Training and Validation of Deep Learning-Based Auto-Segmentation Models for Lung Stereotactic Ablative Radiotherapy Using Retrospective Radiotherapy Planning Contours.

PURPOSE: Deep learning-based auto-segmented contour (DC) models require high quality data for their development, and previous studies have typically used prospectively produced contours, which can be resource intensive and time consuming to obtain. The aim of this study was to investigate the feasibility of using retrospective peer-reviewed radiotherapy planning contours in the training and evaluation of DC models for lung stereotactic ablative radiotherapy (SABR). METHODS: Using commercial deep learning-based auto-segmentation software, DC models for lung SABR organs at risk (OAR) and gross tumor volume (GTV) were trained using a deep convolutional neural network and a median of 105 contours per structure model obtained from 160 publicly available CT scans and 50 peer-reviewed SABR planning 4D-CT scans from center A. DCs were generated for 50 additional planning CT scans from center A and 50 from center B, and compared with the clinical contours (CC) using the Dice Similarity Coefficient (DSC) and 95% Hausdorff distance (HD). RESULTS: Comparing DCs to CCs, the mean DSC and 95% HD were 0.93 and 2.85mm for aorta, 0.81 and 3.32mm for esophagus, 0.95 and 5.09mm for heart, 0.98 and 2.99mm for bilateral lung, 0.52 and 7.08mm for bilateral brachial plexus, 0.82 and 4.23mm for proximal bronchial tree, 0.90 and 1.62mm for spinal cord, 0.91 and 2.27mm for trachea, and 0.71 and 5.23mm for GTV. DC to CC comparisons of center A and center B were similar for all OAR structures. CONCLUSIONS: The DCs developed with retrospective peer-reviewed treatment contours approximated CCs for the majority of OARs, including on an external dataset. DCs for structures with more variability tended to be less accurate and likely require using a larger number of training cases or novel training approaches to improve performance. Developing DC models from existing radiotherapy planning contours appears feasible and warrants further clinical workflow testing.

Technical Note: Cardiac synchronized volumetric modulated arc therapy for stereotactic arrhythmia radioablation - Proof of principle.

PURPOSE: Ventricular tachycardia (VT) is a rapid, abnormal heart rhythm that can lead to sudden cardiac death. Current treatment options include antiarrhythmic drug therapy and catheter ablation, both of which have only modest efficacy and have potential complications. Cardiac radiosurgery has the potential to be a noninvasive and efficient treatment option for VT. Cardiac motion, however, must be accounted for to ensure accurate dose delivery to the target region. Cardiac synchronized volumetric modulated arc therapy (CSVMAT) aims to minimize the dose delivered to normal tissues by synchronizing beam delivery with a cardiac signal, irradiating only during the quiescent intervals of the cardiac cycle (when heart motion is minimal) and adjusting the beam delivery speed in response to heart rate changes. METHODS: A CSVMAT plan was adapted from a conventional VMAT plan and delivered on a Varian TrueBeam linear accelerator. The original VMAT plan was divided into three interleaved CSVMAT phases, each consisting of alternating beam-on and beam-off segments synchronized to a sample heart rate. Trajectory log files were collected for the original VMAT and CSVMAT deliveries and the dose distributions were measured with Gafchromic EBT-XD film. RESULTS: Analysis of the trajectory log files showed successful synchronization with the sample cardiac signal. Film analysis comparing the original VMAT and CSVMAT dose distributions returned a gamma passing rate of 99.14% (2%/2 mm tolerance). CONCLUSIONS: The film results indicated excellent agreement between the dose distributions of the original and cardiac synchronized beam deliveries. This study demonstrates a proof of principle cardiac synchronization strategy for precise radiation treatment plan delivery and adjustment to a variable heart rate. The cardiac synchronized technique may be advantageous in radioablation for VT.

Monte Carlo validation of the TrueBeam 10XFFF phase-space files for applications in lung SABR.

