventy-five patients received intracranial SRS (57% to 1 lesion), while 22 patients received extracranial treatments (50% to the extremities). Simulation often required additional time to be completed, finishing a median 18 min (IQR 5-40) after the goal end time. The median time between simulation completion and end of the same-day treatment was 7.8 h (IQR 7.4-8.6). Treatment technique and the number of target volumes had a significant impact on planning time. The median treatment end time was 5:13 PM (IQR 4:46 PM-6:01 PM), with 74% ending by 6 PM. CONCLUSIONS:A linac-based program to treat patients with SRS/SBRT in an expedited fashion was established and successfully treated patients in a same-day timeline. Careful selection of planning techniques to limit plan complexity and adding automation in time-consuming parts of the process are crucial when developing expedited workflows.

Objective.Standardizing clinical CT images enables consistent analysis of image data for target delineation, radiomics, and machine learning in personalized medicine. The process of managing the scanned protocols in various CT scanners for radiotherapy simulation is underexplored in the literature. This study uses noise evaluation and prediction models to harmonize CT protocols across scanner models and manufacturers, ensuring reliable data for radiotherapy planning.Approach.A global noise index (GNI) was calculated from 1581 clinical CT exams obtained on five scanners (three from vendorPand two from vendorS: SpandSc). Exams were categorized by anatomical site. GNI was assessed (I) within the same model, (II) between models from the same manufacturer, and (III) across manufacturers. One-way ANOVA (I, III) and studentt-tests (II) evaluated significance (p< 0.05). Predictive models were created and validated with 90 further exams, establishing a reference GNI (GNIref) for future optimization.Results.GNI showed minor variations among P-type and betweenSpandScscanners, butSscanners differed fromP. Predictive model error ranged from 0.8 to 1.5 hounsfield units (HU). GNI differences betweenSandPscanners were <1 HU for head, neck, and paraspinal protocols, butSscanners had 1.5-2 HU higher GNI for the abdomen, pelvis, breast, and lungs.Conclusion.Scanners of the same model show slight variation; minor noise differences exist between manufacturers. Predictive modeling can estimate CT noise and support protocol optimization. The reference GNI of an anatomical site can be derived from sufficient CT exams, with or without the predictive mode; a 1-2 HU difference in GNIrefis achievable if the protocol is properly translated.

BACKGROUND:Free-breathing gated CBCT (gCBCT) is commonly prescribed for lung cancer patients undergoing respiratory gating radiotherapy. Recently, the nonstop gated CBCT (ngCBCT) has been proposed to significantly reduce scanning time and imaging dose while preserving high image quality. PURPOSE:To implement the novel ngCBCT imaging technique on a C-arm linear accelerator (LINAC) and quantitatively compare its scan time and imaging dose with those of the current clinical gCBCT. METHODS:ngCBCT was implemented via a customized XML file in the developer mode of a C-arm LINAC, while gCBCT was acquired in the clinical mode. Both techniques employed the same thorax imaging protocol (half fan, full trajectory). Scan times were calculated from the timestamps of acquired projection data. Imaging dose was characterized using the weighted Cone-Beam Dose Index (CBDIw), measured with a standard CTDI body phantom and two pencil chambers placed centrally and peripherally. Respiratory motion was simulated using a CIRS motion platform with both Cos4 waveforms (3-6 s cycles) and seven clinical patient breathing traces. Gating duty cycles of 30%-60% were tested for Cos4 motion, while the same gating window was reproduced for each patient's breathing trace. RESULTS:Scan times for gCBCT ranged from 1.8 to 5 min, influenced by the gating duty cycle, breathing period, and waveform periodicity. In contrast, ngCBCT consistently achieved scan times of approximately 1 min. The imaging dose (CBDIw) for ngCBCT was reduced to 26.7%-60.1% of that for gCBCT, closely matching the respective gating duty cycles. CONCLUSION:This study demonstrates that ngCBCT acquisition is feasible on C-arm LINAC and offers substantial improvements in scan time and dose reduction compared to current clinical gCBCT. This novel technique has the potential to enhance patient comfort and broaden access to respiratory gating radiotherapy.

