A hybrid volumetric dose verification method for single‐isocenter multiple‐target cranial SRS

Abstract A commercial semi‐empirical volumetric dose verification system (PerFraction [PF], Sun Nuclear Corp.) extracts multi‐leaf collimator positions from the electronic portal imaging device movies collected during a pre‐treatment run, while the rest of the delivered control point information is harvested from the accelerator log files. This combination is used to reconstruct dose on a patient CT dataset with a fast superposition/convolution algorithm. The method was validated for single‐isocenter multi‐target SRS VMAT treatments against absolute radiochromic film measurements in a cylindrical phantom. The targets ranged in size from 0.8 to 3.6 cm and in number from 3 to 10 per plan. A total of 17 films rotated at different angles around the cylinder axis were analyzed. Each of 27 total targets was intercepted by at least one film, and 2–4 different films were analyzed per plan. Film dose was always scaled to the ion chamber measurement in a high‐dose, low‐gradient area deliberately created at the isocenter. The planar dose agreement between PF and film using 3%(Global dose‐difference normalization)/1 mm gamma analysis was on average 99.2 ± 1.1%. The point dose difference in the low‐gradient area in the middle of every target was below 3%, while PF‐reconstructed and film dose centroids for individual targets showed submillimeter agreement when measured on a well aligned accelerator. Volumetrically, all voxels in all plans agreed between PF and the primary treatment planning system at the 3%/1 mm level. With proper understanding of its advantages and shortcomings, the tool can be applied to patient‐specific QA in routine radiosurgical clinical practice.


| INTRODUCTION
Brain metastases are a common oncological diagnosis 1 and intracranial stereotactic radiosurgery (SRS) has evolved as an important modality of treatment/palliation for that disease. 2,3 It was demonstrated that even with multiple metastases the SRS treatment could provide reasonable local control, 4,5 and a multi-institutional observational study suggested that clinical outcomes for patients with 5-10 individual metastases treated by SRS alone may be non-inferior to those with 2-4 targets. 6,7 While conceptually straightforward in principle, multi-target SRS poses logistical challenges. As the number of treated metastases increases, the traditional SRS paradigm of one isocenter per lesion leads to prohibitively long treatment times and had to be revisited with the goal of simultaneously treating multiple targets. Interestingly, while dynamic conformal arcs were the mainstay of linac-based radiosurgery for years, the feasibility of single-isocenter multiple-target (SIMT) approach was first demonstrated with a relatively new volumetric modulated arc therapy (VMAT) technique. 8,9 The most recent commercial implementation (HyperArc, Varian Medical Systems) offers refinements in terms of planning automation and collision prevention. 10 Alternatively, Huang et al. 11 proposed a concept of single isocenter dynamic conformal arcs (SIDCA), whereby each lesion is treated by a dedicated group of dynamic conformal arcs but all groups share the same isocenter positioned between all targets. This allows for more efficient dynamic arc treatment, as only one isocenter setup is necessary, and the couch angles and arc directions can be optimized for fastest delivery.
A version of SIDCA is commercially implemented in Automatic Brain Metastases Planning Element software by BrainLab. 12,13 It creates a series of dynamic arcs and each lesion can be covered by all or some of them, depending on the relative position, to minimize normal tissue irradiation. Both techniques by necessity produce treatment plans containing complex MLC apertures, and it is prudent to perform patient-specific end-to-end test prior to commencing the treatment. 14 The number of small, off-center targets poses a unique challenge to dosimetry devices commonly used for such tests. The approach should possess high spatial resolution as the lesions could be of the order of 1 cm or less in size. At the same time, the targets could be fairly wide spread, which negates the advantages of dedicated "stereotactic" detector arrays with small detector pitch, that typically have a relatively small active area under the assumption that the lesion would be located at isocenter. 15 Moreover, the targets randomly placed in three dimensions naturally call for a 3D verification approach. The only true 3D dosimeters with high spatial resolution are radiochromic gels/polymers, 16,17 one of which was successfully used for VMAT-based SIMT validation. 18 However, volumetric radiochromic dosimetry is sufficiently cumbersome at this point to prevent its use for routine patient-specific quality assurance. 19 Therefore, a more practical method is needed that combines some high spatial resolution measurements with 3D dose reconstruction over a volume of an adult head. One such approach, which we validate in this paper, is a hybrid technique whereby information collected from the accelerator electronic portal imaging device (EPID) and delivery log files is supplied as input to the independent dose calculation algorithm that reconstructs the expected deliverable dose distribution on the patient CT dataset. 20 2 | METHODS

