Inter-fractional portability of deep learning models for lung target tracking on cine imaging acquired in MRI-guided radiotherapy.

MRI-guided radiotherapy systems enable beam gating by tracking the target on planar, two-dimensional cine images acquired during treatment. This study aims to evaluate how deep-learning (DL) models for target tracking that are trained on data from one fraction can be translated to subsequent fractions. Cine images were acquired for six patients treated on an MRI-guided radiotherapy platform (MRIdian, Viewray Inc.) with an onboard 0.35 T MRI scanner. Three DL models (U-net, attention U-net and nested U-net) for target tracking were trained using two training strategies: (1) uniform training using data obtained only from the first fraction with testing performed on data from subsequent fractions and (2) adaptive training in which training was updated each fraction by adding 20 samples from the current fraction with testing performed on the remaining images from that fraction. Tracking performance was compared between algorithms, models and training strategies by evaluating the Dice similarity coefficient (DSC) and 95% Hausdorff Distance (HD95) between automatically generated and manually specified contours. The mean DSC for all six patients in comparing manual contours and contours generated by the onboard algorithm (OBT) were 0.68 ± 0.16. Compared to OBT, the DSC values improved 17.0 - 19.3% for the three DL models with uniform training, and 24.7 - 25.7% for the models based on adaptive training. The HD95 values improved 50.6 - 54.5% for the models based on adaptive training. DL-based techniques achieved better tracking performance than the onboard, registration-based tracking approach. DL-based tracking performance improved when implementing an adaptive strategy that augments training data fraction-by-fraction.

A quality assurance framework for routine monitoring of deep learning cardiac substructure computed tomography segmentation models in radiotherapy.

BACKGROUND: For autosegmentation models, the data used to train the model (e.g., public datasets and/or vendor-collected data) and the data on which the model is deployed in the clinic are typically not the same, potentially impacting the performance of these models by a process called domain shift. Tools to routinely monitor and predict segmentation performance are needed for quality assurance. Here, we develop an approach to perform such monitoring and performance prediction for cardiac substructure segmentation. PURPOSE: To develop a quality assurance (QA) framework for routine or continuous monitoring of domain shift and the performance of cardiac substructure autosegmentation algorithms. METHODS: A benchmark dataset consisting of computed tomography (CT) images along with manual cardiac substructure delineations of 241 breast cancer radiotherapy patients were collected, including one "normal" image domain of clean images and five "abnormal" domains containing images with artifact (metal, contrast), pathology, or quality variations due to scanner protocol differences (field of view, noise, reconstruction kernel, and slice thickness). The QA framework consisted of an image domain shift detector which operated on the input CT images and a shape quality detector on the output of an autosegmentation model, and a regression model for predicting autosegmentation model performance. The image domain shift detector was composed of a trained denoising autoencoder (DAE) and two hand-engineered image quality features to detect normal versus abnormal domains in the input CT images. The shape quality detector was a variational autoencoder (VAE) trained to estimate the shape quality of the auto-segmentation results. The output from the image domain shift and shape quality detectors was used to train a regression model to predict the per-patient segmentation accuracy, measured by Dice coefficient similarity (DSC) to physician contours. Different regression techniques were investigated including linear regression, Bagging, Gaussian process regression, random forest, and gradient boost regression. Of the 241 patients, 60 were used to train the autosegmentation models, 120 for training the QA framework, and the remaining 61 for testing the QA framework. A total of 19 autosegmentation models were used to evaluate QA framework performance, including 18 convolutional neural network (CNN)-based and one transformer-based model. RESULTS: When tested on the benchmark dataset, all abnormal domains resulted in a significant DSC decrease relative to the normal domain for CNN models ( p < 0.001 $p < 0.001$ ), but only for some domains for the transformer model. No significant relationship was found between the performance of an autosegmentation model and scanner protocol parameters ( p = 0.42 $p = 0.42$ ) except noise ( p = 0.01 $p = 0.01$ ). CNN-based autosegmentation models demonstrated a decreased DSC ranging from 0.07 to 0.41 with added noise, while the transformer-based model was not significantly affected (ANOVA, p = 0.99 $p=0.99$ ). For the QA framework, linear regression models with bootstrap aggregation resulted in the highest mean absolute error (MAE) of 0.041 ± 0.002 $0.041 \pm 0.002$ , in predicted DSC (relative to true DSC between autosegmentation and physician). MAE was lowest when combining both input (image) detectors and output (shape) detectors compared to output detectors alone. CONCLUSIONS: A QA framework was able to predict cardiac substructure autosegmentation model performance for clinically anticipated "abnormal" domain shifts.

Interobserver variability in target definition for stereotactic arrhythmia radioablation.

