S Lim

IMRT QA using machine learning: A multi‐institutional validation

Abstract Purpose To validate a machine learning approach to Virtual intensity‐modulated radiation therapy (IMRT) quality assurance (QA) for accurately predicting gamma passing rates using different measurement approaches at different institutions. Methods A Virtual IMRT QA framework was previously developed using a machine learning algorithm based on 498 IMRT plans, in which QA measurements were performed using diode‐array detectors and a 3%local/3 mm with 10% threshold at Institution 1. An independent set of 139 IMRT measurements from a different institution, Institution 2, with QA data based on portal dosimetry using the same gamma index, was used to test the mathematical framework. Only pixels with ≥10% of the maximum calibrated units (CU) or dose were included in the comparison. Plans were characterized by 90 different complexity metrics. A weighted poison regression with Lasso regularization was trained to predict passing rates using the complexity metrics as input. Results The methodology predicted passing rates within 3% accuracy for all composite plans measured using diode‐array detectors at Institution 1, and within 3.5% for 120 of 139 plans using portal dosimetry measurements performed on a per‐beam basis at Institution 2. The remaining measurements (19) had large areas of low CU, where portal dosimetry has a larger disagreement with the calculated dose and as such, the failure was expected. These beams need further modeling in the treatment planning system to correct the under‐response in low‐dose regions. Important features selected by Lasso to predict gamma passing rates were as follows: complete irradiated area outline (CIAO), jaw position, fraction of MLC leafs with gaps smaller than 20 or 5 mm, fraction of area receiving less than 50% of the total CU, fraction of the area receiving dose from penumbra, weighted average irregularity factor, and duty cycle. Conclusions We have demonstrated that Virtual IMRT QA can predict passing rates using different measurement techniques and across multiple institutions. Prediction of QA passing rates can have profound implications on the current IMRT process.

TU‐FF‐A3‐04: An In Vivo Comparative Study of the MV and KV Cone Beam Computed Tomography Image Quality of a Lung Patient

Purpose: To compare image quality, reconstruction artifacts and tumorvisibility for kV and MV cone‐beam computed tomography(CBCT) scans reconstructed with the same algorithm. Method and Materials: A protocol lung‐cancer patient was set up in the identical treatment position for kV and MVCBCT using a Varian On‐Board ImagerCBCT and an inhouse MVCBCT imaging system. For both scans the gantry made a 1‐minute, 360° continuous rotation. For the MVCBCT, ∼460 projection images were acquired at 6MV for ∼13 MU; for kVCBCT ∼600 projections were acquired using 125 kVp, 80 mA and 25‐ms exposure time per projection, resulting in ∼2cGy at isocenter. Reconstruction was performed using the Feldkamp back projection algorithm. Both scans were registered to the treatment plan CT. The visibility of three selected regions (bronchus, vertebrae, heart) is compared using the corresponding signal‐to‐noise ratio (SNR). The contrast ratio (CR) and contrast‐to‐noise ratio (CNR) at the tumor are also compared for ease of tumor identification. Results: The SNR of bronchus, vertebrae and heart are 25, 34 and 33 respectively for MVCBCT while the corresponding values in kV scan are 17, 33 and 42. For tumor identifiability, CNR and CR are 11 and 2 respectively for MV scan, and 10 and 2 for kV scan. The CNR of the vertebrae in MV and kV cases are 2 and 6. Time to register the kV image is approximately 50% less than MV image. Similar breathing artifacts are present in both scans. Conclusions: Both kV and MV scans deliver usable images. The tumor can be discriminated from the lung background. Higher bone contrast in kV scan helps to reduce time required to register the scan with the planning CT.Conflict of Interest: Research sponsored by NCI Grant P01‐CA59017 and Varian Medical Systems; Research agreement with Varian Medical Systems.

