Epid‐based in vivo dose verification for lung stereotactic treatments delivered with multiple breath‐hold segmented volumetric modulated arc therapy

Abstract We evaluated an EPID‐based in‐vivo dosimetry (IVD) method for the dose verification and the treatment reproducibility of lung SBRT‐VMAT treatments in clinical routine. Ten patients with lung metastases treated with Elekta VMAT technique were enrolled. All patients were irradiated in five consecutive fractions, with total doses of 50 Gy. Set‐up was carried out with the Elekta stereotactic body frame. Eight patients were simulated and treated using the Active Breath Control (ABC) system, a spirometer enabling patients to maintain a breath‐hold at a predetermined lung volume. Two patients were simulated and treated in free‐breathing using an abdominal compressor. IVD was performed using the SOFTDISO software. IVD tests were evaluated by means of (a) ratio R between daily in‐vivo isocenter dose and planned dose and (b) γ‐analysis between EPID integral portal images in terms of percentage of points with γ‐value smaller than one (γ %) and mean γ‐values (γ mean) using a 3%(global)/3 mm criteria. Alert criteria of ±5% for R ratio, γ % < 90%, and γ mean > 0.67 were chosen. 50 transit EPID images were acquired. For the patients treated with ABC spirometer, the results reported a high level of accuracy in dose delivery with 100% of tests within ±5%. The γ‐analysis showed a mean value of γ mean equal to 0.21 (range: 0.04–0.56) and a mean γ % equal to 96.9 (range: 78–100). Relevant discrepancies were observed only for the two patients treated without ABC, mainly due to a blurring dose effect due to residual respiratory motion. Our method provided a fast and accurate procedure in clinical routine for verifying delivered dose as well as for detecting errors.


| INTRODUCTION
The technological advancements in immobilization and imaging, together with the ability to deliver high conformal doses and to account for organ motion have led to a widespread implementation of stereotactic body radiotherapy (SBRT) in a number of clinical settings. 1 Over the last few years, different new techniques including volumetric modulated arc therapy (VMAT) have been successfully applied to SBRT treatments, owing to high-dose conformity, improved sparing of healthy tissues and fast delivery time. [2][3][4][5] In SBRT treatments effective tumor motion control must be considered much more compelling because reduced margins are needed to avoid the irradiation of a large amount of normal tissue to high doses. Various methods have been developed to explicitly account for respiration motion in SBRT treatments, as respiratory gating techniques, breath-hold techniques, and forced shallow breathing techniques. 6 In particular, Active Breath Control (ABC) methods by means of spirometers have been used to actively hold the patient's breath at a certain level (e.g., at moderate deep inhalation) during the beam-on time, providing an accurate method to improve the localization of lung and liver tumors during SBRT delivery. 7,8 This strategy has been recently implemented for SBRT treatments of extracranial metastases, with the aim to increase the accuracy of treatment delivery and the intrafraction treatment reproducibility.
The integration of several complex techniques such as SBRT, VMAT, and breath-hold into a single therapeutic strategy requires a high-level of confidence in the accuracy of the entire treatment delivery process. The impact of treatment delivery errors represents a major concern for these complex techniques, suggesting a strong argument in favor of in-vivo dosimetry (IVD). 9 In particular, due to very large dose over a few number of fractions used in SBRT treatments, any error in treatment delivery can be much more detrimental as compared to conventional fractionated therapy, with a major risk of completely nullify the curative intent or produce serious damage to the patient. Among the different available dosimetric systems, amorphous silicon electronic portal imaging devices (aSi-EPID) have demonstrated unique favorable characteristics for IVD purposes (high two-dimensional resolution and fast image acquisition) and several algorithms were developed to reconstruct the dose within the patient in terms of point dose, 2D-dose distribution or 3D dose distribution. 10 At our institution, we developed a generalized procedure for the daily in-vivo isocenter dose reconstruction of radiation treatments, [11][12][13][14][15] leading to the foundation of an Italian national project financed by the Istituto Nazionale di Fisica Nucleare 16 and to the development of the SOFTDISO software (Best Medical Italy, Italy).
The analysed results for conformal radiotherapy reported clinically relevant differences between planned and delivered dose, detecting the presence of dose discrepancies in more than 10% of the tests with respect to our tolerance levels. 17 Recent publications have investigated the new challenges of epid-based IVD for more complex treatments as IMRT or VMAT, showing the feasibility and the sensitivity of this approach to detect dose discrepancies also for these complex techniques. [18][19][20] By routine clinical use of EPID-based IVD, major dosimetric discrepancies due to anatomical variations 21 and serious errors including plan transfer errors due to record-and-verify network failure were detected. 22 A large clinical experience has been recently carried out at the Radiotherapy Centre of the Fondazione Policlinico Universitario A. Gemelli in Rome by means of an automated Epid-based IVD procedure for more than 800 patients. The application of IVD procedure has allowed the authors to detect on average 6% of VMAT plans and 21% of 3D-CRT plans outside at least one of their tolerance levels. 23 Experiences with Epid-based IVD for SBRT treatments are even more rare, with very few publications showing the technical feasibility of this strategy. McCowan et al. have recently validated an inhouse physics-based model which utilizes EPID images to reconstruct the dose in patient during SBRT-VMAT treatments. 24 The authors reported satisfactory results for lung and spine cases, with pass rates better than 93% with 3%-3 mm γ-index tolerance level, showing the suitability of this approach for clinical implementation. We have recently applied our IVD algorithm, widely used for 3Dconformal, IMRT and VMAT, to lung SBRT treatments, with the aim to supply, in quasi real-time, both the isocenter dose and the γ-analysis of transit EPID images. In this paper, we presented our current experience on real patients with EPID-based IVD for the dose verification of lung SBRT treatments delivered with breath-hold multisegmented VMAT technique.