Background: Stereotactic arrhythmia radioablation (STAR) is a potential new therapy for patients with refractory ventricular tachycardia (VT). The arrhythmogenic substrate (target) is synthesized from clinical and electro-anatomical information. This study was designed to evaluate the baseline interobserver variability in target delineation for STAR. Methods: Delineation software designed for research purposes was used. The study was split into three phases. Firstly, electrophysiologists delineated a well-defined structure in three patients (spinal canal). Secondly, observers delineated the VT-target in three patients based on case descriptions. To evaluate baseline performance, a basic workflow approach was used, no advanced techniques were allowed. Thirdly, observers delineated three predefined segments from the 17-segment model. Interobserver variability was evaluated by assessing volumes, variation in distance to the median volume expressed by the root-mean-square of the standard deviation (RMS-SD) over the target volume, and the Dice-coefficient. Results: Ten electrophysiologists completed the study. For the first phase interobserver variability was low as indicated by low variation in distance to the median volume (RMS-SD range: 0.02-0.02 cm) and high Dice-coefficients (mean: 0.97 ± 0.01). In the second phase distance to the median volume was large (RMS-SD range: 0.52-1.02 cm) and the Dice-coefficients low (mean: 0.40 ± 0.15). In the third phase, similar results were observed (RMS-SD range: 0.51-1.55 cm, Dice-coefficient mean: 0.31 ± 0.21). Conclusions: Interobserver variability is high for manual delineation of the VT-target and ventricular segments. This evaluation of the baseline observer variation shows that there is a need for methods and tools to improve variability and allows for future comparison of interventions aiming to reduce observer variation, for STAR but possibly also for catheter ablation.

Initial clinical experience building a dual CT- and MR-guided adaptive radiotherapy program.

Introduction: Our institution was the first in the world to clinically implement MR-guided adaptive radiotherapy (MRgART) in 2014. In 2021, we installed a CT-guided adaptive radiotherapy (CTgART) unit, becoming one of the first clinics in the world to build a dual-modality ART clinic. Herein we review factors that lead to the development of a high-volume dual-modality ART program and treatment census over an initial, one-year period. Materials and Methods: The clinical adaptive service at our institution is enabled with both MRgART (MRIdian, ViewRay, Inc, Mountain View, CA) and CTgART (ETHOS, Varian Medical Systems, Palo Alto, CA) platforms. We analyzed patient and treatment information including disease sites treated, radiation dose and fractionation, and treatment times for patients on these two platforms. Additionally, we reviewed our institutional workflow for creating, verifying, and implementing a new adaptive workflow on either platform. Results: From October 2021 to September 2022, 256 patients were treated with adaptive intent at our institution, 186 with MRgART and 70 with CTgART. The majority (106/186) of patients treated with MRgART had pancreatic cancer, and the most common sites treated with CTgART were pelvis (23/70) and abdomen (20/70). 93.0% of treatments on the MRgART platform were stereotactic body radiotherapy (SBRT), whereas only 72.9% of treatments on the CTgART platform were SBRT. Abdominal gated cases were allotted a longer time on the CTgART platform compared to the MRgART platform, whereas pelvic cases were allotted a shorter time on the CTgART platform when compared to the MRgART platform. Our adaptive implementation technique has led to six open clinical trials using MRgART and seven using CTgART. Conclusions: We demonstrate the successful development of a dual platform ART program in our clinic. Ongoing efforts are needed to continue the development and integration of ART across platforms and disease sites to maximize access and evidence for this technique worldwide.

Prospective In Silico Evaluation of Cone-Beam Computed Tomography-Guided StereoTactic Adaptive Radiation Therapy (CT-STAR) for the Ablative Treatment of Ultracentral Thoracic Disease.

Purpose: We conducted a prospective, in silico study to evaluate the feasibility of cone-beam computed tomography (CBCT)-guided stereotactic adaptive radiation therapy (CT-STAR) for the treatment of ultracentral thoracic cancers (NCT04008537). We hypothesized that CT-STAR would reduce dose to organs at risk (OARs) compared with nonadaptive stereotactic body radiation therapy (SBRT) while maintaining adequate tumor coverage. Methods and Materials: Patients who were already receiving radiation therapy for ultracentral thoracic malignancies underwent 5 additional daily CBCTs on the ETHOS system as part of a prospective imaging study. These were used to simulate CT-STAR, in silico. Initial, nonadaptive plans (PI) were created based on simulation images and simulated adaptive plans (PA) were based on study CBCTs. 55 Gy/5 fractions was prescribed, with OAR constraint prioritization over PTV coverage under a strict isotoxicity approach. PI were applied to patients' anatomy of the day and compared with daily PA using dose-volume histogram metrics, with selection of superior plans for simulated delivery. Feasibility was defined as completion of the end-to-end adaptive workflow while meeting strict OAR constraints in ≥80% of fractions. CT-STAR was performed under time pressures to mimic clinical adaptive processes. Results: Seven patients were accrued, 6 with intraparenchymal tumors and 1 with a subcarinal lymph node. CT-STAR was feasible in 34 of 35 simulated fractions. In total, 32 dose constraint violations occurred when the PI was applied to anatomy-of-the-day across 22 of 35 fractions. These violations were resolved by the PA in all but one fraction, in which the proximal bronchial tree dose was still numerically improved through adaptation. The mean difference between the planning target volume and gross total volume V100% in the PI and the PA was -0.24% (-10.40 to 9.90) and -0.62% (-11.00 to 8.00), respectively. Mean end-to-end workflow time was 28.21 minutes (18.02-50.97). Conclusions: CT-STAR widened the dosimetric therapeutic index of ultracentral thorax SBRT compared with nonadaptive SBRT. A phase 1 protocol is underway to evaluate the safety of this paradigm for patients with ultracentral early-stage NSCLC.

