Technical Note: Characterization of clinical linear accelerator triggering latency for motion management system development

Abstract

Purpose

Latencies for motion management systems have previously been presented as guidelines for system development and implementation. These guidelines consider the overall system latency, including data acquisition, algorithm processing, and linac triggering time. However, during system development, the triggering latency of the clinical linear accelerator is often considered fixed. This paper presents a method to decouple the linac‐only triggering latency from the total system latency such that latency can be considered in terms of only the linac‐independent aspects of the system.

Methods

The linac‐only latency was investigated by considering the time at which a linac response was observed relative to the time at which a beam‐on/off triggering signal was sent to the linac. The relative time between the two signals was analyzed using a multichannel oscilloscope with input signals from a custom gating box to manually trigger the beam state as well as a diode positioned at beam isocenter to monitor the linac response. The beam‐on/off latency was measured at multiple energies (6/18 MV) and repetition rates (100–600 MU/min) to investigate beam setting dependencies.

Results

The measured latency was observed to be dependent on the accelerator settings for repetition rate and energy, with beam‐on latencies decreasing with increasing repetition rate and decreasing energy. In contrast, the opposite trend was present for the observed beam‐off latency. At 600 MU/min, beam‐on/off latencies were observed to be 3.37/1.45 ms for a 6 MV beam and 6.02/0.73 ms for an 18 MV beam. Negative latencies were possible for beam‐off measurements due to the mechanical latency being less than the pulse separation at given repetition rates.

Conclusions

The linac latency associated with triggering the beam‐on/off was determined to have a minor contribution to the total allowable system latency; thus, the majority of the total system latency can be attributed to linac‐independent factors.

1. Introduction

Motion management systems for radiation therapy treatment delivery platforms that utilize gating for intrafraction motion compensation inherently rely on a short response time between position detection and turning the beam on or off. The time between a feature entering or leaving the gating window and the time at which the beam is turned on/off defines the latency of the motion management system. The full system latency can be broken down into components including the image acquisition and streaming, the image processing, and the linac beam‐on/off triggering latencies. While the full system latency is the most important latency to quantify when considering potential margin reduction and the accuracy of current systems, an understanding of the triggered linac‐only latency (time between the signal being sent to turn the linac on/off and the corresponding observed change in beam state at isocenter) can be extremely useful as investigators and companies continue to develop new motion management approaches.^{1, 2, 3, 4}

When developing new motion management systems, developers have some control over the image acquisition method and parameters as well as the image processing methods used to determine feature locations. However, the linac latency is inherent to the treatment machine being used. Therefore, as developers try to limit the overall system latency (<500 ms as recommended by AAPM Task Group 76⁵), it is useful to consider the necessary speed of the system with the linac‐only latency removed. In this manner, it allows for developers to consider the system latency threshold with only the linac‐independent components of the latency. Ultimately, quantifying the inherent beam‐on/off latency will provide researchers with a better understanding of the necessary speeds for both image acquisition and processing.

While much attention has been devoted to investigating the total system latency associated with commonly used motion management and delivery systems,^{6, 7, 8, 9, 10, 11} to the best of our knowledge, there is only one prior group of investigators reporting linac‐only latency.¹² This previous linac‐only investigation was performed on a Novalis system in which a linac time delay of 60 ± 20 ms was observed.¹² Considering that full system latency on Novalis systems has been reported between 90 and 200 ms,^{7, 9} the linac‐only latency of 60 ms is significant. While the measurements of the linac‐only latency and the full system latency were made by separate researchers and with varying methods, these results suggest that linac latency may play a large role in the overall system latency, thus warranting further investigation. This work investigated the linac‐only latency associated with triggering a Varian Clinac 21EX beam to determine the relative time that must be considered for linac triggering in the development of a motion management solution.