Aim Artificial intelligence (AI) based auto-segmentation aids radiation therapy (RT) workflows and is being adopted in clinical environments facilitated by the increased availability of commercial solutions for organs at risk (OARs). In addition, open-source imaging datasets support training for new auto-segmentation algorithms. Here, we studied if the female and male anatomies are equally represented among these solutions. Materials and Methods Inquiries were sent to eight vendors regarding their clinically available OAR auto-segmentation solutions for each gender. The Cancer Imaging Archive (TCIA) was also screened for publicly available imaging datasets specific to the female and the male anatomy. Results All vendors provided AI based auto-segmentation solutions for the male pelvis and female breasts, while 5/8 vendors provided solutions for the female pelvis. The female breast and the female pelvis solutions were released at a median of 0.6 years and 2.3 years, respectively, after the release of the male pelvis solutions. Among 27 TCIA datasets identified, 15 involved the female anatomy (breast: 10; pelvis: 5) and 12 involved the male pelvis but no female-specific dataset included OAR segmentations, while three male pelvis datasets included OARs (ejaculatory duct, neurovascular bundle, penile bulb and verumontanum). Conclusion Commercial AI auto-segmentation solutions and open-source imaging datasets include considerably more solutions and OAR segmentations for male cancer over female cancer sites. This gender disparity is likely to propagate throughout the RT pipeline.

BACKGROUND:Free-breathing gated cone-beam computed tomography (gCBCT), which captures a specific anatomy coinciding with a preset gating window in the breathing cycle, is routinely prescribed to gating lung SBRT patients for pretreatment setup verification. However, a half-fan gCBCT scan can take 2-8 min (for a typical gating duty cycle of 30%-60% and patient breathing period of 3-6 s) on a C-arm linear accelerator because the gantry movement is interrupted and resumed by the respiratory gating signal multiple times over the scan. The long scan time increases patient on-table time, leading to discomfort and a higher likelihood of patient movement. Meanwhile, extra kV projections are acquired while the gantry is accelerating for the gCBCT scan, resulting in a higher imaging dose compared to 3D CBCT. PURPOSE:To investigate the feasibility of a novel imaging paradigm named "nonstop gated CBCT (ngCBCT)" that improves upon current clinical gCBCT by substantially reducing the scan time and imaging dose while retaining high-quality images. METHODS:The ngCBCT is implemented by allowing the gantry to rotate continuously, with the kV x-ray beam activated only when the breathing signal falls within the preset gating window. Raw gCBCT projections of two gating lung SBRT patients were retrospectively retrieved and intentionally sampled based on each patient's breathing cycle to emulate the ngCBCT acquisitions. The datasets include both half-fan and full-fan acquisitions, representing the primary clinical scan geometries. Three reconstruction algorithms-Feldkamp-Davis-Kress (FDK), penalized likelihood iterative reconstruction (PL), and prior-image-based iterative reconstruction (PIBR)-were applied to these ngCBCT emulations to evaluate reconstruction performances on the non-uniform and under-sampled projections resulting from this acquisition strategy. RESULTS:The FDK reconstructions of ngCBCT are degraded with streak artifacts and have insufficient quality for clinical use. While PL yields improved reconstructions over FDK, the PIBR method consistently delivers the best visual and quantitative results with the aid of patient-specific prior images. CONCLUSION:The proposed ngCBCT technique addresses the key limitations of current clinical gCBCT by substantially reducing data acquisition time and imaging dose. The ngCBCT with PIBR achieves adequate image quality and offers a promising opportunity for pretreatment setup verification in gating lung SBRT.

Purpose Deformable image registration (DIR) represents an inherently ill-posed problem, and its quality highly depends on the algorithm and user input, which can severely affect its applications in adaptive radiation therapy. We propose an automated framework for integrating DIR uncertainty into dose accumulation. Methods and Materials A hyperparameter perturbation approach was applied to estimate an ensemble of deformation vector fields for a given computed tomography (CT) to cone beam CT (CBCT) DIR. For each voxel, a principal component analysis was performed on the distribution of homologous points to construct voxel-specific DIR uncertainty confidence ellipsoids. During the resampling process for dose mapping, the complete dose within each ellipsoid was evaluated via interpolation to estimate the upper and lower dose limits for the particular voxel. We applied the proposed framework in a retrospective dose accumulation study of 20 patients with lung cancer who underwent image guided radiation therapy with weekly CBCTs. Results The average computational time was around 30 minutes, making the approach clinically feasible for automated offline evaluations. The uncertainty (ie, largest ellipsoid semiaxis length) for the fifth week CBCT DIR was 3.8 ± 1.8, 2.5 ± 0.7, 1.5 ± 0.4, 3.2 ± 1.3, and 4.5 ± 1.8 mm for the gross tumor volume, esophagus, spinal cord, lungs, and heart, respectively. Confidence ellipsoids were markedly elongated, with the largest semiaxis 5.5 and 2.5 times longer than the other axes. The dosimetric uncertainties were mainly within 4 Gy but exhibited significant spatial variation because of the interplay between dose gradient and DIR uncertainty. When DIR uncertainty was considered in dose accumulation, 3 cases exceeded institutional limits for dose-volume histogram metrics, highlighting the importance of considering the inherent uncertainty of DIR. Conclusions This framework has the potential to facilitate the clinical implementation of dose accumulation, which can improve clinical decision-making in adaptive radiation therapy and provide more personalized radiation therapy treatments.