2.A | Simulation, active breath control and treatment planning
Ten patients with lung metastases treated with Elekta VMAT were enrolled. Patients set-up was carried out with the stereotactic body frame (SBF, Elekta, Crawley, UK), an immobilization device used to define target position by a stereotactic coordinates system instead of anatomical landmarks or skin markers, 28 with an attached "vacuum pillow" customized to each patient. For eight of the ten enrolled patients, the breathing control was performed using the ABC system (Elekta, Crawley, UK), a spirometer able to immobilize the respiratory motion repeatedly and reproducibly for a period of time that can be comfortably tolerated by the patient. 7,8 Before final CT acquisition for planning, patients underwent a 3-day training to assess their comfort and compliance, the lung capacity and the optimal breath hold length/level. At the end, all patients underwent a moderate deep inspiration breath-hold at 75% to 80% of maximum inspiration capacity, and a breath-hold length of 20-30 s. During this training, three CT scan studies were acquired in order to evaluate the interfraction reproducibility of tumor position.
Two patients were not compliant to perform deep inspiration breath-hold using the ABC system due to their small lung tidal volume. For these two patients a forced shallow breathing was per- Dose prescription was 50 Gy in five fractions for all patients.
VMAT plans were generated with Ergo++ treatment planning system (TPS) (Elekta, Crawley, England). This is an anatomy-based TPS that supplies a simplified approach to create VMAT plans by manually predefining a series of aperture shapes in conjunction with the beam's eye view (BEV) of the target and organs at risk. Planning procedure was reported in details in a previous study. 4 However, because the delivery of a SBRT-VMAT arc takes more than a single tolerable breath-hold, we designed a solution to perform a full arc rotation delivery splitting the arc into short sub-arcs, for each of which the delivery time was defined according to the patient predefined breath-hold period. In the first step, the aperture shape for each control point within a sub-arc was determined by the BEV of the target and the adjacent critical structures, automatically adapting the leaf edge to the outline of the PTV. Then, the beam weights for all the control points were optimized by inverse planning based on the simulated annealing optimization algorithm, so defining the doserate/monitor units number ratio for each control point. All plans were optimized with a single full arc. The entire gantry rotation was described in the optimization process by a sequence of 90 control points, i.e., one every 4°.
Patient set-up was checked before every treatment fraction using the portal images obtained by two perpendicular square open 10 × 10 cm 2 beams (each of which delivers two monitor units) and their comparison with the corresponding digitally reconstructed radiographs obtained from the planning CT dataset. Any deviation greater than 3 mm in the isocenter position was immediately corrected. The EPID portal image obtained at the end of the delivery of the VMAT arc in the first treatment fraction was assumed as the reference image for the subsequent daily γ-analysis. In other words, the first portal image was used as a surrogate of EPID transit signals to ensure the highest treatment reproducibility. This means that once the patient setup is corrected before each treatment fraction, IVD γ-analysis should supply correct results if there are no linac or no breath-control system failures.