2. Materials and methods

The beam‐on and beam‐off linac latency was investigated for a Varian Clinac 21EX (Varian Medical Systems, Inc., Palo Alto, CA) clinical linear accelerator at the University of Wisconsin Medical Radiation Research Center. An in‐house beam triggering system was developed to allow for manual triggering of the beam in service mode at user‐specified times. The triggering system was controlled using MATLAB (R2017A, MathWorks, Torrance, CA) and interfaced with a data acquisition (DAQ) device (U3‐HV, LabJack Corporation, Lakewood, CO) to output a voltage characteristic of the desired linac beam state (beam‐on, +5V; beam‐off, 0V). The output voltage was then routed to a custom‐made circuit that converted the voltage output from the data acquisition device to the signal expected by the linac to designate the beam state.

To characterize the beam latency, a method similar to that proposed by Wiersma et al.⁶ was implemented in which a multichannel oscilloscope (TDS 2004B, Tektronix, Inc., Beaverton, OR) was used to provide a high temporal resolution comparison between the beam triggering signal and the detected response. The triggering signal initiation time was detected by splitting the output signal from the DAQ device to both the oscilloscope and the linac controller. The rising or falling edge of the linac triggering signal was used to characterize the time at which the signal initiation was sent to the linac (ideal linac delivery time). To characterize the observed time of response (actual linac delivery time), a diode (IsoRad, Sun Nuclear Corporation, Melbourne, FL) was placed at the beam isocenter in a 15 × 15 cm field with 10 cm virtual water backscatter, and the output was measured (Fig. 1). The diode was connected to the oscilloscope to measure the time at which a response was observed and was considered on the same time scale as the linac triggering signal. The oscilloscope was triggered on the linac initiation rise/fall time to isolate the region where the linac latency was observed. Figure 2 provides a block diagram depicting the experimental setup.

Figure 1.

Setup of the diode used to measure the observed response of the linac. The diode was positioned at isocenter of a 15 × 15 cm field on top of 10 cm virtual water for backscatter. [Color figure can be viewed at wileyonlinelibrary.com]

Figure 2.

The experimental setup for the linac latency measurement is depicted. Latency measurements were performed by considering the temporal outputs from the DAQ device to quantify the beam‐on/off signal to the linac and also the diode response to quantify the time of the observed response. Both signals were observed on an oscilloscope and measured to determine the relative latency. [Color figure can be viewed at wileyonlinelibrary.com]

The beam‐on/off latency was determined by comparing the time at which the beam switching signal occurred relative to the time that a change in the diode response was observed. The observed beam‐on response was characterized by the rising edge of the first linac pulse detected, and the observed beam‐off was characterized by the time at which the signal from the last linac pulse fell to 80% of the maximum signal. For the beam‐off analysis, the response signal acquired from the diode was smoothed using a Savitzky–Golay filter to better quantify the 80%‐point relative to the maximum value. The dependence on both the energy and repetition rate settings was investigated by taking measurements at 6 and 18 MV as well as linac repetition rate settings ranging from 100 to 600 MU/min. For each energy and repetition rate, 10 measurements were performed for both beam‐on and beam‐off evaluations. Oscilloscope outputs were recorded for a duration of 25–50 ms with 2,500 samples over the span of the viewing window, resulting in a time resolution of 0.01–0.02 ms.

A second investigation was performed to determine the range of beam‐on latency values possible for each beam parameter setting. This was performed with a similar setup; but instead of using the DAQ device, a waveform generator was used to trigger the beam. The signal generator was set to output a square wave that alternated between 0V and +5V at a rate of 14.995 Hz. Therefore, the pulse repetition frequency (prf) of the linac pulses was very nearly a multiple of the triggering signal frequency (prf of linac is 360 Hz for 6 MV). Again, an oscilloscope was used to compare the linac triggering signal and the observed linac pulses. The oscilloscope was triggered on the rising edge of the linac trigger signal from 0V to +5V. This produced an output where the linac triggering signal remained static as viewed on the oscilloscope; however, the linac output pulses moved slowly across the oscilloscope. This effect was due to the slight differences in the frequency between the prf and the linac triggering signal. Essentially, with each iteration, the triggering signal was being sent closer in time to the next expected linac pulse until the next pulse could no longer be resolved due to the linac latency. As the linac pulses approached the linac triggering signal (representing a short linac latency), they would begin to drop out due to the limit of the mechanical latency (lower limit of linac latency due solely to linac signaling and beam generation). The nearest and furthest in time that the first linac pulse was located relative to the linac triggering signal defined the minimum and maximum latency observed. Frame‐by‐frame analysis of 60 s video sequences was used to quantify the range of possible values.

