Dosimetric characterization of a new surface-conforming electron MLC prototype.

The purpose is to reduce normal tissue radiation toxicity for electron therapy through the creation of a surface-conforming electron multileaf collimator (SCEM). The SCEM combines the benefits of skin collimation, electron conformal radiotherapy, and modulated electron radiotherapy. An early concept for the SCEM was constructed. It consists of leaves that protrude towards the patient, allowing the leaves to conform closely to irregular patient surfaces. The leaves are made of acrylic to decrease bremsstrahlung, thereby decreasing the out-of-field dose. Water tank scans were performed with the SCEM in place for various field sizes for all available electron energies (6, 9, 12, and 15 MeV) with a 0.5 cm air gap to the water surface at 100 cm source-to-surface distance (SSD). These measurements were compared with Cerrobend cutouts with the field size-matched at 100 and 110 cm SSD. Output factor measurements were taken in solid water for each energy at dmax for both the cerrobend cutouts and SCEM at 100 cm SSD. Percent depth dose (PDD) curves for the SCEM shifted shallower for all energies and field sizes. The SCEM also produced a higher surface dose relative to Cerrobend cutouts, with the maximum being a 9.8% increase for the 3 cm × 9 cm field at 9 MeV. When compared to the Cerrobend cutouts at 110 cm SSD, the SCEM showed a significant decrease in the penumbra, particularly for lower energies (i.e., 6 and 9 MeV). The SCEM also showed reduced out-of-field dose and lower bremsstrahlung production than the Cerrobend cutouts. The SCEM provides significant improvement in the penumbra and out-of-field dose by allowing collimation close to the skin surface compared to Cerrobend cutouts. However, the added scatter from the SCEM increases shallow PDD values. Future work will focus on reducing this scatter while maintaining the penumbra and out-of-field benefits the SCEM has over conventional collimation.

Standardizing and improving dose predictions for head and neck cancers using complete sets of OAR contours.

BACKGROUND: Radiotherapy dose predictions have been trained with data from previously treated patients of similar sites and prescriptions. However, clinical datasets are often inconsistent and do not contain the same number of organ at risk (OAR) structures. The effects of missing contour data in deep learning-based dose prediction models have not been studied. PURPOSE: The purpose of this study was to investigate the impacts of incomplete contour sets in the context of deep learning-based radiotherapy dose prediction models trained with clinical datasets and to introduce a novel data substitution method that utilizes automated contours for undefined structures. METHODS: We trained Standard U-Nets and Cascade U-Nets to predict the volumetric dose distributions of patients with head and neck cancers (HNC) using three input variations to evaluate the effects of missing contours, as well as a novel data substitution method. Each architecture was trained with the original contour (OC) inputs, which included missing information, hybrid contour (HC) inputs, where automated OAR contours generated in software were substituted for missing contour data, and automated contour (AC) inputs containing only automated OAR contours. 120 HNC treatments were used for model training, 30 were used for validation and tuning, and 44 were used for evaluation and testing. Model performance and accuracy were evaluated with global whole body dose agreement, PTV coverage accuracy, and OAR dose agreement. The differences in these values between dataset variations were used to determine the effects of missing data and automated contour substitutions. RESULTS: Automated contours used as substitutions for missing data were found to improve dose prediction accuracy in the Standard U-Net and Cascade U-Net, with a statistically significant difference in some global metrics and/or OAR metrics. For both models, PTV coverage between input variations was unaffected by the substitution technique. Automated contours in HC and AC datasets improved mean dose accuracy for some OAR contours, including the mandible and brainstem, with a greater improvement seen with HC datasets. Global dose metrics, including mean absolute error, mean error, and percent error were different for the Standard U-Net but not for the Cascade U-Net. CONCLUSION: Automated contours used as a substitution for contour data improved prediction accuracy for some but not all dose prediction metrics. Compared to the Standard U-Net models, the Cascade U-Net achieved greater precision.

Microdosimetric and radiobiological effects of gold nanoparticles at therapeutic radiation energies.