Knowledge-based adaptive planning quality assurance using dosimetric indicators for stereotactic adaptive radiotherapy for pancreatic cancer.

INTRODUCTION: We aimed to develop knowledge-based tools for robust adaptive radiotherapy (ART) planning to determine on-table adaptive DVH metric variations or planning process errors for stereotactic pancreatic ART. We developed volume-based dosimetric identifiers to identify deviations of ART plans from simulation plans. MATERIALS AND METHODS: Two patient cohorts who were treated on MR-Linac for pancreas cancer were included in this retrospective study; a training cohort and a validation cohort. All patients received 50 Gy in 5 fractions. PTV-OPT was generated by subtracting the critical organs plus a 5 mm-margin from PTV. Several metrics that potentially can identify failure-modes were calculated including PTV & PTV_OPT V95% and PTV & PTV_OPT D95%/D5%. The difference between each DVH metric in each adaptive plan with the DVH metric in simulation plan was calculated. The 95% confidence interval (CI) of the variations in each DVH metric was calculated for the patient training cohort. Variations in DVH metrics that exceeded the 95% CI for all fractions in training and validation cohort were flagged for retrospective investigation for root-cause analysis to determine their predictive power for identifying failure-modes. RESULTS: The CIs for the PTV & PTV_OPT V95% and PTV & PTV_OPT D95%/D5% were ± 13%, ± 5%, ± 0.1, ± 0.03, respectively. We estimated the positive predictive value and negative predictive value of our method to be 77% and 89%, respectively, for the training cohort, and 80% for both in the validation cohort. DISCUSSION: We developed dosimetric indicators for ART planning QA to identify population-based deviations or planning errors during online adaptive process for stereotactic pancreatic ART. This technology may be useful as an ART clinical trial QA tool and improve overall ART quality at an institution.

Phantom-based Quality Assurance of a Clinical Dose Accumulation Technique Used in an Online Adaptive Radiation Therapy Platform.

Purpose: This study aimed to develop a routine quality assurance method for a dose accumulation technique provided by a radiation therapy platform for online treatment adaptation. Methods and Materials: Two commonly used phantoms were selected for the dose accumulation QA: Electron density and anthropomorphic pelvis. On a computed tomography (CT) scan of the electron density phantom, 1 target (gross tumor volume [GTV]; insert at 6 o'clock), a subvolume within this target, and 7 organs at risk (OARs; other inserts) were contoured in the treatment planning system (TPS). Two adaptation sessions were performed in which the GTV was recontoured, first at 7 o'clock and then at 5 o'clock. The accumulated dose was exported from the TPS after delivery. Deformable vector fields were also exported to manually accumulate doses for comparison. For the pelvis phantom, synthetic Gaussian deformations were applied to the planning CT image to simulate organ changes. Two single-fraction adaptive plans were created based on the deformed planning CT and cone beam CT images acquired onboard the radiation therapy platform. A manual dose accumulation was performed after delivery using the exported deformable vector fields for comparison with the system-generated result. Results: All plans were successfully delivered, and the accumulated dose was both manually calculated and derived from the TPS. For the electron density phantom, the average mean dose differences in the GTV, boost volume, and OARs 1 to 7 were 0.0%, -0.2%, 92.0%, 78.4%, 1.8%, 1.9%, 0.0%, 0.0%, and 2.3%, respectively, between the manually summed and platform-accumulated doses. The gamma passing rates for the 3-dimensional dose comparison between the manually generated and TPS-provided dose accumulations were >99% for both phantoms. Conclusions: This study demonstrated agreement between manually obtained and TPS-generated accumulated doses in terms of both mean structure doses and local 3-dimensional dose distributions. Large disagreements were observed for OAR1 and OAR2 defined on the electron density phantom due to OARs having lower deformation priority over the target in addition to artificially large changes in position induced for these structures fraction-by-fraction. The tests applied in this study to a commercial platform provide a straightforward approach toward the development of routine quality assurance of dose accumulation in online adaptation.

Deep learning-based motion compensation for four-dimensional cone-beam computed tomography (4D-CBCT) reconstruction.