3. Results

Figures 3(a)–3(d) provide an example of the trigger signal and the diode response for the beam‐on/off measurements of a 6 MV beam at both 100 and 600 MU/min. The solid black lines represent the time at which the beam‐on/off signal was sent to the linac, and the dashed black lines represent the time at which a diode response was observed. The lighter gray region between the solid and dashed lines represents the latency observed in each case. As the latencies observed were on the same order of magnitude as the time between linac pulses, it was possible for negative beam‐off latencies to be observed. In this case, the last linac pulse observed occurred before the beam‐off signal was sent, and the signal was received by the linac before emitting a subsequent pulse. In this regard, since the beam‐off latency was impacted by the rate of the pulses emitted, a worst‐case beam‐off latency was calculated and shown with a vertical dash‐dot line in the figure. The worst‐case beam‐off latency was calculated as the time between the beam‐off trigger (solid black line) and the time just before the next expected pulse, based on the average pulse frequency for each repetition rate. Figures 3(e)–3(h) provides a similar example for an 18 MV beam.

Figure 3.

The diode response relative to the triggering signal for beam‐on (a,c,e,g) and beam‐off (b,d,f,h) is presented for a 6 MV (a–d) and 18 MV (e–h) beam at 100 MU/min (a,b,e,f) and 600 MU/min (c,d,g,h). The solid line indicates the signal initiation time, the dashed line indicates the observed response time, and the dash‐dot line indicates the worst‐case response time (beam‐off only). The light gray region represents the observed latency and the dark gray region represents the worst‐case latency (beam‐off only). [Color figure can be viewed at wileyonlinelibrary.com]

The linac latency was investigated with two energies (6 and 18 MV) and six repetition rates (100–600 MU/min) to determine if there was any latency dependence associated with the beam delivery settings. For the observed beam‐on latency, there tended to be a decreased latency with increasing repetition rate as well as with decreasing energy. Alternatively, for the observed beam‐off latency, an opposite effect was seen in which the latency increased with increasing repetition rate and decreased with increasing energy. This opposite effect was likely due to the potential for negative beam‐off latencies and the uncertainty associated with where the beam‐off signal occurs between subsequent pulses. When investigating the beam‐off latency in terms of the worst‐case latency, the results became more aligned with what was observed for the beam‐on case for which the latency decreased with increasing repetition rate and decreasing energy. The dependencies for both the repetition rate and the beam energy are presented for the observed beam‐on, observed beam‐off, and the worst‐case beam‐off latencies in Fig. 4. Each data point is the average of ten independent measurements and the error bars indicate the standard deviation of the measurements. Ultimately, at 600 MU/min, the average observed beam‐on latency was 3.37 ± 2.88 ms for the 6 MV beam and 6.02 ± 2.49 ms for the 18 MV beam, while the observed beam‐off latency was 1.45 ± 2.58 ms for the 6 MV beam and 0.73 ± 2.03 ms for the 18 MV beam. When considering the beam‐off latency in terms of the worst‐case calculation, the latency increased to 2.56 ± 1.77 ms for the 6 MV beam and 2.40 ± 1.53 ms for the 18 MV beam. The range of possible values as determined from the data acquired using the waveform generator is provided for the beam‐on latency in Fig. 5. The minimum and maximum observed values for both 6 and 18 MV are plotted with the shaded regions indicating the range of latency values possible. The possible range of values at 6 MV was between 0.2 and 19.0 ms, while the range at 18 MV was between 0.2 and 38.0 ms.