PURPOSE: To establish the clinical acceptability of universal Monte Carlo phase-space data for the 10XFFF (flattening filter free) photon beam on the Varian TrueBeam Linac, including previously unreported data for small fields, output factors, and inhomogeneous media. The study was particularly aimed at confirming the suitability for use in simulations of lung stereotactic ablative radiotherapy treatment plans. METHODS: Monte Carlo calculated percent depth doses (PDDs), transverse profiles, and output factors for the TrueBeam 10 MV FFF beam using generic phase-space data that have been released by the Varian MC research team were compared with in-house measurements and published data from multiple institutions (ten Linacs from eight different institutions). BEAMnrc was used to create field size specific phase-spaces located underneath the jaws. Doses were calculated with DOSXYZnrc in a water phantom for fields ranging from 1 × 1 to 40 × 40 cm(2). Particular attention was paid to small fields (down to 1 × 1 cm(2)) and dose per pulse effects on dosimeter response for high dose rate 10XFFF beams. Ion chamber measurements were corrected for changes in ion collection efficiency (P(ion)) with increasing dose per pulse. MC and ECLIPSE ANISOTROPIC ANALYTICAL ALGORITHM (AAA) calculated PDDs were compared to Gafchromic film measurement in inhomogeneous media (water, bone, lung). RESULTS: Measured data from all machines agreed with Monte Carlo simulations within 1.0% and 1.5% for PDDs and in-field transverse profiles, respectively, for field sizes >1 × 1 cm(2) in a homogeneous water phantom. Agreements in the 80%-20% penumbra widths were better than 2 mm for all the fields that were compared. For all the field sizes considered, the agreement between their measured and calculated output factors was within 1.1%. Monte Carlo results for dose to water at water/bone, bone/lung, and lung/water interfaces as well as within lung agree with film measurements to within 2.8% for 10 × 10 and 3 × 3 cm(2) field sizes. This represents a significant improvement over the performance of the ECLIPSE AAA. CONCLUSIONS: The 10XFFF phase-space data offered by the Varian Monte Carlo research team have been validated for clinical use using measured, interinstitutional beam data in water and with film dosimetry in inhomogeneous media.

Monte Carlo modeling of HD120 multileaf collimator on Varian TrueBeam linear accelerator for verification of 6X and 6X FFF VMAT SABR treatment plans.

A Monte Carlo (MC) validation of the vendor-supplied Varian TrueBeam 6 MV flattened (6X) phase-space file and the first implementation of the Siebers-Keall MC MLC model as applied to the HD120 MLC (for 6X flat and 6X flattening filter-free (6X FFF) beams) are described. The MC model is validated in the context of VMAT patient-specific quality assurance. The Monte Carlo commissioning process involves: 1) validating the calculated open-field percentage depth doses (PDDs), profiles, and output factors (OF), 2) adapting the Siebers-Keall MLC model to match the new HD120-MLC geometry and material composition, 3) determining the absolute dose conversion factor for the MC calculation, and 4) validating this entire linac/MLC in the context of dose calculation verification for clinical VMAT plans. MC PDDs for the 6X beams agree with the measured data to within 2.0% for field sizes ranging from 2 × 2 to 40 × 40 cm2. Measured and MC profiles show agreement in the 50% field width and the 80%-20% penumbra region to within 1.3 mm for all square field sizes. MC OFs for the 2 to 40 cm2 square fields agree with measurement to within 1.6%. Verification of VMAT SABR lung, liver, and vertebra plans demonstrate that measured and MC ion chamber doses agree within 0.6% for the 6X beam and within 2.0% for the 6X FFF beam. A 3D gamma factor analysis demonstrates that for the 6X beam, > 99% of voxels meet the pass criteria (3%/3 mm). For the 6X FFF beam, > 94% of voxels meet this criteria. The TrueBeam accelerator delivering 6X and 6X FFF beams with the HD120 MLC can be modeled in Monte Carlo to provide an independent 3D dose calculation for clinical VMAT plans. This quality assurance tool has been used clinically to verify over 140 6X and 16 6X FFF TrueBeam treatment plans.

A Varian DynaLog file-based procedure for patient dose-volume histogram-based IMRT QA.