BACKGROUND: Motion-compensated (MoCo) reconstruction shows great promise in improving four-dimensional cone-beam computed tomography (4D-CBCT) image quality. MoCo reconstruction for a 4D-CBCT could be more accurate using motion information at the CBCT imaging time than that obtained from previous 4D-CT scans. However, such data-driven approaches are hampered by the quality of initial 4D-CBCT images used for motion modeling. PURPOSE: This study aims to develop a deep-learning method to generate high-quality motion models for MoCo reconstruction to improve the quality of final 4D-CBCT images. METHODS: A 3D artifact-reduction convolutional neural network (CNN) was proposed to improve conventional phase-correlated Feldkamp-Davis-Kress (PCF) reconstructions by reducing undersampling-induced streaking artifacts while maintaining motion information. The CNN-generated artifact-mitigated 4D-CBCT images (CNN enhanced) were then used to build a motion model which was used by MoCo reconstruction (CNN+MoCo). The proposed procedure was evaluated using in-vivo patient datasets, an extended cardiac-torso (XCAT) phantom, and the public SPARE challenge datasets. The quality of reconstructed images for XCAT phantom and SPARE datasets was quantitatively assessed using root-mean-square-error (RMSE) and normalized cross-correlation (NCC). RESULTS: The trained CNN effectively reduced the streaking artifacts of PCF CBCT images for all datasets. More detailed structures can be recovered using the proposed CNN+MoCo reconstruction procedure. XCAT phantom experiments showed that the accuracy of estimated motion model using CNN enhanced images was greatly improved over PCF. CNN+MoCo showed lower RMSE and higher NCC compared to PCF, CNN enhanced and conventional MoCo. For the SPARE datasets, the average (± standard deviation) RMSE in mm-1 for body region of PCF, CNN enhanced, conventional MoCo and CNN+MoCo were 0.0040 ± 0.0009, 0.0029 ± 0.0002, 0.0024 ± 0.0003 and 0.0021 ± 0.0003. Corresponding NCC were 0.84 ± 0.05, 0.91 ± 0.05, 0.91 ± 0.05 and 0.93 ± 0.04. CONCLUSIONS: CNN-based artifact reduction can substantially reduce the artifacts in the initial 4D-CBCT images. The improved images could be used to enhance the motion modeling and ultimately improve the quality of the final 4D-CBCT images reconstructed using MoCo.

Prospective imaging comparison of anatomic delineation with rapid kV cone beam CT on a novel ring gantry radiotherapy device.

INTRODUCTION: A kV imager coupled to a novel, ring-gantry radiotherapy system offers improved on-board kV-cone-beam computed tomography (CBCT) acquisition time (17-40 seconds) and image quality, which may improve CT radiotherapy image-guidance and enable online adaptive radiotherapy. We evaluated whether inter-observer contour variability over various anatomic structures was non-inferior using a novel ring gantry kV-CBCT (RG-CBCT) imager as compared to diagnostic-quality simulation CT (simCT). MATERIALS/METHODS: Seven patients undergoing radiotherapy were imaged with the RG-CBCT system at breath hold (BH) and/or free breathing (FB) for various disease sites on a prospective imaging study. Anatomy was independently contoured by seven radiation oncologists on: 1. SimCT 2. Standard C-arm kV-CBCT (CA-CBCT), and 3. Novel RG-CBCT at FB and BH. Inter-observer contour variability was evaluated by computing simultaneous truth and performance level estimation (STAPLE) consensus contours, then computing average symmetric surface distance (ASSD) and Dice similarity coefficient (DSC) between individual raters and consensus contours for comparison across image types. RESULTS: Across 7 patients, 18 organs-at-risk (OARs) were evaluated on 27 image sets. Both BH and FB RG-CBCT were non-inferior to simCT for inter-observer delineation variability across all OARs and patients by ASSD analysis (p < 0.001), whereas CA-CBCT was not (p = 0.923). RG-CBCT (FB and BH) also remained non-inferior for abdomen and breast subsites compared to simCT on ASSD analysis (p < 0.025). On DSC comparison, neither RG-CBCT nor CA-CBCT were non-inferior to simCT for all sites (p > 0.025). CONCLUSIONS: Inter-observer ability to delineate OARs using novel RG-CBCT images was non-inferior to simCT by the ASSD criterion but not DSC criterion.

A clinical and time savings evaluation of a deep learning automatic contouring algorithm.