Figure 4.

Values are presented at varying repetition rates and energies for the (a) observed beam‐on, (b) observed beam‐off, and (c) worst‐case beam‐off latencies. Each point is the average of ten independent measurements with error bars indicating the standard deviation of the measurements. [Color figure can be viewed at wileyonlinelibrary.com]

Figure 5.

The range of latencies that were observed for a 6 and 18 MV beam at repetition rates between 100 and 600 MU/min [Color figure can be viewed at wileyonlinelibrary.com]

4. Discussion

The linac‐only latency for a Varian Clinac 21EX linear accelerator was investigated to characterize the relative impact of linac triggering on the overall system latency. The beam‐on linac latency was less than 25 ms for all energies and repetition rates investigated. The latency was less than 10 ms for all repetition rates at 6 MV and repetition rates above 200 MU/min at 18 MV. A dependence in the latency was observed for changing energy and repetition rate parameters. At 6 MV, the beam‐on latency initially decreased for repetition rates of 100–400 MU/min, before a relatively constant latency was reached from 400 to 600 MU/min. The approximately 4 ms latency observed between 400 and 600 MU/min was believed to be the most representative of the mechanical latency of the machine. The observed variability with energy and repetition rate was likely due to the linac pulse separation for each setting. At the nominal dose rate of 600 MU/min, the Clinac 21EX operates at a prf of 360 pps for 6 MV and 180 pps for 18 MV, which corresponds to a constant 2.78 and 5.56 ms pulse separation, respectively. As the repetition rate is decreased, the temporal pulse separation can increase at portions of the linac pulse train. Decreased repetition rates in Varian linacs are achieved through pulse dropping in which pulses are removed from the nominal 600 MU/min repetition rate by delaying the electron gun pulses relative to the radiofrequency pulses. By making these signals noncoincident, there is no beam pulse emitted. For instance, at 400 MU/min, every third beam pulse is removed from the pulse train, such that at 6 MV, there are two pulses separated by 2.78 ms, followed by a pulse separated by 5.56 ms. Correspondingly, at 100 MU/min, there is one linac pulse followed by the dropping of five pulses such that the observed pulses are 16.7 ms apart. In regard to the results observed, the increase in the time between linac pulses at lower repetition rates led to an increased uncertainty of when the triggering signal was sent relative to the next expected pulse. In practice, the linac triggering signal from the DAQ device was sent at a random point in time relative to the pulse train, and thus, there was a degree of uncertainty regarding where it occurred relative to the next possible pulse. This uncertainty was present for all repetition rates, but much more pronounced at lower pulse repetition frequencies, leading to increases in the observed linac latency. Essentially, as the time between neighboring pulses increased to a value greater than the time necessary for the linac signaling, the linac latency became more dependent on the pulse repetition frequency. This was further demonstrated with the increase in the observed standard deviation of the measurements at lower repetition rates. A similar repetition rate dependency was observed for the 18 MV beam, however, at an overall higher latency relative to the 6 MV beam. This effect was believed to be again tied to the halving of the pulse repetition frequency for the 18 MV beam, and the corresponding increase in the time between pulses relative to 6 MV (360 pulses per second at 6 MV, 180 pulses per second at 18 MV).

Analyzing the range of possible beam‐on latencies further strengthened the argument that much of the observed difference with energy and repetition rate was due to the time between linac pulses. The minimum values observed across the repetition rates produced relatively consistent results at approximately 0.5 ms, with all minimum values being less than 1.5 ms. This minimum latency observation was likely the best indicator of the true mechanical latency associated with triggering the beam. However, while the values got as low as 0.5 ms for the minimum latency, this was by no means a hard threshold at which latency would occur if the triggering signal was sent with perfect timing. It was observed that the linac pulses began to drop out before that minimum value was truly reached. Additionally, it would be expected that as one linac pulse became too close to the triggering signal, it would simply drop out and the latency would be defined by the next linac pulse. While this was the case in most scenarios, there were occasions when multiple subsequent pulses would drop out simultaneously, having the latency then be defined by what was previously the third or the fourth pulse observed. These effects were not present every time; however, it does indicate that the linac latency cannot solely be attributed to the time between pulses, and there are other operating parameters that may affect the true observable latency at each repetition rate.

