A novel quality assurance procedure for trajectory log validation using phantom‐less real‐time latency corrected EPID images

Abstract The use of trajectory log files for routine patient quality assurance is gaining acceptance. Such use requires the validation of the trajectory log itself. However, the accurate localization of a multileaf collimator (MLC) leaf while it is in motion remains a challenging task. We propose an efficient phantom‐less technique using the EPID to verify the dynamic MLC positions with high accuracy. Measurements were made on four Varian TrueBeams equipped with M120 MLCs. Two machines were equipped with the S1000 EPID; two were equipped with the S1200 EPID. All EPIDs were geometrically corrected prior to measurements. Dosimetry mode EPID measurements were captured by a frame grabber card directly linked to the linac. All leaf position measurements were corrected both temporally and geometrically. The readout latency of each panel, as a function of pixel row, was determined using a 40 × 1.0 cm2 sliding window (SW) field moving at 2.5 cm/s orthogonal to the row readout direction. The latency of each panel type was determined by averaging the results of two panels of the same type. Geometric correction was achieved by computing leaf positions with respect to the projected isocenter position as a function of gantry angle. This was determined by averaging the central axis position of fields at two collimator positions of 90° and 270°. The radiological to physical leaf end position was determined by comparison of the measured gap with that determined using a feeler gauge. The radiological to physical leaf position difference was found to be 0.1 mm. With geometric and latency correction, the proposed method was found to be improve the ability to detect dynamic MLC positions from 1.0 to 0.2 mm for all leaves. Latency and panel residual geometric error correction improve EPID‐based MLC position measurement. These improvements provide for the first time a trajectory log QA procedure.


| INTRODUCTION
Trajectory log files are records automatically generated at the end of each field delivered in treatment mode by a radiotherapy machine.
These log files are a sequence of snap shots of the machine state taken at fixed intervals. The interval depends on the machine type, being as short as 20 ms for new models. The machine state record includes parameters, such as energy, dose rate, position of each multileaf collimator (MLC) leaf, gantry angle, collimator angle, and couch position.
Because of their availability, correlation with the actual treatment delivery 1,2 and high precision, the use of trajectory log files is gaining popularity as a QA tool for patient specific and routine QA. [1][2][3][4][5][6][7] Note, however, that the parameters recorded are taken from the machine's control system; they are not measurements and can be in error. MLC positions reported in the log have been found to deviate from the actual delivered positions. 1 It is essential, therefore, to have a mechanism to both commission and to periodically QA the log file. 2 Eckhause 2 proposed a phantom-based methodology to verify trajectory logs using the electronic portal imaging device (EPID). With the fiducials on the phantom used as landmarks, it was shown that it was possible to triangulate the leaf positions accurately enough to verify the positions reported in the log file. As this technique requires a phantom, the EPID was placed at source to detector distance (SDD) of 150 cm with a stationary gantry. This not only limits the verification of MLC within AE13 cm from isocenter for aS1000 panel but also limits the ability to check rotational delivery techniques such as VMAT.
Recently, Zwan et al 8,9 proposed an EPID-based phantom-less methodology for MLC QA. This allowed the EPID to be extended to the isocenter (SDD 100 cm) to image all the MLC leaves and to localize their position to within AE1.0 mm, which is sufficient to satisfy the tolerance required by TG-142. 10 Their approach was novel in that instead of using integrated images, the positions of the MLC leaves were determined while they were in motion using EPID images acquired at approximately 10 Hz. An accurate localization of moving leaves would make possible the verification of the MLC positions reported in the trajectory log. However, the accuracy is insufficient to be used to validate trajectory logs.
A factor contributing to the uncertainty is that the moment during the dose delivery at which a moving leaf in imaged was not corrected for timing delays or latencies in the readout of the EPID image.
Because the leaf is moving, this results in a shift of the imaged position.
In this study, we describe a phantom-less method to improve the accuracy leaf localization by correcting for EPID readout timing latencies. 11 EPID positioning errors that depend on gantry angle, and MLC centerline calibration errors are also corrected for, resulting in a QA procedure with accuracy sufficient to QA trajectory log analysis. Such a QA process would increase efficiency and reduce the workload of clinical physicists.

