Cross-Vendor Comparison of Beam Hold Duration (Latency)

Josh Naylor, MPE, MSc (00:00):

So, I’m going to be talking about latency. The title was there down at the bottom, right in Welsh. I’m not Welsh, but I had to go to Wales three times to get these measurements on the other vendor system. So that’s the explanation for that. And this was actually on those trips. So today I’m going to talk about what latency is and why it matters. Why should you care about it as much as I do what the latency is for a therapist if you don’t have SGRT or if you have beam hold switched off, how long it takes to, to kind of notice the patient’s moved. And then we’ll take a look at the manufacturer’s claims, the specifications from each of them, and the guidance documents that are available with tolerances and kinds of methods. Then I’ll talk about those methods that have been published and are available to us to use to measure latency. We’ll look at the results that I got from all of the major SGRT vendors, compare them and then consider the clinical significance of those results.

Josh Naylor, MPE, MSc (01:06):

So latency is the term that I tend to use, but in the literature, there’s a whole host of terms describing this. Some of them quite wordy. But basically, what we’re talking about with the type of latency that I’ve been measuring end to end is when the patient moves, how long does it take for the radiation to shut off? And why does it matter? Like, why do we need to do these measurements?

Josh Naylor, MPE, MSc (01:31):

So I think probably most of us here are physicists and there’s many therapists here as well. I feel that the reason we all are here is to ensure the right dose of radiation goes to the right place in the patient, right? And hits the tumor and not the healthy tissue around it. So we, all of our quality control tests and much of our treatment planning is focused on that, getting the right dose to the right place. So latency is affecting the positional aspect of that. So if the latency is too long, we’re going to be get delivering the dose to the wrong place in the patient.

Josh Naylor, MPE, MSc (02:08):

So, as I said before we had SGRT and if you have beam hold switched off on your SGRT system, the therapist is watching the patient. And if the patient moves that need to move probably a few centimetres to see it on the CCTV, there are all these processes in the brain that you have to go through to actually press off on the LINAC.

Josh Naylor, MPE, MSc (02:30):

And if you want to, you can dig out your phone and try that QR code there and measure your own reaction time. But here’s some data on kind of reaction times. The average is about 500 milliseconds, but actually it’s a really widespread, so for many people, maybe in a busy clinic, they’ve got a lot on their mind. It might may well take quite a lot longer than that 500 milliseconds.

Josh Naylor, MPE, MSc (02:54):

So yeah, the manufacturer specifications range from around that 500 mark all the way down to 50. Now that 50 is probably not right? Well, I’ll say probably it’s definitely not right. They, what they’ve done there is basically taken 20 frames per second, which Brain Lab can do, or they say they can do, and then divided one second by 20 frames per second to get that 50 milliseconds. So it’s not, not really comparable to the other ones.

Josh Naylor, MPE, MSc (03:26):

So yeah, just to illustrate this with some cheesy PowerPoint end-to-end latency. We’ve got the radiation on on the LINAC and the patient moves, the SGRT system detects that movement processes the signal, and when it exceeds the real time deltas, as they’re called in AlignRT, it holds the radiation on the LINAC and the LINAC itself has to go through some processing.

Josh Naylor, MPE, MSc (03:51):

So in this end-to-end latency, you’ve got that LINAC component. So to quantify that, I did the exact same method with the MV panel. I’ll come back to that method for RPM or RGSC, and that’s about 38 milliseconds on one of our LINAC. So that’s kind of as low as you could go. And of course, it’s more simplistic because it’s just one dimension. It’s not six degrees of freedom. But that’s kind of representative of that LINAC component.

Josh Naylor, MPE, MSc (04:20):

So in the guidance documents they’re basically mostly pretty vague. There’s, there’s not much help for using these about tolerances. TG142 quotes 100 milliseconds, and I just want to kind of emphasize that of expected, it’s got there. A lot of people have misread that. And there’s a number of peer-reviewed journal articles where people have said the tolerance is less than a hundred milliseconds and that’s mistaken it. So if it was 500 milliseconds, was your measurement at commissioning just to spell it out and really labor the point the your tolerance would be 400 to 600 milliseconds. So don’t make the same mistake. Lots of people in in the literature have.

Josh Naylor, MPE, MSc (05:05):

TG302 came along and reassuringly told us that latency measurements are challenging. That’s definitely true. And then they introduced some a little weight. I would say the tolerance is here less than a second for breast DIBH. And the 0.5 there is referenced back to TG76, but if you go back to that original reference, it’s actually talking about real-time tracking and the prediction models. So yeah, I wouldn’t pay too much heed to that.

Josh Naylor, MPE, MSc (05:37):

And I actually discussed with one of the authors of this document who also authored TG302, and she said that although they were released in the same year, this document is the one to rely on for the tolerances, for the numbers because TG302 actually took two or three years to go through the kind of peer review process. So this 200 milliseconds, it’s the first time in any kind of international guidance documents, we’ve had an absolute value rather than the 100 milliseconds of expected. And it does say for free breathing gating. So you might think, should we be applying that to non-gating? Just conventional SGRT. We’ll come back to that towards the end of this, towards the end of this presentation.