In the present study, we describe a method based on the analysis of the dynamic MLC log files (DynaLog) generated by the controller of a Varian linear accelerator in order to perform patient-specific IMRT QA. The DynaLog files of a Varian Millennium MLC, recorded during an IMRT treatment, can be processed using a MATLAB-based code in order to generate the actual fluence for each beam and so recalculate the actual patient dose distribution using the Eclipse treatment planning system. The accuracy of the DynaLog-based dose reconstruction procedure was assessed by introducing ten intended errors to perturb the fluence of the beams of a reference plan such that ten subsequent erroneous plans were generated. In-phantom measurements with an ionization chamber (ion chamber) and planar dose measurements using an EPID system were performed to investigate the correlation between the measured dose changes and the expected ones detected by the reconstructed plans for the ten intended erroneous cases. Moreover, the method was applied to 20 cases of clinical plans for different locations (prostate, lung, breast, and head and neck). A dose-volume histogram (DVH) metric was used to evaluate the impact of the delivery errors in terms of dose to the patient. The ionometric measurements revealed a significant positive correlation (R² = 0.9993) between the variations of the dose induced in the erroneous plans with respect to the reference plan and the corresponding changes indicated by the DynaLog-based reconstructed plans. The EPID measurements showed that the accuracy of the DynaLog-based method to reconstruct the beam fluence was comparable with the dosimetric resolution of the portal dosimetry used in this work (3%/3 mm). The DynaLog-based reconstruction method described in this study is a suitable tool to perform a patient-specific IMRT QA. This method allows us to perform patient-specific IMRT QA by evaluating the result based on the DVH metric of the planning CT image (patient DVH-based IMRT QA).

A Monte Carlo approach to validation of FFF VMAT treatment plans for the TrueBeam linac.

PURPOSE: To commission and benchmark a vendor-supplied (Varian Medical Systems) Monte Carlo phase-space data for the 6 MV flattening filter free (FFF) energy mode on a TrueBeam linear accelerator for the purpose of quality assurance of clinical volumetric modulated arc therapy (VMAT) treatment plans. A method for rendering the phase-space data compatible with BEAMnrc/DOSXYZnrc simulation software package is presented. METHODS: Monte Carlo (MC) simulations were performed to benchmark the TrueBeam 6 MV FFF phase space data that have been released by the Varian MC Research team. The simulations to benchmark the phase space data were done in three steps. First, the original phase space which was created on a cylindrical surface was converted into a format that was compatible with BEAMnrc. Second, BEAMnrc was used to create field size specific phase spaces located underneath the jaws. Third, doses were calculated with DOSXYZnrc in a water phantom for fields ranging from 1 × 1 to 40 × 40 cm(2). Calculated percent depth doses (PDD), transverse profiles, and output factors were compared with measurements for all the fields simulated. After completing the benchmarking study, three stereotactic body radiotherapy (SBRT) VMAT plans created with the Eclipse treatment planning system (TPS) were calculated with Monte Carlo. Ion chamber and film measurements were also performed on these plans. 3D gamma analysis was used to compare Monte Carlo calculation with TPS calculations and with film measurement. RESULTS: For the benchmarking study, MC calculated and measured values agreed within 1% and 1.5% for PDDs and in-field transverse profiles, respectively, for field sizes >1 × 1 cm(2). Agreements in the 80%-20% penumbra widths were better than 2 mm for all the fields that were compared. With the exception of the 1 × 1 cm(2) field, the agreement between measured and calculated output factors was within 1%. It is of note that excellent agreement in output factors for all field sizes including highly asymmetric fields was achieved without accounting for backscatter into the beam monitor chamber. For the SBRT VMAT plans, the agreement between Monte Carlo and ion chamber point dose measurements was within 1%. Excellent agreement between Monte Carlo, treatment planning system and Gafchromic film dose distribution was observed with over 99% of the points in the high dose volume passing the 3%, 3 mm gamma test. CONCLUSIONS: The authors have presented a method for making the Varian IAEA compliant 6 MV FFF phase space file of the TrueBeam linac compatible with BEAMnrc/DOSXYZnrc. After benchmarking the modified phase space against measurement, they have demonstrated its potential for use in MC based quality assurance of complex delivery techniques.

Evaluation of the AAA Treatment Planning Algorithm for SBRT Lung Treatment: Comparison with Monte Carlo and Homogeneous Pencil Beam Dose Calculations.