PURPOSE: The purpose of this study was to quantify the microscopic dose distribution surrounding gold nanoparticles (GNPs) irradiated at therapeutic energies and to measure the changes in cell survival in vitro caused by this dose enhancement. METHODS: The dose distributions from secondary electrons surrounding a single gold nanosphere and single gold nanocube of equal volume were both simulated using MCNP6. Dose enhancement factors (DEFs) in the 1 µm3 volume surrounding a GNP were calculated and compared between a nanosphere and nanocube and between 6 and 18 MV energies. This microscopic effect was explored further by experimentally measuring the cell survival of C-33a cervical cancer cells irradiated at 18 MV with varying doses of energy and concentrations of GNPs. Survival of cells receiving no irradiation, a 3 Gy dose, and a 6 Gy dose of 18 MV energy were determined for each concentration of GNPs. RESULTS: It was observed that the dose from electrons surrounding the gold nanocube surpasses that of a gold nanosphere up to a distance of 1.1 µm by 18.5% for the 18 MV energy spectrum and by 23.1% for the 6 MV spectrum. DEFs ranging from ∼2 to 8 were found, with the maximum DEF resulting from the case of the gold nanocube irradiated at 6 MV energy. Experimentally, for irradiation at 18 MV, incubating cells with 6 nM (0.10% gold by mass) GNPs produces an average 6.7% decrease in cell survival, and incubating cells with 9 nM (0.15% gold by mass) GNPs produces an average 14.6% decrease in cell survival, as compared to cells incubated and irradiated without GNPs. CONCLUSION: We have successfully demonstrated the potential radiation dose enhancing effects in vitro and microdosimetrically from gold nanoparticles.

Unique approaches for testicular shielding during total-body irradiation for pediatric patients.

Total-body irradiation (TBI) has been used as a part of the conditioning regimen for patients undergoing hematopoietic stem cell transplantation for certain nonmalignant conditions such as sickle cell disease. Although effective, TBI can cause lasting side effects for pediatric patients. One of these potential side effects includes oligospermia or even permanent azoospermia. Although many investigators have studied ways to shield the testicles during the TBI for nonmalignant conditions, there is no set standard. We describe the technical aspects of effective techniques to shield the testicles of male pediatric patients undergoing TBI. We verified that our techniques reduced the testicular dose by approximately 80%-85% of the TBI prescription dose in four male pediatric patients, keeping the dose well below the documented doses that can cause permanent infertility and hypogonadism.

Monte Carlo modeling of the Elekta Versa HD and patient dose calculation with EGSnrc/BEAMnrc.

INTRODUCTION: Numerous studies have proven the Monte Carlo method to be an accurate means of dose calculation. Although there are several commercial Monte Carlo treatment planning systems (TPSs), some clinics may not have access to these resources. We present a method for routine, independent patient dose calculations from treatment plans generated in a commercial TPS with our own Monte Carlo model using free, open-source software. MATERIALS AND METHODS: A model of the Elekta Versa HD linear accelerator was developed using the EGSnrc codes. A MATLAB script was created to take clinical patient plans and convert the DICOM RTP files into a format usable by EGSnrc. Ten patients' treatment plans were exported from the Monaco TPS to be recalculated using EGSnrc. Treatment simulations were done in BEAMnrc, and doses were calculated using Source 21 in DOSXYZnrc. Results were compared to patient plans calculated in the Monaco TPS and evaluated in Verisoft with a gamma criterion of 3%/2 mm. RESULTS: Our Monte Carlo model was validated within 1%/1-mm accuracy of measured percent depth doses and profiles. Gamma passing rates ranged from 82.1% to 99.8%, with 7 out of 10 plans having a gamma pass rate over 95%. Lung and prostate patients showed the best agreement with doses calculated in Monaco. All statistical uncertainties in DOSXYZnrc were less than 3.0%. CONCLUSION: A Monte Carlo model for routine patient dose calculation was successfully developed and tested. This model allows users to directly recalculate DICOM RP files containing patients' plans that have been exported from a commercial TPS.

Machine learning models to predict the delivered positions of Elekta multileaf collimator leaves for volumetric modulated arc therapy.