Purpose:To compare the dosimetry and treatment efficiency of lung stereotactic body radiation therapy (SBRT) using the deep inspiration breath hold (DIBH), free breathing (FB), and respiratory gating (RG) strategies. Methods and Materials:308 lung SBRT patients with middle to lower zone lung tumors were included in this retrospective study. The prescriptions were 1000 cGy x 5 fractions, 1200 cGy x 4 fractions, or 1800 cGy x 3 fractions. They were all treated with a volumetric modulated arc therapy (VMAT) technique and 6 MV flattening filter free (FFF) beam on C-arm linear accelerators, but using different motion management strategies (151 DIBH, 136 FB, 21 RG). The lung dose (mean lung dose (MLD), V5, V20) and treatment time (on table, imaging & verification, delivery) of these patients were retrospectively collected for statistical comparison. Results:The average doses (MLD, V5, V20) to the ipsilateral lung were 408.2 cGy, 20.1 %, 5.7 % for the DIBH cohort, 569.8 cGy, 27.6 %, 8.4 % for the FB cohort, and 519.6 cGy, 23.5 %, 7.5 % for the RG patients. Correspondingly, the average time (on table/imaging & verification/delivery) for the three patient cohorts was 22.3/16.0/6.3 min, 13.6/10.5/3.1 min, and 22.7/14.6/8.1 min, respectively. Conclusion:Quantitative comparison of lung dose and treatment efficiency for three commonly used motion management strategies in lung SBRT is reported. While the relative advantages and disadvantages of these strategies are well recognized, our findings further confirm these differences and provide clinicians with quantitative data to support informed decision-making in clinical practice.

BACKGROUND:In image-guided radiotherapy (IGRT), four-dimensional cone-beam computed tomography (4D-CBCT) is critical for assessing tumor motion during a patient's breathing cycle prior to beam delivery. However, generating 4D-CBCT images with sufficient quality requires significantly more projection images than a standard 3D-CBCT scan, leading to extended scanning times and increased imaging dose to the patient. PURPOSE:To introduce a novel spatiotemporal Gaussian neural representation framework to reconstruct high-temporal dynamic CBCT images from 1-minute acquisition, preserving motion dynamics and fine spatial details without relying on prior images or motion models. METHODS:Our framework employs a differentiable 4D Gaussian representation initialized from average CBCT images. Gaussian points are characterized by position, covariance, rotation, and density, offering a compact and dynamic model for CBCT scenes. A Gaussian deformation network, incorporating a HexPlane encoder and multi-head decoder, predicts Gaussian deformations to minimize L1 and structural similarity index measure (SSIM) losses between rendered and measured projections. Adaptive Gaussian control refines the representation by pruning underutilized Gaussians and densifying points in high-gradient regions. The method was benchmarked on the AAPM SPARE challenge datasets and further validated with clinical CBCT scans from a Varian TrueBeam system. For the AAPM SPARE challenge datasets, the performance of the proposed method was evaluated using the root-mean-squared-error (RMSE) and the structural similarity index (SSIM) in the four regions of interest: Body, Lung, PTV, and Bone. The geometric accuracy was evaluated by calculating the registration error when aligning the tumor to the ground truth using the Elastix package, focusing on pixels within the planning target volume (PTV). To demonstrate our method's capability in high-temporal motion dynamic modeling using extremely undersampled projections, the clinical half-fan projections from a 1-minute Varian TrueBeam acquisition were sorted into 50 phases with approximately 18 projections per phase, significantly finer than the commonly used 10-phase binning. RESULTS:Compared to the AAPM SPARE challenge participant methods, our method achieved superior geometric accuracy in terms of PTV alignment error, and comparable RMSE and SSIM when no prior 4DCT or motion model is used for our reconstruction. For PTV alignment, our method achieved translational and rotational errors of 0.54 mm (LR), 0.76 mm (SI), 1.36 mm (AP), 0.55° (rAP), and 0.93° (rSI), and 1.31° (rLR), respectively. For high temporal dynamic CBCT reconstruction, our method successfully reconstructed a 50-phase CBCT from a 1-minute Varian Truebeam half-fan scan, demonstrating effective streak artifact suppression, respiratory motion preservation, and fine detail restoration. Reconstruction on a single NVIDIA RTX A6000 GPU required approximately 30-80 min, depending on the number of Gaussian points used (ranging from 50 to 400K), to reconstruct CBCT from 680 projections acquired with a 30 × 40 cm detector. Our code and reconstruction results can be found at: https://github.com/fuyabo/4DGS_for_4DCBCT/tree/main. CONCLUSIONS:The spatiotemporal Gaussian framework is a novel data-driven dynamic CBCT reconstruction technique that features excellent geometric accuracy in terms of PTV alignment and high-temporal motion modeling, indicating promise for tumor motion assessment and high-temporal respiratory motion modeling based on a 1-minute half-fan scan prior to beam delivery.