Automatic contouring algorithms may streamline clinical workflows by reducing normal organ-at-risk (OAR) contouring time. Here we report the first comprehensive quantitative and qualitative evaluation, along with time savings assessment for a prototype deep learning segmentation algorithm from Siemens Healthineers. The accuracy of contours generated by the prototype were evaluated quantitatively using the Sorensen-Dice coefficient (Dice), Jaccard index (JC), and Hausdorff distance (Haus). Normal pelvic and head and neck OAR contours were evaluated retrospectively comparing the automatic and manual clinical contours in 100 patient cases. Contouring performance outliers were investigated. To quantify the time savings, a certified medical dosimetrist manually contoured de novo and, separately, edited the generated OARs for 10 head and neck and 10 pelvic patients. The automatic, edited, and manually generated contours were visually evaluated and scored by a practicing radiation oncologist on a scale of 1-4, where a higher score indicated better performance. The quantitative comparison revealed high (> 0.8) Dice and JC performance for relatively large organs such as the lungs, brain, femurs, and kidneys. Smaller elongated structures that had relatively low Dice and JC values tended to have low Hausdorff distances. Poor performing outlier cases revealed common anatomical inconsistencies including overestimation of the bladder and incorrect superior-inferior truncation of the spinal cord and femur contours. In all cases, editing contours was faster than manual contouring with an average time saving of 43.4% or 11.8 minutes per patient. The physician scored 240 structures with > 95% of structures receiving a score of 3 or 4. Of the structures reviewed, only 11 structures needed major revision or to be redone entirely. Our results indicate the evaluated auto-contouring solution has the potential to reduce clinical contouring time. The algorithm's performance is promising, but human review and some editing is required prior to clinical use.

Simulation-Free Radiation Therapy: An Emerging Form of Treatment Planning to Expedite Plan Generation for Patients Receiving Palliative Radiation Therapy.

Purpose: Herein we report the clinical and dosimetric experience for patients with metastases treated with palliative simulation-free radiation therapy (SFRT) at a single institution. Methods and Materials: SFRT was performed at a single institution. Multiple fractionation regimens were used. Diagnostic imaging was used for treatment planning. Patient characteristics as well as planning and treatment time points were collected. A matched cohort of patients with conventional computed tomography simulation radiation therapy (CTRT) was acquired to evaluate for differences in planning and treatment time. SFRT dosimetry was evaluated to determine the fidelity of SFRT. Descriptive statistics were calculated on all variables and statistical significance was evaluated using the Wilcoxon signed rank test and t test methods. Results: Thirty sessions of SFRT were performed and matched with 30 sessions of CTRT. Seventy percent of SFRT and 63% of CTRT treatments were single fraction. The median time to plan generation was 0.88 days (0.19-1.47) for SFRT and 1.90 days (0.39-5.23) for CTRT (P = .02). The total treatment time was 41 minutes (28-64) for SFRT and 30 minutes (21-45) for CTRT (P = .02). In the SFRT courses, the maximum and mean deviations in the actual delivered dose from the approved plans for the maximum dose were 4.1% and 0.07%, respectively. All deliveries were within a 5% threshold and deemed clinically acceptable. Conclusions: Palliative SFRT is an emerging technique that allowed for a statistically significant lower time to plan generation and was dosimetrically acceptable. This benefit must be weighed against increased total treatment time for patients receiving SFRT compared with CTRT, and appropriate patient selection is critical.

Simulated computed tomography-guided stereotactic adaptive radiotherapy (CT-STAR) for the treatment of locally advanced pancreatic cancer.

BACKGROUND AND PURPOSE: We conducted a prospective, in silico imaging clinical trial to evaluate the feasibility and potential dosimetric benefits of computed tomography-guided stereotactic adaptive radiotherapy (CT-STAR) for the treatment of locally advanced pancreatic cancer (LAPC). MATERIALS AND METHODS: Eight patients with LAPC received five additional CBCTs on the ETHOS system before or after their standard of care radiotherapy treatment. Initial plans were created based on their initial simulation anatomy (PI) and emulated adaptive plans were created based on their anatomy-of-the-day (PA). The prescription was 50 Gy/5 fractions. Plans were created under a strict isotoxicity approach, in which organ-at-risk (OAR) constraints were prioritized over planning target volume coverage. The PI was evaluated on the patient's anatomy-of-the-day, compared to the daily PA, and the superior plan was selected. Feasibility was defined as successful completion of the workflow in compliance with strict OAR constraints in ≥80% of fractions. RESULTS: CT-STAR was feasible in silico for LAPC and improved OAR and/or target dosimetry in 100% of fractions. Use of the PI based on the patient's anatomy-of-the-day would have yielded a total of 94 OAR constraint violations and ≥1 hard constraint violation in 40/40 fractions. In contrast, 39/40 PA met all OAR constraints. In one fraction, the PA minimally exceeded the large bowel constraint, although dosimetrically improved compared to the PI. Total workflow time per fraction was 36.28 minutes (27.57-55.86). CONCLUSION: CT-STAR for the treatment of LAPC cancer proved feasible and was dosimetrically superior to non-adapted CT-stereotactic body radiotherapy.

Robustness and reproducibility of an artificial intelligence-assisted online segmentation and adaptive planning process for online adaptive radiation therapy.