PURPOSE: Accurate positioning of multileaf collimator (MLC) leaves during volumetric modulated arc therapy (VMAT) is essential for accurate treatment delivery. We developed a linear regression, support vector machine, random forest, extreme gradient boosting (XGBoost), and an artificial neural network (ANN) for predicting the delivered leaf positions for VMAT plans. METHODS: For this study, 160 MLC log files from 80 VMAT plans were obtained from a single institution treated on 3 Elekta Versa HD linear accelerators. The gravity vector, X1 and X2 jaw positions, leaf gap, leaf position, leaf velocity, and leaf acceleration were extracted and used as model inputs. The models were trained using 70% of the log files and tested on the remaining 30%. Mean absolute error (MAE), root mean square error (RMSE), the coefficient of determination R2 , and fitted line plots showing the relationship between delivered and predicted leaf positions were used to evaluate model performance. RESULTS: The models achieved the following errors: linear regression (MAE = 0.158 mm, RMSE = 0.225 mm), support vector machine (MAE = 0.141 mm, RMSE = 0.199 mm), random forest (MAE = 0.161 mm, RMSE = 0.229 mm), XGBoost (MAE = 0.185 mm, RMSE = 0.273 mm), and ANN (MAE = 0.361 mm, RMSE = 0.521 mm). A significant correlation between a plan's gamma passing rate (GPR) and the prediction errors of linear regression, support vector machine, and random forest is seen (p < 0.045). CONCLUSIONS: We examined various models to predict the delivered MLC positions for VMAT plans treated with Elekta linacs. Linear regression, support vector machine, random forest, and XGBoost achieved lower errors than ANN. Models that can accurately predict the individual leaf positions during treatment can help identify leaves that are deviating from the planned position, which can improve a plan's GPR.

Effects of varying statistical uncertainty using a Monte Carlo based treatment planning system for VMAT.

PURPOSE: To determine the severity of the effects on VMAT dose calculations caused by varying statistical uncertainties (SU) per control point in a Monte Carlo based treatment planning system (TPS) and to assess the impact of the uncertainty during dose volume histogram (DVH) evaluation. METHODS: For this study, 13 archived patient plans were selected for recalculation. Treatment sites included prostate, lung, and head and neck. These plans were each recalculated five times with varying uncertainty levels using Elekta's Monaco Version 5.11.00 Monte Carlo Gold Standard XVMC dose calculation algorithm. The statistical uncertainty per control point ranged from 2 to 10% at intervals of 2%, while the grid spacing was set at 3 mm for all calculations. Indices defined by the RTOG describing conformity, coverage, and homogeneity were recorded for each recalculation. RESULTS: For all indices tested, one-way ANOVA tests failed to reject the null hypothesis that there is no significant difference between SU levels (p>0.05). Using the Bland-Altman analysis method, it was determined that we can expect the indices (with the exception of CIRTOG) to be within 1% of the lowest uncertainty calculation when calculating at 4% SU per control point. Beyond that, we can expect the indices to be within 3% of the lowest uncertainty calculation. CONCLUSION: Increasing the SU per control point exponentially decreased the amount of time required for dose calculations, while creating minimal observable differences in DVHs and isodose lines.

A customizable aluminum compensator system for total body irradiation.

PURPOSE: To develop a simplified aluminum compensator system for total body irradiation (TBI) that is easy to assemble and modify in a short period of time for customized patient treatments. METHODS: The compensator is composed of a combination of 0.3 cm thick aluminum bars, two aluminum T-tracks, spacers, and metal bolts. The system is mounted onto a plexiglass block tray. The design consists of 11 fixed sectors spanning from the patient's head to feet. The outermost sectors utilize 7.6 cm wide aluminum bars, while the remaining sectors use 2.5 cm wide aluminum bars. There is a magnification factor of 5 from the compensator to the patient treatment plane. Each bar of aluminum is interconnected at each adjacent sector with a tongue and groove arrangement and fastened to the T-track using a metal washer, bolt, and nut. Inter-bar leakage of the compensator was tested using a water tank and diode. End-to-end measurements were performed with an ion chamber in a solid water phantom and also with a RANDO phantom using internal and external optically stimulated luminescent detectors (OSLDs). In-vivo patient measurements from the first 20 patients treated with this aluminum compensator were compared to those from 20 patients treated with our previously used lead compensator system. RESULTS: The compensator assembly time was reduced to 20-30 min compared to the 2-4 h it would take with the previous lead design. All end-to-end measurements were within 10% of that expected. The median absolute in-vivo error for the aluminum compensator was 3.7%, with 93.8% of measurements being within 10% of that expected. The median error for the lead compensator system was 5.3%, with 85.1% being within 10% of that expected. CONCLUSION: This design has become the standard compensator at our clinic. It allows for quick assembly and customization along with meeting the Task Group 29 recommendations for dose uniformity.