2.A | System description
The method evaluated in this paper is a part of PerFRACTION (PF) software suite (Sun Nuclear Corp, Melbourne, FL) that provides a number of options for pre-and on-treatment patient-specific dosimetric analysis. We focused on the pre-treatment patient-specific QA (called Fraction 0) and chose the input configuration that, in our opinion, provided the most advantageous balance between the empirical and calculation portions of the analysis. The software runs on a central dedicated Windows server and all routine user interactions occur through a web browser-based interface. At the heart of the method is the graphics processing unit-accelerated superposition/convolution dose calculation algorithm described and validated previously. 21,22 The beam model can be customized by the vendor to fit the user's data, although a generic model for the accelerator class configuration proved sufficient in this work.
The system is compatible with contemporary Varian and Elekta linacs. The verification process starts with transferring the patient CT and finalized Plan, Structure, and Dose DICOM RT objects from the treatment planning system to PF. This establishes a new patient/ plan in the system. The same plan is transferred to the record-andverify (R&V) system and is then delivered to the EPID operating in a cine mode. The compressed (MPEG) EPID movies, one per beam, are stored after the delivery in a specified network directory that is monitored by PF, automatically transferred to the PF server, and associated with the individual beam(s) found in the RT Plan object.
The accelerator log files are processed in the exact same fashion.
The EPID image frames are then synchronized to the log files to determine the exact duration of time when each EPID frame was acquired. This is achieved by first creating a series of predicted images based on the projection of the RT Plan fluence to the plane of the EPID. The predicted images of every segment (or multitude of segments) are then compared to the measured frames to find the maximum similarity. The measured frame with maximum similarity is considered to be acquired during the same segments as the best matching predicted image.
With the synchronization process completed, the frames are then  The PF calculation voxel size is the larger of the TPS or the minimum set in PF, which was 2.5 mm in this work. This voxel size was set to obtain a reasonable compromise between the calculation speed and accuracy and the dose distribution is not distinguishable from the one calculated with a 2 mm voxel 14 at the 1% dose-difference/1 mm distance to agreement level. The resulting semi-empirical dose distribution can be compared to the planned one by standard gamma analysis 23 and dose-volume histogram evaluation.

2.B.2 | Treatment planning
The datasets were transferred to the TPS (Pinnacle v. 14.0, Philips Radiation Oncology Systems, Fitchburg, WI) and the isocenter was placed based on the known locations of the film fiducials visible on CT scans. The next step was devising regions of interest (ROI) for planning. Six plans of two types were created. The first three plans in Table 1 contain only three spherical target ROIs each, with the goal of creating conformal plans without additional constraints. Each target is intersected in the middle by at least one film plane. Plans 4-6 are rooted in real patient datasets. The patient RT Structure DICOM objects were processed to make transfer to the phantom CT possible. The organs-at-risk (OAR) had to be moved some to ensure that they were positioned within the cylinder. The targets were also nudged to intersect with at least one film plane each. An example arrangement can be seen in Fig. 1A, which shows the targets (red) above, below, and intersecting the coronal plane. Two to four planes were measured per plan. Overall, 17 films intersecting 27 targets were analyzed. In addition to those, each plan contained a 2 cm diameter spherical structure (green in Fig. 1A) drawn at the isocenter and planned to achieve uniform 18 Gy dose for normalization purposes.
VMAT optimization employed two full coplanar and two partial The prescriptions followed RTOG 0320 protocol, 24     to radiosurgical applications, the alignment of measured and reconstructed profiles at the 50% level (dose centroid) was evaluated.

2.C | Dose comparison
Horizontal and vertical profiles were drawn in RIT through the center of each target on every film image. The vertical profile always corresponded to the craniocaudal direction. The horizontal profile anatomical direction varied with film orientation, anywhere from anteroposterior to lateral, and the results were segregated accordingly. The metric was, in most cases, the difference in the coordinates of the midpoints between the 50% level dose profile points. In a few instances where the targets were too close to each other to produce clearly isolated dose peaks on the film, the 65% profile points were used to calculate the dose centroid.

3.A | Gamma analysis results
The gamma analysis results (3%G/1 mm) are detailed in Table 2.

3.B | Peak target dose
Both PF and Pinnacle show agreement with measurement largely to within ±3%, which is satisfactory, particularly for the targets less

3.C | Profile alignment
The results of dose profiles alignment between PF and film are presented in Table 3. Within the range of accelerator motions employed in the plans, submillimeter average displacements between the reconstructed and planned dose distribution centroids can be inferred.

| DISCUSSION
While the recent AAPM TG-218 report 29  definitively tests the integrity of the data transfer chain all the way from the TPS to the accelerator, which is one of the most important aspects of the patient-specific end-to-end tests. The beam model quality, which is the frequent culprit in the end-to-end head and neck phantom irradiation failures 32 is also tested, but not by direct dose measurements. We would argue that this level of scrutiny is acceptable for routine QA (as opposed to system commissioning), and furthermore 3D reconstruction with small voxels is more comprehensive than, for example, experimental planar sampling with large detector pitch arrays. Studying the sensitivity of the method to induced MLC errors is outside the scope of this work, but it is likely to be similar to the results demonstrated by others in the related work with high-resolution systems. 33 while any gradual parameter drift is more appropriately addressed by an ongoing comprehensive QA program. Regarding systematic delivery deficiencies, with modern digital accelerators, the known issues such as the overshoot phenomenon 35 are largely considered mitigated. 20 For example, for the TrueBeam accelerator with its 20 ms controller interrogation cycle and strict delivery linearity enforcement inside each control point, 36 even the gantry acceleration trajectory is highly predictable and reproducible. 37 Therefore, while not going as far as endorsing the log file analysis as a "premier SRS/SBRT QA tool," 38 we nevertheless suggest that coupled with thorough TPS commissioning and comprehensive ongoing accelerator QA program, the hybrid verification method validated in this paper is a viable tool that could be applied in clinical practice.

| CONCLUSIONS
A semi-empirical volumetric dose verification system extracts MLC positions from the EPID movies, while the rest of the delivery control point information comes from the accelerator log files. This combination is used to reconstruct dose on a patient CT dataset with a fast superposition/convolution algorithm. The method was comprehensively validated for single-isocenter multi-target VMAT SRS treatments against absolute film measurements. With proper understanding of its advantages and shortcomings, the tool can be used in routine clinical practice.

ACKNOWLEDG MENTS
This work was supported in part by a grant from Sun Nuclear Corp.

CONFLI CT OF INTEREST
SA is a graduate student supported by an SNC grant and VF is the PI on the project. JK is an SNC employee.