Clinical implementation of online adaptive radiation therapy requires initial and ongoing performance assessment of the underlying auto-segmentation and adaptive planning algorithms, although a straightforward and efficient process for this in phantom is lacking. The purpose of this work was to investigate robustness and repeatability of the artificial intelligence-assisted online segmentation and adaptive planning process on the Varian Ethos adaptive platform, and to develop an end-to-end test strategy for online adaptive radiation therapy. Five synthetic deformations were generated and applied to a computed tomography image of an anthropomorphic pelvis phantom, and reference treatment plans were generated from each of the resulting deformed images. The undeformed phantom was repeatedly imaged, and the online adaptive process was performed including auto-segmentation, review and manual correction of contours, and adaptive plan creation. One adaptive fractions in five different deformation scenarios were performed. The manually corrected contours had a high degree of consistency (> 93% Dice similarity coefficient and < 1.0 mm mean surface distance) across repeated fractions, with no significant variation across the synthetic deformation instance except for bowel (p = 0.026, one-way ANOVA). Adaptive treatment plans also resulted in highly consistent dose-volume values for targets and organs at risk. A straightforward and efficient process was developed and used to quantify a set of organ specific contouring and dosimetric action levels to help establish uncertainty bounds for an end-to-end test on the Varian Ethos system.

In Silico Trial of Computed Tomography-Guided Stereotactic Adaptive Radiation Therapy (CT-STAR) for the Treatment of Abdominal Oligometastases.

PURPOSE: We conducted a prospective, in silico clinical imaging study (NCT04008537) to evaluate the feasibility of cone beam computed tomography-guided stereotactic adaptive radiation therapy (CT-STAR) for the treatment of abdominal oligometastases. We hypothesized that CT-STAR produces improved dosimetry compared with nonadapted CT-stereotactic body radiation therapy (SBRT). METHODS AND MATERIALS: Eight patients receiving stereotactic body radiation therapy for abdominal oligometastatic disease received 5 additional kV cone beam CTs on the ETHOS system. These additional cone beam CTs were used for imaging during an emulator treatment session. Initial plans were created based on their simulation (PI) and emulated adaptive plans were based on anatomy-of-the-day. The prescription was 50 Gy out of 5 fractions. Organ-at-risk (OAR) constraints were prioritized over planning target volume coverage under a strict isotoxicity approach. The PI was applied to the patient's anatomy-of-the-day and compared with the reoptimized adaptive plans using dose-volume histogram metrics, with selection of the superior plan. Feasibility was defined as completion of the adaptive workflow and compliance with strict OAR constraints in ≥80% of fractions. Fractions were performed under time pressures by a physician and physicist to mimic the adaptive process. RESULTS: CT-STAR was feasible, with successful workflow completion in 38 out of 40 (95%) fractions. PI application to daily anatomy created OAR constraint violations in 30 out of 40 (75%) fractions. There were 8 stomach, 18 duodenum, 16 small bowel, and 11 large bowel PI OAR constraint violations. In contrast, OAR violations occurred in 2 out of 40 (5%) adaptive plans (both small bowel violations, both improved from the PI). CT-STAR also improved gross tumor volume V100 and D95 coverage in 25 out of 40 (63%) and 20 out of 40 (50%) fractions, respectively. Zero out of 40 (0%) fractions were deemed nonfeasible due to poor image quality and/or inability to delineate structures. Adaptation time per fraction was a median of 22.59 minutes (10.97-47.23). CONCLUSIONS: CT-STAR resolved OAR hard constraint violations and/or improved target coverage in silico compared with nonadapted CT-guided stereotactic body radiation therapy for the ablation of abdominal oligometastatic disease. Although limitations of this study include its small sample size and in silico design, the consistently high-quality cone beam CT images captured and comparable timing metrics to prior adaptive studies suggest that CT- STAR is a viable treatment paradigm for the ablation of abdominal oligometastatic disease. Clinical trials are in development to further evaluate CT-STAR in the clinic.

Techno-Economic Feasibility Analysis of a Fully Mobile Radiation Oncology System Using Monte Carlo Simulation.