Latent space arc therapy optimization.

Volumetric modulated arc therapy planning is a challenging problem in high-dimensional, non-convex optimization. Traditionally, heuristics such as fluence-map-optimization-informed segment initialization use locally optimal solutions to begin the search of the full arc therapy plan space from a reasonable starting point. These routines facilitate arc therapy optimization such that clinically satisfactory radiation treatment plans can be created in a reasonable time frame. However, current optimization algorithms favor solutions near their initialization point and are slower than necessary due to plan overparameterization. In this work, arc therapy overparameterization is addressed by reducing the effective dimension of treatment plans with unsupervised deep learning. An optimization engine is then built based on low-dimensional arc representations which facilitates faster planning times.

Variability of Breast Surface Positioning Using an Active Breathing Coordinator for a Deep Inspiration Breath Hold Technique.

Purpose The Elekta Active Breathing CoordinatorTM (ABC) is used to control breathing and guide deep inspiration breath hold (DIBH). It has been shown to be accurate in lung cancers, but limited analysis has been performed on the spatial accuracy and reproducibility of the breast surface. The use of optical surface-image guidance for patient positioning has grown in popularity and is an alternative solution for breast DIBH. This study aims to evaluate the breast surface variability of an ABC-guided DIBH by using a three-dimensional (3D) surface imaging system to record surface position. Methods Ten participants were placed in the treatment position, and breathing baselines and inhalation volume threshold baselines were monitored and recorded using the ABC. Over 60 minutes, the breathing patterns were recorded by the ABC and CatalystHDTM (C-RAD, Uppsala, Sweden). For each breath hold, the valve of the ABC closed at the baseline inhalation threshold and a 3D surface image was acquired. For each point on the baseline breast surface, a 3D vector was calculated to the subsequent breath hold surface as well as a root mean square (RMS) vector magnitude for the entire surface. Results The average and standard deviation for the RMS difference between the baseline and subsequent evaluated images were 7.12 ± 2.70 mm. Conclusion This study shows that while the ABC-guided inhalation volume is kept constant, a non-negligible variability of the breast surface position exists. Special considerations should be used in clinical situations, where the positioning of the surface is considered more important than inhalation volume.

A sensitivity analysis of probability maps in deep-learning-based anatomical segmentation.

PURPOSE: Deep-learning-based segmentation models implicitly learn to predict the presence of a structure based on its overall prominence in the training dataset. This phenomenon is observed and accounted for in deep-learning applications such as natural language processing but is often neglected in segmentation literature. The purpose of this work is to demonstrate the significance of class imbalance in deep-learning-based segmentation and recommend tuning of the neural network optimization objective. METHODS: An architecture and training procedure were chosen to represent common models in anatomical segmentation. A family of 5-block 2D U-Nets were independently trained to segment 10 structures from the Cancer Imaging Archive's Head-Neck-Radiomics-HN1 dataset. We identify the optimal threshold for our models according to their Dice score on the validation datasets and consider perturbations about the optimum. A measure of structure prominence in segmentation datasets is defined, and its impact on the optimal threshold is analyzed. Finally, we consider the use of a 2D Dice objective in addition to binary cross entropy. RESULTS: We observe significant decreases in perceived model performance with conventional 0.5-thresholding. Perturbations of as little as ±0.05 about the optimum threshold induce a median reduction in Dice score of 11.8% for our models. There is statistical evidence to suggest a weak correlation between training dataset prominence and optimal threshold (Pearson r = 0.92 and p ≈ 10 - 4 ). We find that network optimization with respect to the 2D Dice score itself significantly reduces variability due to thresholding but does not unequivocally create the best segmentation models when assessed with distance-based segmentation metrics. CONCLUSION: Our results suggest that those practicing deep-learning-based contouring should consider their postprocessing procedures as a potential avenue for improved performance. For intensity-based postprocessing, we recommend a mixed objective function consisting of the traditional binary cross entropy along with the 2D Dice score.

On the evaluation of mobile target trajectory between four-dimensional computer tomography and four-dimensional cone-beam computer tomography.