| MATERIALS AND METHODS
To validate a trajectory log, it is necessary to acquire EPID images of the leaves while they are in motion. Images were acquired in cine mode at approximately 10 frame per second (fps) via iTool Capture In EPID images of a sliding window (SW) slit field moving uniformly from right to left across the EPID panel, the slit appears slightly skewed with the leaf closest to the gantry appearing to be trailing the leaf furthest from the gantry by approximately 2 mm ( Fig. 1).
It is hypothesized that the skewness is caused by the multiplexed readout of the EPID panel. The readout is row-wise, with the row closest to the gantry being readout first. 11 The time elapses between the imaging and readout of the last leaf pair (row 1) and that of the first   were determined as the maximum gradient points of the penumbrae defined by the 80% and 20% of the maximum value. The dynamic displacements of each leaf pair, x Aj (k) and x Bj (k), were defined as the difference of the instantaneous leaf positions, X Aj (k) and X Bj (k), and the corresponding stationary positions, X Aj | v=0 and X Bj | v=0 (Fig. 2): The mean center position of the leaf pair, j, is defined as: The time delay, t d,j , of panel of a leaf pair j relative to a reference leaf pair r can be determined by: The relationship between t d,j and position, y, perpendicular to the leaf travel direction from leaf j to l can also be expressed as: And was derived with the linear regression. Here, leaf pair number 60, r = 60, closest to the first line readout was first used to determine the t d,j relationship for both panels. All the frames cap- Combining (4) and (5) x Bj,ref t ð Þ, at the gantry angle θ at time t relative to the radiation isocenter.
The To verify the methodology for dynamic trajectory log, a set of five test patterns were delivered.      The radiological to physical gap correction for the M120 MLC was found to be 0.1 mm at isocenter for the EPID. With the 0.5 mm miscalibration, EPID-based measurements showed an average of 0.5 mm gap discrepancy from the plan positions (in black circle) as shown in Fig. 5. Interestingly, the EPID-based measurements showed the same magnitude of discrepancy with logfile-based measurements (red cross). Figure 6 shows the measured leaf gap compared to the plan, black circles, and trajectory log, red crosses, reported gap of each leaf after the gap was adjusted to the correct position with a feeler gauge. Both the plan and trajectory gap comparisons show no significant gap error. While these results indicate that the trajectory log analysis was insensitive to the calibration error, the EPID-based analysis was found to be sensitive to the same error.

| RESULTS
Using (5), the residual panel position error in the x direction (parallel to the MLC motion) was determined to be 0.84 mm at gantry 0.
As this error was the same order of magnitude of the potential MLC error, the correction was deemed significant and incorporated in all the dynamic analysis. To verify the methodology for dynamic trajectory log, a set of five test patterns were delivered. Table 1 shows the summary and the descriptions of the test patterns. The mean, range, and standard deviation of the Δx Aj t, θ ð Þj traj and Δx Bj t,θ ð Þj traj of each plan with and without the correction were reported to assess the performance of this methodology. Figure 7 shows the comparison between the uncorrected and corrected measurements compar-   The skewness in the comparison was found to be directional dependent as shown in Fig. 7(e). When the speed of the field increased from 0 to 2.5 cm/s, the range of position errors increased from 0.22 to 0.92 mm and the standard deviation also increased from 0.06 to 0.54 mm ( Table 2).   Figure 9 shows an example the same VMAT delivery when a failing motor, B37 (blue arrow), was identified by corrected EPID image but was not identified by the trajectory log. In this case, the average position and gap error were found to be 0.6 and 0.8 mm, respectively.

| DISCUSSION
In this study, we have presented a phantom-less method to measure the dynamic MLC leaf positions with submillimeter accuracy which the conventional integrated image technique, traditionally used in portal dosimetry, cannot achieve. Similar to previous studies, 12,13 the maximum error of the EPID at SDD 150 cm was found to be within 0.5 mm. However, the maximum error was found to be in the order of 1.0 mm at SDD 100 cm which is significantly larger the error at SDD 150 cm. This can be attributed to the single SDD in the IsoCal procedure. More interestingly, the magnitude and trajectory of the flex maps were found to be machine specific. At SDD 100 cm, the precision without the machine specific panel geometric residual and temporal correction is insufficient for trajectory log verification. The results from this study showed that the readout time corrections and geometric residual corrections were able to improve the precision of both static and dynamic absolute MLC position measurements. As the method proposed by this study does not require a phantom or its setup, additional workload for clinical physicists is minimized. The accuracy reported here is comparable to the earlier work Eckhause et al, 2 where a phantom was used to measure the position of MLC leaves.
In general, the results from the trajectory log were found to be very consistent with the measured data for both static and dynamic delivery. In this study, it was found that the log file was insensitive to MLC calibration errors. Similar to a recent study, 1 we also found the trajectory could miss certain MLC errors. It is important to have an independent and highly accurate method of measuring static and dynamic MLC position if log files are routinely used for patient specific QA. With the increasing need of high throughput, a phantom-less method can also improve the scalability and ease of implementation.
As this study measures the positions of MLC leaves while they are in motion, correction for the temporal delays of the EPID readout was found to be an essential correction.