PURPOSE: Disparities in radiation oncology (RO) can be attributed to geographic location, socioeconomic status, race, sex, and other societal factors. One potential solution is to implement a fully mobile (FM) RO system to bring radiotherapy to rural areas and reduce barriers to access. We use Monte Carlo simulation to quantify techno-economic feasibility with uncertainty, using two rural Missouri scenarios. METHODS: Recently, a semimobile RO system has been developed by building an o-ring linear accelerator (linac) into a mobile coach that is used for temporary care, months at a time. Transitioning to a more FM-RO system, which changes location within a given day, presents technical challenges including logistics and quality assurance. This simulation includes cancer census in both northern and southeastern Missouri, multiple treatment locations within a given day, and associated expenditures and revenues. A subset of patients with lung, breast, and rectal diseases, treated with five fractions, was simulated in the FM-RO system. RESULTS: The FM-RO can perform all necessary quality assurance tests as suggested in national medical physics guidelines within 1.5 hours, thus demonstrating technological feasibility. In northern and southeastern Missouri, five-fraction simulations' net incomes were, in US dollars (USD), $1.55 ± 0.17 million (approximately 74 patients/year) and $3.65 USD ± 0.25 million (approximately 98 patients/year), respectively. The number of patients seen had the highest correlation with net income as well as the ability to break-even within the simulation. The model does not account for disruptions in care or other commonly used treatment paradigms, which may lead to differences in estimated economic return. Overall, the mobile system achieved a net benefit, even for the most negative simulation scenarios. CONCLUSION: Our simulations suggest technologic success and economic viability for a FM-RO system within rural Missouri and present an interesting solution to address other geographic disparities in access to radiotherapy.

Implementation of Magnetic Resonance Safety Program for Radiation Oncology.

During the last decade, radiation oncology departments have integrated magnetic resonance imaging (MRI) equipment, procedures, and expertise into their practices. MRI safety is an important consideration because a large percentage of patients receiving radiation therapy have histories of multiple surgeries and implanted devices. However, MRI safety guidelines and workflows were traditionally designed for radiology departments. This report presents an MR safety program designed for a radiation oncology department to address its specific needs.

Robustness of deep learning segmentation of cardiac substructures in noncontrast computed tomography for breast cancer radiotherapy.

PURPOSE: To develop and evaluate deep learning-based autosegmentation of cardiac substructures from noncontrast planning computed tomography (CT) images in patients undergoing breast cancer radiotherapy and to investigate the algorithm sensitivity to out-of-distribution data such as CT image artifacts. METHODS: Nine substructures including aortic valve (AV), left anterior descending (LAD), tricuspid valve (TV), mitral valve (MV), pulmonic valve (PV), right atrium (RA), right ventricle (RV), left atrium (LA), and left ventricle (LV) were manually delineated by a radiation oncologist on noncontrast CT images of 129 patients with breast cancer; among them 90 were considered in-distribution data, also named as "clean" data. The image/label pairs of 60 subjects were used to train a 3D deep neural network while the remaining 30 were used for testing. The rest of the 39 patients were considered out-of-distribution ("outlier") data, which were used to test robustness. Random rigid transformations were used to augment the dataset during training. We investigated multiple loss functions, including Dice similarity coefficient (DSC), cross-entropy (CE), Euclidean loss as well as the variation and combinations of these, data augmentation, and network size on overall performance and sensitivity to image artifacts due to infrequent events such as the presence of implanted devices. The predicted label maps were compared to the ground-truth labels via DSC and mean and 90th percentile symmetric surface distance (90th-SSD). RESULTS: When modified Dice combined with cross-entropy (MD-CE) was used as the loss function, the algorithm achieved a mean DSC = 0.79 ± 0.07 for chambers and  0.39 ± 0.10 for smaller substructures (valves and LAD). The mean and 90th-SSD were 2.7 ± 1.4 and 6.5 ± 2.8 mm for chambers and 4.1 ± 1.7 and 8.6 ± 3.2 mm for smaller substructures. Models with MD-CE, Dice-CE, MD, and weighted CE loss had highest performance, and were statistically similar. Data augmentation did not affect model performances on both clean and outlier data and model robustness was susceptible to network size. For a certain type of outlier data, robustness can be improved via incorporating them into the training process. The execution time for segmenting each patient was on an average 2.1 s. CONCLUSIONS: A deep neural network provides a fast and accurate segmentation of large cardiac substructures in noncontrast CT images. Model robustness of two types of clinically common outlier data were investigated and potential approaches to improve them were explored. Evaluation of clinical acceptability and integration into clinical workflow are pending.

Findings of the AAPM Ad Hoc committee on magnetic resonance imaging in radiation therapy: Unmet needs, opportunities, and recommendations.

The past decade has seen the increasing integration of magnetic resonance (MR) imaging into radiation therapy (RT). This growth can be contributed to multiple factors, including hardware and software advances that have allowed the acquisition of high-resolution volumetric data of RT patients in their treatment position (also known as MR simulation) and the development of methods to image and quantify tissue function and response to therapy. More recently, the advent of MR-guided radiation therapy (MRgRT) - achieved through the integration of MR imaging systems and linear accelerators - has further accelerated this trend. As MR imaging in RT techniques and technologies, such as MRgRT, gain regulatory approval worldwide, these systems will begin to propagate beyond tertiary care academic medical centers and into more community-based health systems and hospitals, creating new opportunities to provide advanced treatment options to a broader patient population. Accompanying these opportunities are unique challenges related to their adaptation, adoption, and use including modification of hardware and software to meet the unique and distinct demands of MR imaging in RT, the need for standardization of imaging techniques and protocols, education of the broader RT community (particularly in regards to MR safety) as well as the need to continue and support research, and development in this space. In response to this, an ad hoc committee of the American Association of Physicists in Medicine (AAPM) was formed to identify the unmet needs, roadblocks, and opportunities within this space. The purpose of this document is to report on the major findings and recommendations identified. Importantly, the provided recommendations represent the consensus opinions of the committee's membership, which were submitted in the committee's report to the AAPM Board of Directors. In addition, AAPM ad hoc committee reports differ from AAPM task group reports in that ad hoc committee reports are neither reviewed nor ultimately approved by the committee's parent groups, including at the council and executive committee level. Thus, the recommendations given in this summary should not be construed as being endorsed by or official recommendations from the AAPM.