PURPOSE: For mobile lung tumors, four-dimensional computer tomography (4D CT) is often used for simulation and treatment planning. Localization accuracy remains a challenge in lung stereotactic body radiation therapy (SBRT) treatments. An attractive image guidance method to increase localization accuracy is 4D cone-beam CT (CBCT) as it allows for visualization of tumor motion with reduced motion artifacts. However, acquisition and reconstruction of 4D CBCT differ from that of 4D CT. This study evaluates the discrepancies between the reconstructed motion of 4D CBCT and 4D CT imaging over a wide range of sine target motion parameters and patient waveforms. METHODS: A thorax motion phantom was used to examine 24 sine motions with varying amplitudes and cycle times and seven patient waveforms. Each programmed motion was imaged using 4D CT and 4D CBCT. The images were processed to auto segment the target. For sine motion, the target centroid at each phase was fitted to a sinusoidal curve to evaluate equivalence in amplitude between the two imaging modalities. The patient waveform motion was evaluated based on the average 4D data sets. RESULTS: The mean difference and root-mean-square-error between the two modalities for sine motion were -0.35 ± 0.22 and 0.60 mm, respectively, with 4D CBCT slightly overestimating amplitude compared with 4D CT. The two imaging methods were determined to be significantly equivalent within ±1 mm based on two one-sided t tests (p < 0.001). For patient-specific motion, the mean difference was 1.5 ± 2.1 (0.8 ± 0.6 without outlier), 0.4 ± 0.3, and 0.8 ± 0.6 mm for superior/inferior (SI), anterior/posterior (AP), and left/right (LR), respectively. CONCLUSION: In cases where 4D CT is used to image mobile tumors, 4D CBCT is an attractive localization method due to its assessment of motion with respect to 4D CT, particularly for lung SBRT treatments where accuracy is paramount.

Managing tumor changes during radiotherapy using a deep learning model.

PURPOSE: We propose a treatment planning framework that accounts for weekly lung tumor shrinkage using cone beam computed tomography (CBCT) images with a deep learning-based model. METHODS: Sixteen patients with non-small-cell lung cancer (NSCLC) were selected with one planning CT and six weekly CBCTs each. A deep learning-based model was applied to predict the weekly deformation of the primary tumor based on the spatial and temporal features extracted from previous weekly CBCTs. Starting from Week 3, the tumor contour at Week N was predicted by the model based on the input from all the previous weeks (1, 2 … N - 1), and was evaluated against the manually contoured tumor using Dice coefficient (DSC), precision, average surface distance (ASD), and Hausdorff distance (HD). Information about the predicted tumor was then entered into the treatment planning system and the plan was re-optimized every week. The objectives were to maximize the dose coverage in the target region while minimizing the toxicity to the surrounding healthy tissue. Dosimetric evaluation of the target and organs at risk (heart, lung, esophagus, and spinal cord) was performed on four cases, comparing between a conventional plan (ignoring tumor shrinkage) and the shrinkage-based plan. RESULTS: he primary tumor volumes decreased on average by 38% ± 26% during six weeks of treatment. DSCs and ASD between the predicted tumor and the actual tumor for Weeks 3, 4, 5, 6 were 0.81, 0.82, 0.79, 0.78 and 1.49, 1.59, 1.92, 2.12 mm, respectively, which were significantly superior to the score of 0.70, 0.68, 0.66, 0.63 and 2.81, 3.22, 3.69, 3.63 mm between the rigidly transferred tumors ignoring shrinkage and the actual tumor. While target coverage metrics were maintained for the re-optimized plans, lung mean dose dropped by 2.85, 0.46, 2.39, and 1.48 Gy for four sample cases when compared to the original plan. Doses in other organs such as esophagus were also reduced for some cases. CONCLUSION: We developed a deep learning-based model for tumor shrinkage prediction. This model used CBCTs and contours from previous weeks as input and produced reasonable tumor contours with a high prediction accuracy (DSC, precision, HD, and ASD). The proposed framework maintained target coverage while reducing dose in the lungs and esophagus.

Quantifying false positional corrections due to facial motion using SGRT with open-face Masks.