SU‐E‐T‐534: Beam and MLC Commissioning and Assessment of Three Commercial Treatment Planning Systems

PURPOSE To assess and compare the open beam and multi-leaf collimator modeling of Pinnacle, Ecilpse (AAA and Acuros) and RayStation planning systems. METHOD AND MATERIALS The 6MV photon beam of a Varian TrueBeam with Millennium 120 MLC was used for this study. Measurements made with combinations of ion chamber, radiochromic film, and diodes in water and plastic phantoms. Depth and crossplane profiles of open square fields shaped by jaws or MLC ranged from 3×3 to 40×40cm2 and from 0 to 20 cm depth. Depth dose, flatness (80% of FWHM), and penumbra (20-80%) of calculated and measured profiles were compared. Various MLC test patterns were calculated and compared with measurements to assess the modeling of the round leaf edge, tongue-and-groove, and interleaf transmissions. RESULTS Calculated depth doses are within 1.0% and flatness is within 2% for all field sizes and depths. Jaw penumbrae are within 2mm and 3mm for 20×20 and 30×30cm2 at 10cm depth respectively. MLC penumbrae (20-80%) of the three systems are within 0.3mm and 1.0mm for a 3×3cm2 and 10×10cm2 MLC apertures. Notably, to match the measured MLC round-edge transmission, the half thickness (10% transmission) leaf-tip width of the current RayStation MLC model has to be broadened to 10mm. All three systems appear to adequately model the tongue-and-groove. Pinnacle explicitly models the interleaf transmission while Eclipse and RayStation simply use average MLC transmission. CONCLUSIONS All three systems are capable of generating clinically acceptable beam models for open fields. Based upon the round-edge profile, Eclipse and Pinnacle provide better MLC models than RayStation. Among the three systems, Eclipse took the least time and effort to commission these features.

MO‐FF‐A2‐04: An Accurate Mechanical Quality Assurance Procedure for a New High Performance Linac

Purpose T o establish QA procedures for the mechanical systems of a new linac that has been developed to deliver radiation, using image‐guidance, to the target with improved spatial accuracy. The procedures are required to be able to detect mechanical errors of much less than 1 mm, be independent of the linacs own readouts and calibration procedures, and be fast enough for the physicist to perform on a monthly or more frequent schedule. Method and Materials In image guided delivery, mechanical properties that will affect the spatial accuracy with which dose is delivered include: • radiation isocenter ‐imaging origin displacement, • position errors of the jaws and MLC leaves, • accuracy with which the patient support couch can respond to a change in position request. To measure these quantities, we use the machines kV, MV, and infra‐red imaging systems. We report on the techniques used, and the estimates of their accuracy. Results The preliminary estimate of the measurement uncertainty of the radiation isocenter — imaging origin offset is ± 0.3 mm. The observed offset is within the measurement error. The accuracy of the field size seen in the MV images is ± 0.2mm. Couch accuracy for shifts of up to 2 cm, the magnitudes expected using image guidance, was found to be within the measurement error. The ability to control the machine using scripts allows gantry and collimator positioning, couch positioning, beam delivery and imaging in all modes, to be sequenced and performed automatically. Thus the time required for a complete mechanical QA procedure is greatly shortened. Conclusion The imaging components of a new linac can be positioned with sufficient reproducibility and accuracy to allow their use in a mechanical QA program that can achieve the sub‐mm accuracy needed for this machine. Research supported by Varian Medical Systems

SU‐G‐BRC‐05: Conundrum for VMAT Cranial Multiple Lesions Treated with HD120 MLC

PURPOSE To commission a custom 6MV-SRS-AAA Eclipse beam model for VMAT multiple lesions cranial SRS treatment on a Varian TrueBeam STx. METHODS Six clinical plans were created using a customized beam model with dosimetric-leaf-gap(DLG) optimized for clinical treatments. Each plan had 4-6 non-isocentric targets with size from 0.2 to 7.1cc. All fields were measured with EBT3 film in the coronal plane in a solid water phantom and with an AS1000 EPID using gantry rotation. In addition, an end-to-end test was performed with coronal and sagittal films in an anthropomorphic phantom verifying dosimetry and localization accuracy. Portal dose distributions were generated with a custom portal dosimetry algorithm(PDIP). Measured dose distributions were compared with calculations using average dose difference (DD), and gamma function, γ. Using a 1.25mm grid, the γ criteria, local DD ≤ 3% and 2mm distance-to-agreement, were applied in regions with dose 50% of maximum. RESULTS The respective DD and γ for all films were <±2% and >94.2%. The portal dose γ scores for all the plans were >94.9%. However, local regions with underdose >10%, were observed when targets were treated with the 5mm leaves. The same plans re-optimized with two isocenters such that all lesions were under the 2.5mm leaves did not show this effect. The DD and localization error of the end-to-end test were within 3.4% and 1.0mm respectively. CONCLUSION The custom AAA beam model is capable of calculating acceptable dosimetry for targets using only the 2.5 mm leaves. This restricts lesions to within ±4cm of isocenter. The observed underdose beneath the 5mm leaves is attributed to a limitation in Eclipse that uses a single DLG representing the DLGs of both 2.5mm and 5mm leaves. If lesions are >4cm from isocenter, a multiple isocenter technique should be considered to allow the use of only the 2.5mm leaves.