Evaluation of Motion Compensation Methods for Noninvasive Cardiac Radioablation of Ventricular Tachycardia.

PURPOSE: Noninvasive cardiac radioablation is increasingly used for treatment of refractory ventricular tachycardia. Attempts to limit normal tissue exposure are important, including managing motion of the target. An interplay between cardiac and respiratory motion exists for cardiac radioablation, which has not been studied in depth. The objectives of this study were to estimate target motion during abdominal compression free breathing (ACFB) and respiratory gated (RG) deliveries and to investigate the quality of either implanted cardioverter defibrillator lead tip or the diaphragm as a gating surrogate. METHODS AND MATERIALS: Eleven patients underwent computed tomography (CT) simulation with an ACFB 4-dimensional CT (r4DCT) and an exhale breath-hold cardiac 4D-CT (c4DCT). The target, implanted cardioverter defibrillator lead tip and diaphragm trajectories were measured for each patient on the r4DCT and c4DCT using rigid registration of each 4D phase to the reference (0%) phase. Motion ranges for ACFB and exhale (40%-60%) RG delivery were estimated from the target trajectories. Surrogate quality was estimated as the correlation with the target motion magnitudes. RESULTS: Mean (range) target motion across patients from r4DCT was as follows: left/right (LR), 3.9 (1.7-6.9); anteroposterior (AP), 4.1 (2.2-5.4); and superoinferior (SI), 4.7 (2.2-7.9) mm. Mean (range) target motion from c4DCT was as follows: LR, 3.4 (1.0-4.8); AP, 4.3 (2.6-6.5); and SI, 4.1 (1.4-8.0) mm. For an ACFB, treatment required mean (range) margins to be 4.5 (3.1-6.9) LR, 4.8 (3-6.5) AP, and 5.5 (2.3-8.0) mm SI. For RG, mean (range) internal target volume motion would be 3.6 (1.1-4.8) mm LR, 4.3 (2.6-6.5) mm AP, and 4.2 (2.2-8.0) mm SI. The motion correlations between the surrogates and target showed a high level of interpatient variability. CONCLUSIONS: In ACFB patients, a simulated exhale-gated approach did not lead to large projected improvements in margin reduction. Furthermore, the variable correlation between readily available gating surrogates could mitigate any potential advantage to gating and should be evaluated on a patient-specific basis.

Technical Note: Self-shielding evaluation and radiation leakage measurement of a jawless ring gantry linac with a beam stopper.

PURPOSE: To characterize the shielding design and leakage radiation from a newly released ring gantry linac (Halcyon, Varian Medical Systems). METHODS: To assess the radiation leakage surrounding headshield and the radiation level after the beam stopper, measurements were made with GafChromic films. To evaluate the in-room radiation levels, the radiation leakage in the isocenter plane was measured with a large volume spherical ionization chamber (Exradin A6, Standard Imaging). A lead enclosure was constructed to shield the chamber from the low energy scatter radiation from the room. The radiation level at multiple locations was measured with the MLC fully closed and gantry at 0, 45, 90, 135, 180, 225, 270, and 315 degrees. The leakage radiation passing through multiple concrete slabs with various thickness was recorded in a narrow beam geometry to determine the tenth value layer (TVL). RESULTS: A uniform leakage (<0.05%) at 1 m from electron beam line was measured surrounding the linac head with the maximum leakage measured at the top of the head enclosure. The highest radiation level (<0.08%) was measured near the edge of the beam stopper when projected to the measurement plane. The maximum radiation levels due to the head leakage at 15 locations inside the treatment room were recorded and a radiation map was plotted. The maximum leakage was measured at points that along the electron beam line while the gantry at 90 or 270 degree and at the end of head enclosure (0.314%, 0.4 m from electron beamline). The leakage TVL value is found to be 226 mm in a narrow beam geometry with the concrete density of 2.16 g/cm3 or 134.6 lb/cu.ft. CONCLUSION: An overall uniform leakage was measured surrounding linac head. The beam stopper shields the primary radiation with the highest valued measured near the edge of beam stopper. The leakage TVL values are derived and less than the values reported for conventional C-arm linac.