PURPOSE: Studies have evaluated the viability of using open-face masks as an immobilization technique to treat intracranial and head and neck cancers. This method offers less stress to the patient with comparable accuracy to closed-face masks. Open-face masks permit implementation of surface guided radiation therapy (SGRT) to assist in positioning and motion management. Research suggests that changes in patient facial expressions may influence the SGRT system to generate false positional corrections. This study aims to quantify these errors produced by the SGRT system due to face motion. METHODS: Ten human subjects were immobilized using open-face masks. Four discrete SGRT regions of interest (ROIs) were analyzed based on anatomical features to simulate different mask openings. The largest ROI was lateral to the cheeks, superior to the eyebrows, and inferior to the mouth. The smallest ROI included only the eyes and bridge of the nose. Subjects were asked to open and close their eyes and simulate fear and annoyance and peak isocenter shifts were recorded. This was performed in both standard and SRS specific resolutions with the C-RAD Catalyst HD system. RESULTS: All four ROIs analyzed in SRS and Standard resolutions demonstrated an average deviation of 0.3 ± 0.3 mm for eyes closed and 0.4 ± 0.4 mm shift for eyes open, and 0.3 ± 0.3 mm for eyes closed and 0.8 ± 0.9 mm shift for eyes open. The average deviation observed due to changing facial expressions was 1.4 ± 0.9 mm for SRS specific and 1.6 ± 1.6 mm for standard resolution. CONCLUSION: The SGRT system can generate false positional corrections for face motion and this is amplified at lower resolutions and smaller ROIs. These errors should be considered in the overall tolerances and treatment plan when using open-face masks with SGRT and may warrant additional radiographic imaging.

A detailed experimental and Monte Carlo analysis of gold nanoparticle dose enhancement using 6 MV and 18 MV external beam energies in a macroscopic scale.

Dose enhancement due to gold nanoparticles (GNPs) has been quantified experimentally and through Monte Carlo simulations for external beam radiation therapy energies of 6 and 18 MV. The highest enhancement was observed for the 18 MV beam at the highest GNP concentration tested, amounting to a DEF of 1.02. DEF is shown to increase with increasing concentration of gold and increasing energy in the megavoltage energy range. The largest difference in measured vs. simulated DEF across all data sets was 0.3%, showing good agreement.

DNA Dosimeter Measurement of Relative Biological Effectiveness for 160 kVp and 6 MV X Rays.

In this work, we developed a DNA dosimeter, consisting of 4-kb DNA strands attached to magnetic streptavidin beads and labeled with fluorescein, to detect double-strand breaks (DSBs). The purpose here was to evaluate whether the DNA dosimeter readings reflect the relative biological effects of 160 kVp and 6 MV X rays. AVarian 600 C/D linac (6 MV) and a Faxitron cabinet X-ray system (160 kVp), both calibrated using traceable methods, were used to deliver high- and low-energy photons, respectively, to DNA dosimeters and multiple cell lines (mNs-5, HT-22 and Daoy). The responses were fit versus dose, and were used to quantify the dose of low-energy photons that produced the same response as that of the high-energy photons, at doses of 3, 6 and 9 Gy. The equivalent doses were utilized to calculate the relative biological effectiveness (RBEDSB and RBEcell survival). Additionally, a neutral comet assay was performed to measure the amount of intracellular DNA DSB, and ultimately the RBEcomet assay. The results of this work showed 160-kVp photon RBE values and 95% confidence intervals of 1.12 ± 0.04 (mNS-5), 1.16 ± 0.06 (HT-22), 1.25 ± 0.09 (Daoy) and 1.21 ± 0.24 (DNA dosimeter) at 9 Gy and 1.32 ± 0.16 (comet assay) at 3 Gy. Within the current error, the DNA dosimeter measured RBEDSB values in agreement with the RBEcell survival and assay from the cell survival and comet assay RBEcomet measurements. These results suggest that the DNA dosimeter can measure the changes in the radiobiological effects from different energy photons.

Deep learning-based survival analysis for brain metastasis patients with the national cancer database.