SU‐C‐213CD‐02: The Use of Optically Stimulated Luminescent Dosimeters in a Cone Beam Quality Assurance Testing

Purpose: The number of cone beam techniques available on LINACs and the large variation of dose with position means that a large number of measurements would be required to perform the annual QA as recommended by TG142. We evaluate the accuracy of a more efficient method of measuring the imaging dose of cone beam scanning modes using optically stimulated luminescent (OSL) (Microdot, Landauer) dosimeters. Methods: Special inserts were constructed to hold OSL dosimeters at the 5 standard points of measurement (1 central, 4 peripheral) of a commercial CT dose phantom. The phantom was loaded with OSLs and scanned for each of four cone beam modes: two half‐fan modes (pelvis, low dose thorax) and two full fan modes (head, pelvis spot‐light). KVP and mAs settings were at default values, the head technique used 100KVP, the others used 125 KVP. All scans were then re‐measured using a 0.6cc ion chamber positioned at the same points. Ion chamber and OSL results were compared. A single calibration point, the measurement with the highest dose, was used to convert the readout to cGy. Results: Doses measured by the ion chamber varied from 0.17 cGy to 4.6 cGy. The correspondence between the PMT reading and the chamber reading (RADCAL 9096) was linear, the correlation coefficient being 0.996. The average absolute difference between the chamber and the OSL dose was 0.13 cGy. Conclusions: The use of OSLs was found to be of sufficient accuracy to permit their use for the annual QA recommended by TG 142. The number of scans required for a comprehensive set of dose measurements for 4 scans modes and five measurement points was reduced from 20 to 5, including a single calibration scan. This is a significant reduction in the time needed for TG 142 QA.

SU‐E‐T‐532: Comparison of Dose Distributions Calculated Using Different Planning Systems with Radiochromic Film Measurements in an Inhomogeneous Phantom

PURPOSE Accurate modeling of the dose distribution in a lung tumor is challenging for traditional dose calculation algorithms. We compare the dose distributions of four commercial dose calculation Methods: Raysearch (Raysearch Laboratories) and Pinnacle (Philips Healthcare) collapsed cone, and Eclipse AAA and Eclipse Acuros,(Varian Medical Systems) with measurements using radiochromic film in a lung tumor phantomMethods: A simple lung tumor phantom was constructed using a thermoplastic cylinder 29 mm diameter and 40mm in length (density 1.3 gm/cc) imbedded in cork phantom 25 × 25 × 20 cm of density 0.32 gm/cc. Nine film layers normal to the axis of the cylinder where placed between layers of cork, above, below and through the cylindrical inhomogeneity. The phantom was irradiated with a single asymmetric 10×10 cm 6 MV field with the central axis collinear with the cylinder axis. Thirteen film exposures at 5 cm depth taken with doses 0-10 Gy were used to calibrate the film. The phantom was CT scanned and the DICOM study loaded into each of the treatment planning systems to calculate the dose distribution in the phantom. RESULTS Away from cork-poly interfaces, agreement between the four algorithms was within 3% of the film measurements. For Acuros, the dose at the edge of the cylinder was found to be up to 2% lower than that at the center of the cylinder possibly because of the loss of lateral electron equilibrium. CONCLUSIONS All four algorithms achieved remarkable agreement with the radiochromic film measurement. The Acuros algorithm appeared to more accurately model the peripheral dose deficit in the tumor, although a more detailed study is required for confirmation.

SU-FF-T-79: Accurate and Efficient Monte Carlo Dose Calculation for Electron Beams