The observed beam‐off response exhibited latencies that were less than those observed for the beam‐on response, and in many cases produced negative latencies. A negative latency indicated that the last linac pulse observed occurred before the beam‐off signal was sent to the linac. While conceptually a negative latency is not possible, in practice this occurs when the mechanical latency is smaller than the time between neighboring linac pulses. Similar to what was described for the beam‐on case, it is possible for the triggering signal to occur anywhere in the linac pulse train (whether it be immediately following a pulse or immediately preceding a pulse). In this scenario, if the time between the beam‐off triggering and the next expected pulse was greater than the linac mechanical latency, there would not be any pulses produced after the signal initiation. This effect was accentuated at lower repetition rates where the time between linac pulses was long, leading to large negative observed latencies. Due to the uncertainty associated with the beam‐off measurements, the worst‐case beam‐off latency was also investigated. The worst‐case beam‐off latency was calculated by projecting the beam‐off signal to occur at the time just before the next expected pulse following the beam‐off signal (ensuring a positive latency). The next expected pulse was projected based on the pulse train for the given repetition rate. This investigation provided an upper bound of the beam‐off latency and produced results that were very similar to the beam‐on latency. Overall, the values were slightly lower for the worst‐case beam‐off latency (2.56 ± 1.78 ms for 6 MV at 600 MU/min) compared to the beam‐on latency (3.37 ± 2.88 ms for 6 MV at 600 MU/min). Similar trends were observed in response to changing energy and repetition rate.

The linac latencies observed in this work (<25 ms) were much smaller than the linac latency reported previously by Tenn et al. for a Novalis system (60 ms). While the difference in methodology likely played a role in the difference observed, it is believed to also be due to the difference in the linac model. This work utilized a Varian Clinac 21EX which controls the beam through electron gun and RF coincidence/anticoincidence. This method of beam control, along with a gridded electron gun, facilitates a short switching time. Subsequently, it would be expected for this type of beam control to lead to a lower expected linac latency relative to units that rely upon the adjustment of the pulse repetition frequency for beam control. The large differences observed between the linac investigated in this work and those previously reported reinforce the concept that motion management system latency varies between machine models. These variations emphasize the need to consider latency for each machine and set of parameters to ensure that the latency is being accounted for in the correct manner by the entire system.

When considering the linac latency observed in this work relative to full system latencies previously reported on the same linac model (Varian Clinac 21EX), it can be determined that the linac latency is minimal, yet not insignificant. A previous investigation into the beam‐on/off system latency of a Varian Clinac 21EX with a Varian Real‐time Position Management system indicated a latency of 90/80 ms.⁸ Based on the results reported herein, this would indicate that the linac‐only latency can account for approximately 5–30% of the overall beam‐on system latency, depending on the beam energy and repetition rate. Considering the nominal dose rate of 600 MU/min for a 6 MV beam, the overall impact would be closer to 5%, indicating minimal impact relative to other latency components. Therefore, while linac latency is not entirely negligible, the bulk of the overall latency is expected to be due to data acquisition and processing algorithms for the system investigated in this work.

5. Conclusion

Observed beam‐on/off linac latencies for a Varian Clinac 21EX were determined to have a minimal impact relative to latency expectations for a full motion management system. The latency was observed to depend on the repetition rate and energy settings used for the delivery, with the dependence largely being the result of the time between neighboring pulses that is inherent to the beam settings. Ultimately, while the linac latency must still be considered, overall motion management system latency is expected to be dominated by linac‐independent parameters including the data acquisition and computational methods.

Conflict of interest

The authors have no conflicts to disclose.