PURPOSE: Prognostic indices such as the Brain Metastasis Graded Prognostic Assessment have been used in clinical settings to aid physicians and patients in determining an appropriate treatment regimen. These indices are derivative of traditional survival analysis techniques such as Cox proportional hazards (CPH) and recursive partitioning analysis (RPA). Previous studies have shown that by evaluating CPH risk with a nonlinear deep neural network, DeepSurv, patient survival can be modeled more accurately. In this work, we apply DeepSurv to a test case: breast cancer patients with brain metastases who have received stereotactic radiosurgery. METHODS: Survival times, censorship status, and 27 covariates including age, staging information, and hormone receptor status were provided for 1673 patients by the NCDB. Monte Carlo cross-validation with 50 samples of 1400 patients was used to train and validate the DeepSurv, CPH, and RPA models independently. DeepSurv was implemented with L2 regularization, batch normalization, dropout, Nesterov momentum, and learning rate decay. RPA was implemented as a random survival forest (RSF). Concordance indices of test sets of 140 patients were used for each sample to assess the generalizable predictive capacity of each model. RESULTS: Following hyperparameter tuning, DeepSurv was trained at 32 min per sample on a 1.33 GHz quad-core CPU. Test set concordance indices of 0.7488 ± 0.0049, 0.6251 ± 0.0047, and 0.7368 ± 0.0047, were found for DeepSurv, CPH, and RSF, respectively. A Tukey HSD test demonstrates a statistically significant difference between the mean concordance indices of the three models. CONCLUSION: Our results suggest that deep learning-based survival prediction can outperform traditional models, specifically in a case where an accurate prognosis is highly clinically relevant. We recommend that where appropriate data are available, deep learning-based prognostic indicators should be used to supplement classical statistics.

An open-source tool to visualize potential cone collisions while planning SRS cases.

PURPOSE: To create an open-source visualization program that allows one to find potential cone collisions while planning intracranial stereotactic radiosurgery cases. METHODS: Measurements of physical components in the treatment room (gantry, cone, table, localization stereotactic radiation surgery frame, etc.) were incorporated into a script in MATLAB (MathWorks, Natick, MA) that produces three-dimensional visualizations of the components. A localization frame, used during simulation, fully contains the patient. This frame was used to represent a safety zone for collisions. Simple geometric objects are used to approximate the simulated components. The couch is represented as boxes, the gantry head and cone are represented by cylinders, and the patient safety zone can be represented by either a box or ellipsoid. These objects are translated and rotated based upon the beam geometry and the treatment isocenter to mimic treatment. A simple graphical user interface (GUI) was made in MATLAB (compatible with GNU Octave) to allow users to pass the treatment isocenter location, the initial and terminal gantry angles, the couch angle, and the number of angular points to visualize between the initial and terminal gantry angle. RESULTS: The GUI provides a fast and simple way to discover collisions in the treatment room before the treatment plan is completed. Twenty patient arcs were used as an end-to-end validation of the system. Seventeen of these appeared the same in the software as in the room. Three of the arcs appeared closer in the software than in the room. This is due to the treatment couch having rounded corners, whereas the software visualizes sharp corners. CONCLUSIONS: This simple GUI can be used to find the best orientation of beams for each patient. By finding collisions before a plan is being simulated in the treatment room, a user can save time due to replanning of cases.

Patient-specific dose quality assurance of single-isocenter multiple brain metastasis stereotactic radiosurgery using PTW Octavius 4D.

PURPOSE: Single-isocenter multiple brain metastasis stereotactic radiosurgery is an efficient treatment modality increasing in clinical practice. The need to provide accurate, patient-specific quality assurance (QA) for these plans is met by several options. This study reviews some of these options and explores the use of the Octavius 4D as a solution for patient-specific plan quality assurance. METHODS: The Octavius 4D Modular Phantom (O4D) with the 1000 SRS array was evaluated in this study. The array consists of 977 liquid-filled ion chambers. The center 5.5 cm × 5.5 cm area has a detector spacing of 2.5 mm. The ability of the O4D to reconstruct three-dimensional (3D) dose was validated against a 3D gel dosimeter, ion chamber, and film measurements. After validation, 15 patients with 2-11 targets had their plans delivered to the phantom. The criteria used for the gamma calculation was 3%/1 mm. The portion of targets which were measurable by the phantom was countable. The accompanying software compiled the measured doses allowing each target to be counted from the measured dose distribution. RESULTS: Spatial resolution was sufficient to verify the high dose distributions characteristic of SRS. Amongst the 15 patients there were 74 targets. Of the 74 targets, 61 (82%) of them were visible on the measured dose distribution. The average gamma passing rate was 99.3% (with sample standard deviation of 0.68%). CONCLUSIONS: The high resolution provided by the O4D with 1000 SRS board insert allows for very high-resolution measurement. This high resolution in turn can allow for high gamma passing rates. The O4D with the 1000 SRS array is an acceptable method of performing quality assurance for single-isocenter multiple brain metastasis SRS.