Purpose: To develop a Monte Carlo dose calculation engine for electron beams that is feasible for routine clinical treatment planning.Method and Materials: The dose calculation engine consists of a description of the clinical beams and a dose calculation module. A 12‐component multi‐source model was used to characterize the phase‐space of clinical beams. There are 6 components each for electrons and photons, corresponding to the 3 scrapers, x‐jaws, y‐jaws, and the direct component respectively. In addition, we have developed a method to account for the presence of an arbitrary shaped cutout by modifying the last component of the standard beam model. For the dose calculation module, implementation of the Super‐Monte Carlo method accelerates the calculation by using electron and photon tracks pre‐calculated in water to avoid the computationally intensive sampling processes. These tracks are replayed in the patient computermodel as defined by CT. To account for inhomogeneities, the step size and scattering angle were adjusted according to the CT voxel values and material indexes. The dose calculation engine was verified by comparing with film measurements in several different geometries. Results: The results agreed with film measurements to within 2–5% percent both in homogeneous and heterogeneous phantoms. Our method is faster than the analog Monte Carlo calculation by a factor of 6 to 10 and is comparable in performance to a commercial system. The modified beam models for arbitrary cutouts can be derived in a few seconds. The disk storage for pre‐calculated tracks is about 5.5 GB and 125 MB for the standard beam models.Conclusion: The developed Monte Carlo calculation engine is accurate and efficient. The disk space and computational time required are well within clinical acceptability. It is a highly promising dose calculation tool for routine clinical applications. This work supported in part by NCI grant P01‐CA59017.

Institutional experience with SRS VMAT planning for multiple cranial metastases

Abstract Background and Purpose This study summarizes the cranial stereotactic radiosurgery (SRS) volumetric modulated arc therapy (VMAT) procedure at our institution. Materials and Methods Volumetric modulated arc therapy plans were generated for 40 patients with 188 lesions (range 2–8, median 5) in Eclipse and treated on a TrueBeam STx. Limitations of the custom beam model outside the central 2.5 mm leaves necessitated more than one isocenter pending the spatial distribution of lesions. Two to nine arcs were used per isocenter. Conformity index (CI), gradient index (GI) and target dose heterogeneity index (HI) were determined for each lesion. Dose to critical structures and treatment times are reported. Results Lesion size ranged 0.05–17.74 cm3 (median 0.77 cm3), and total tumor volume per case ranged 1.09–26.95 cm3 (median 7.11 cm3). For each lesion, HI ranged 1.2–1.5 (median 1.3), CI ranged 1.0–2.9 (median 1.2), and GI ranged 2.5–8.4 (median 4.4). By correlating GI to PTV volume a predicted GI = 4/PTV0.2 was determined and implemented in a script in Eclipse and used for plan evaluation. Brain volume receiving 7 Gy (V 7 Gy) ranged 10–136 cm3 (median 42 cm3). Total treatment time ranged 24–138 min (median 61 min). Conclusions Volumetric modulated arc therapy provide plans with steep dose gradients around the targets and low dose to critical structures, and VMAT treatment is delivered in a shorter time than conventional methods using one isocenter per lesion. To further improve VMAT planning for multiple cranial metastases, better tools to shorten planning time are needed. The most significant improvement would come from better dose modeling in Eclipse, possibly by allowing for customizing the dynamic leaf gap (DLG) for a special SRS model and not limit to one DLG per energy per treatment machine and thereby remove the limitation on the Y‐jaw and allow planning with a single isocenter.

SU-F-T-226: QA Management for a Large Institution with Multiple Campuses for FMEA

PURPOSE To redesign our radiation therapy QA program with the goal to improve quality, efficiency, and consistency among a growing number of campuses at a large institution. METHODS A QA committee was established with at least one physicist representing each of our six campuses (22 linacs). Weekly meetings were scheduled to advise on and update current procedures, to review end-to-end and other test results, and to prepare composite reports for internal and external audits. QA procedures for treatment and imaging equipment were derived from TG Reports 142 and 66, practice guidelines, and feedback from ACR evaluations. The committee focused on reaching a consensus on a single QA program among all campuses using the same type of equipment and reference data. Since the recommendations for tolerances referenced to baseline data were subject to interpretation in some instances, the committee reviewed the characteristics of all machines and quantified any variations before choosing between treatment planning system (i.e. treatment planning system commissioning data that is representative for all machines) or machine-specific values (i.e. commissioning data of the individual machines) as baseline data. RESULTS The configured QA program will be followed strictly by all campuses. Inventory of available equipment has been compiled, and additional equipment acquisitions for the QA program are made as needed. Dosimetric characteristics are evaluated for all machines using the same methods to ensure consistency of beam data where possible. In most cases, baseline data refer to treatment planning system commissioning data but machine-specific values are used as reference where it is deemed appropriate. CONCLUSION With a uniform QA scheme, variations in QA procedures are kept to a minimum. With a centralized database, data collection and analysis are simplified. This program will facilitate uniformity in patient treatments and analysis of large amounts of QA data campus-wide, which will ultimately facilitate FMEA.