PURPOSE: To evaluate the dose calculation accuracy of the Varian Eclipse anisotropic analytical algorithm (AAA) for stereotactic body radiation therapy (SBRT), and to investigate the dosimetric consequences of not applying tissue heterogeneity correction on complex SBRT lung plans. MATERIALS AND METHODS: Nine cases of non-small-cell lung cancer (NSCLC) that were previously treated with SBRT at our center were selected for this study. Following Radiation Therapy Oncology Group 0236, the original plans were calculated using pencil beam without heterogeneity correction (PBNC). For this study, these plans were recalculated by applying tissue heterogeneity correction with the AAA algorithm and with the Monte Carlo (MC) method, keeping the number of monitor units the same as the original plans. Two kinds of plan comparison were made. First, the AAA calculations were compared with MC. Second, the treatment plans that were calculated with AAA were compared with the original PBNC calculations. The following dose-volume parameters were used for the comparison: V100%; V90%; the maximum, the minimum, and the mean planning target volume (PTV) doses (Dmax, Dmin, and Dmean, respectively); V20Gy, V15Gy, V10Gy, V5Gy; Dmean for the lung; and Dmax for the critical organs. RESULTS: Comparable results were obtained for AAA and MC calculations: except for Dmax, Dmin, and Dmean, the differences in the patient-average values of all of the PTV dose parameters were less than 2%. The largest average difference was observed for Dmin (3.8 ± 5.4%). Average differences in all the lung dose parameters were under 0.2%, and average differences in normal tissue Dmax were under 0.3 Gy, except for the skin dose. There were appreciable differences in the PTV and normal tissue dose-volume parameters when comparing AAA and PBNC calculations. Except for V100% and V90%, PBNC calculations on average underestimated the dose to the PTV. The largest discrepancy was in the PTV maximum dose, with a patient-averaged difference of 11.1 ± 4.6%. CONCLUSIONS: Based on our MC investigation, we conclude that the Eclipse AAA algorithm is sufficiently accurate for dose calculations of lung SBRT plans involving small 6-MV photon fields. Our results also demonstrate that, although dose calculations at the periphery of the PTV showed good agreement when comparing PBNC with both AAA and MC calculations, there is a potential to significantly underestimate the dose inside the PTV and doses to critical structures if tissue heterogeneity correction is not applied to lung SBRT plans.

Monte Carlo based, patient-specific RapidArc QA using Linac log files.

PURPOSE: A Monte Carlo (MC) based QA process to validate the dynamic beam delivery accuracy for Varian RapidArc (Varian Medical Systems, Palo Alto, CA) using Linac delivery log files (DynaLog) is presented. Using DynaLog file analysis and MC simulations, the goal of this article is to (a) confirm that adequate sampling is used in the RapidArc optimization algorithm (177 static gantry angles) and (b) to assess the physical machine performance [gantry angle and monitor unit (MU) delivery accuracy]. METHODS: Ten clinically acceptable RapidArc treatment plans were generated for various tumor sites and delivered to a water-equivalent cylindrical phantom on the treatment unit. Three Monte Carlo simulations were performed to calculate dose to the CT phantom image set: (a) One using a series of static gantry angles defined by 177 control points with treatment planning system (TPS) MLC control files (planning files), (b) one using continuous gantry rotation with TPS generated MLC control files, and (c) one using continuous gantry rotation with actual Linac delivery log files. Monte Carlo simulated dose distributions are compared to both ionization chamber point measurements and with RapidArc TPS calculated doses. The 3D dose distributions were compared using a 3D gamma-factor analysis, employing a 3%/3 mm distance-to-agreement criterion. RESULTS: The dose difference between MC simulations, TPS, and ionization chamber point measurements was less than 2.1%. For all plans, the MC calculated 3D dose distributions agreed well with the TPS calculated doses (gamma-factor values were less than 1 for more than 95% of the points considered). Machine performance QA was supplemented with an extensive DynaLog file analysis. A DynaLog file analysis showed that leaf position errors were less than 1 mm for 94% of the time and there were no leaf errors greater than 2.5 mm. The mean standard deviation in MU and gantry angle were 0.052 MU and 0.355 degrees, respectively, for the ten cases analyzed. CONCLUSIONS: The accuracy and flexibility of the Monte Carlo based RapidArc QA system were demonstrated. Good machine performance and accurate dose distribution delivery of RapidArc plans were observed. The sampling used in the TPS optimization algorithm was found to be adequate.