2.B | The SOFTDISO software
SOFTDISO is a commercial system for IVD developed within an Italian National research project. 16 The mathematical aspects of the dose reconstruction algorithm were deeply explained in a previous paper. 20 The SOFTDISO software is directly interfaced with the Record & Verify system of the radiotherapy network (Mosaiq, Elekta, Crawley, UK) and consists of two integrated modules. The first one, called the "Patient commissioning module", imports the DICOM files from the CT scanner and the TPS in order to extract all the needed radiological and geometrical parameters. The second module, called the "Test Computation Module", was developed to obtain the R ratio between the daily reconstructed dose at isocenter and the planned isocenter dose. The accuracy of this procedure has been well reported in literature 19 providing a tolerance level of ±5%. The module supplies also the γ-analysis for daily EPID images with respect to a reference one. The γ-analysis tests were evaluated in terms of percentage of points with γ value less than one (γ % ) and in terms of mean γ values (γ mean ). Based on previous experience on other anatomical sites, 20,21 a 3% (global)-3 mm criteria was adopted. Alert criteria of γ % < 90% and γ mean > 0.67 were chosen, to accept only 10% of values above 3%/3 mm and an average discrepancy of 2%/ 2 mm. These criteria seems to be a reasonable choice to provide the detection of significant errors but without a major number of false positives.
The EPID portal images were automatically imported by the SOFTDISO software via DICOM protocol and the IVD results were delivered to a computer screen in quasi real-time, that is within one minute from the end of the arc delivery.

2.C | IVD clinical workflow
Electronic portal imaging devices-based IVD was performed for each treatment fraction for all enrolled patients, with the aim to steadily track the treatment reproducibility (mainly, in terms of setup inaccuracy and/or breath-hold failure). If the R ratio and/or the γ-analysis values should exceed the tolerance levels, then the medical physicist examines the IVD chain in order to detect and remove any possible sources of errors. When the dose discrepancies were unclear, a new CT scan was required for the patient resimulation and replanning. At the end of the treatment, as the final act of a multistep quality assurance process, 29 an IVD report is inserted into the patient's medical chart, providing a record of the actual dose received by individual patients.

| RESULTS
A total of 50 transit EPID images, one image for each SBRT-VMAT plan, was acquired during the treatment fractions of the 10 patients.
Two images were removed from analysis for an electronic acquisition failure. Figure 1 shows the R ratios, γ % and γ mean histogram values for all patients; the black bars refer to the eight patients treated with ABC and the white bars to the two patients treated in free-breathing. Table 1 reports the overall results for R, γ % and γ mean metrics.
For the eight patients treated with ABC spirometer, the results show a very high level of accuracy in dose delivery, with no dose discrepancies in any patient (except one test) and a very high interfraction reproducibility. In particular, only one fraction of a patient of this cohort supplied one of the three metrics out of tolerance level, with γ % value equal to 78% (while R ratio and γ mean values were equal to 0.95 and 0.56, respectively). Figure 2   The integration in SBRT process of advanced techniques as VMAT and gating/breath-hold systems for organ-motion control has the potential to further increase the accuracy and outcomes of SBRT F I G 1 . Distribution of (a) R ratios, (b) γ %, and (c) γ mean values for the 10 patients. Black bars indicate the tests for the patients treated with the ABC system; white bars indicate the tests for the two patients treated in free-breathing.  instead focused on the ability of our IVD procedure to identify treatment-related dose discrepancies also during SBRT-VMAT delivery.
This aim has been well-illustrated by two examples. First, major discrepancies in treatment reproducibility were found for the two patients treated without ABC. In particular, we want to highlight that the observed discrepancies for these two patients were detected only by γ-analysis of the 2D portal images. This finding was mainly due to the fact that in SBRT-VMAT treatments of lung metastases, the isocenter is always located at the center of the lesion so that In particular, our overall results confirm the correctness of the adopted tolerance/action levels of 5% for R ratios, and >90% of γ % pass-rate and <0.67 for γ mean when using a 3%(global)/3 mm criteria.
For the eight patients that underwent ABC treatment, 100% of the R ratio, γ mean and γ % tests resulted within their respective alert criteria, excluding one test for which the reasons for the disagreement have been understood. The adopted tolerance levels clearly demonstrated the power to highlight even small discrepancies in the treatment reproducibility. In particular, the γ % pass-rate metric resulted in the best discriminator for the detection of discrepancies in the treat- An ideal IVD system should produce results instantaneously, during the beam delivery, in order to detect gross error as they occur, e.g., before the delivery of the whole dose to the patient. In the past years, we explored this feasibility for dynamic conformal arc therapy. 38 Our aim was to monitor the treatment reproducibil- cases of major discrepancies. 43 Obviously, a true real-time IVD approach is even more stringent for single fraction radiosurgical treatments.

| CONLUSIONS
In conclusion, our preliminary results suggested that our IVD method provide a fast and accurate procedure in clinical routine for verifying delivered dose as well as for detecting dose discrepancies in lung SBRT treatments. IVD has been shown to identify dose discrepancies that would have been missed with other quality assurance methods. The use of ABC in lung SBRT translates in very high reproducibility treatments.

CONFLI CT OF INTERESTS
We certify that any actual or potential conflicts of interest do not exist regarding this paper; the work is original, has not been accepted for publication, nor is concurrently under consideration elsewhere, and that all the authors have contributed directly to the planning, execution, or analysis of the work reported or to the writing of the paper.