BACKGROUND:The synthesis of CT from CBCT images using AI methods has been explored in radiotherapy to improve adaptive workflows. However, the model training process can be particularly challenging for the abdominal region due to dataset disparities between CT and CBCT images caused by organ motion, low soft tissue contrast, and inconsistencies in air volumes. These factors might impact the implicit prediction uncertainties, which are not actively considered on the synthetized images, overlooking poorly predicted image regions that might lead to inaccuracies in the dose calculation. PURPOSE:To evaluate the impact of the model uncertainty on the predicted Hounsfield Units (HU) and dose calculation on synthetic CT (sCT) for abdominal patients. METHODS:CBCT images from 65 abdominal patients were retrospectively used to generate sCT images. Rigid image registration (RIR) and deformable image registration (DIR) were individually implemented to create two datasets (D1 and D2) to train (80%), validate (10%), and test (10%) three models (M1: Unet, M2: Bayes-Unet, M3: cycle-GAN). Treatment plans were made on the ground truth CT (GTCT) and the sCTs for dose calculation comparison. The model performance was evaluated with mean absolute error (MAE) and root mean square error (RMSE), and the sCT quality was verified with structural similarity index measure (SSIM). Gamma index (2%/2  mm), D95% of PTV, and Dmean of liver were evaluated and compared between the plans calculated on the GTCT and the sCT. The voxel-wise uncertainty map for M1 and M3 were generated by calculating the standard variation of each voxel from training the model independently ten times. For M2 the Monte Carlo DropConnect method was implemented with 100 iterations. Finally, the uncertainty was associated with the accuracy of CT numbers and dose calculation. RESULTS:Across the three models {M1, M2, M3} trained with D1 and D2, the MAE were {50.9 ± 13.3} and {40.9 ± 11.5}, respectively, the RMSE were {68.3 ± 13.5} and {62.2 ± 10.7}, respectively, and the SSIM were {0.89 ± 0.05} and {0.94 ± 0.05}, respectively. For D1 and D2, the gamma rates were {96.3 ± 1.04} and {97.3 ± 0.2}, respectively. No major differences in DVH were noticed between GTCT and sCT (p < 00.1). The correlation between the whole sCT uncertainty maps and gamma index was statistically significant (Spearman's coefficient = 0.84, p < 0.001) and weak between the target volume uncertainty and gamma index (Spearman's coefficient = 0.01, p = 0.89). CONCLUSION:Using DIR resulted in improved performance across all three models. Metrics used to evaluate synthetic image accuracy might not reflect the uncertainty implications in image quality and dose calculations, which suggests the benefit of displaying uncertainty errors in AI generated sCT as a potential strategy to improve the evaluation of intra-fraction changes used for adaptive abdominal radiotherapy.

In image-guided radiotherapy (IGRT), four-dimensional cone-beam computed tomography (4D-CBCT) is critical for assessing tumor motion during a patients breathing cycle prior to beam delivery. However, generating 4D-CBCT images with sufficient quality requires significantly more projection images than a standard 3D-CBCT scan, leading to extended scanning times and increased imaging dose to the patient. To address these limitations, there is a strong demand for methods capable of reconstructing high-quality 4D-CBCT images from a 1-minute 3D-CBCT acquisition. The challenge lies in the sparse sampling of projections, which introduces severe streaking artifacts and compromises image quality. This paper introduces a novel framework leveraging spatiotemporal Gaussian representation for 4D-CBCT reconstruction from sparse projections, achieving a balance between streak artifact reduction, dynamic motion preservation, and fine detail restoration. Each Gaussian is characterized by its 3D position, covariance, rotation, and density. Two-dimensional X-ray projection images can be rendered from the Gaussian point cloud representation via X-ray rasterization. The properties of each Gaussian were optimized by minimizing the discrepancy between the measured projections and the rendered X-ray projections. A Gaussian deformation network is jointly optimized to deform these Gaussian properties to obtain a 4D Gaussian representation for dynamic CBCT scene modeling. The final 4D-CBCT images are reconstructed by voxelizing the 4D Gaussians, achieving a high-quality representation that preserves both motion dynamics and spatial detail. The code and reconstruction results can be found at https://github.com/fuyabo/4DGS_for_4DCBCT/tree/main.