Josh Naylor, MPE, MSc (06:27):

So the methods, these are all published in various articles with more details on the method, but you can use a pin diode, Gafchromic film, I try and avoid that as much as possible, as it’s bit of a pain. And then there’s a really coarse one published that’s a stopwatch just being monitored with a video camera. Some centers have used scintillating crystals. I settled on the MV panel method and there’s a couple of publications on that.

Josh Naylor, MPE, MSc (06:58):

And then at this, this meeting two years ago in New York me and a colleague from Southampton Hospital just down the road from us, our kind of neighboring center, we worked quite hard on developing this MV panel method. So I’m not going to go into great detail about how we did the measurements, but you can go back and check out that talk if you want kind of a blow by blow on how to do a really kind of convenient method, cause it doesn’t require any additional equipment and it’s really reliable and I’ve, that’s what I’ve used for all of my measurements.

Josh Naylor, MPE, MSc (07:33):

So onto the results – I’ve done measurements for Elekta and Varian LINAC, done that for AlignRT and for C-RAD on, on both those LINAC. And then with AlignRT done those measurements for version six and version seven we’re on version seven right now.

Josh Naylor, MPE, MSc (07:53):

There was some nice improvements to latency. It got a bit quicker on version seven and I’ve done that for the region of interest, the conventional SGRT kind of set up and then the patch in the respiratory module for the free breathing gating and the patches slightly quicker. We’ll come back to that later. Also, identify from Varian Catalyst, C-RAD and then Brain Lab. So, nice spread of the, the major vendors.

Josh Naylor, MPE, MSc (08:23):

So the results, we’ve got them all here on one graph I’ll focus on them in turn, but you can see there’s quite a wide range and the manufacturer specs are there as well. Largely the end-to-end latency that I measured is a little bit bigger than the manufacturer spec. That’s to be expected because the manufacturer specification is not end-to-end, they normally state it for the SGRT system itself, which is reasonable because they don’t have any control over the LINAC side of things. So they’re saying what does the SGRT system add for latency?

Josh Naylor, MPE, MSc (09:04):

So just to touch on the Brain Lab one because this was done with a slightly different method by the team at Southampton Hospital. So using a pin diode, they actually got some of their electronic engineers to make that. I use the MV panel cause we don’t have as skilled and friendly electronic engineers as their team. But Brain Lab is up at two 60 or so milliseconds and using the same method as AlignRT systems that’s on Elekta LINAC. Both those103 milliseconds. So yeah, to come back to the MV panel method measurements that I’ve done, the AlignRT ones at my center, I’ve got kind of two or three years of annual QA across our four LINAC and then C-RAD I did on both Elekta and Varian at two different sites in Wales and Varian I did on two LINAC at one center. So they’re all TrueBeam except for that one that’s labeled the second bar on the chart. And you’ll probably immediately notice the massive discrepancy between the two C-RAD systems. Now that’s not because of the LINAC, that’s the obvious kind of first thought cause one’s Elekta, one’s Varian. But with a AlignRT we see no difference between Elekta and Varian and LINAC. So I chatted this over cause I was really confused about it. I thought, oh no, did I screw up the measurements on the variant one there with the six 30 milliseconds And I discussed it with a contact in Sweden who’s done a ton of work with C-RAD. Her PhD was using C-RAD for SGRT and her feeling is that it’s to do with the deformable algorithm, the registration algorithm at certain points C-RAD switches from rigid to deformable and the latency seems to potentially, these are all theories ’cause I was just discussing it with her on Monday. The latency seems to increase with the deformable algorithm. And then also there’s this feature called surface averaging in C-RAD and it basically averages the, it’s for breathing averages that breathing over the default setting is four seconds. So it’s actually looking at the previous four seconds and averaging out the breathing and that seems to add to the latency and it would seem that the 142nd value there on the electro LINAC for C-RAD, those parameters were not met. So the latency was a bit better. But yeah, Varian Identify it’s within the manufacturer specification, but it’s way up there compared to the 120 milliseconds on AlignRT. And we’ll look at the clinical significance of these measurements in a moment.

Josh Naylor, MPE, MSc (11:46):

So, one thing to also note on C-RAD, this is a screenshot from when I was doing the measurements, although on the Elekta LINAC we had quite a quick latency of about 140 milliseconds. The display you can see there where it says isocenter shift in the top right with the red and green graph, it actually takes a noticeable kind of few seconds to refresh. So your therapist might be sat there thinking, oh, they’re still intolerance, but it’s just because the graph hasn’t updated and there’s this kind of lag, but the radiation has shut off already. So it is a bit little strange when we were doing it was quite tricky to get the measurements.

Josh Naylor, MPE, MSc (12:27):

So yeah, the clinical impact is kind of the more important thing because it helps us put these measurements in context. So to look first of all at this for, I’ve used the example of SBRT lung because we’ve got that high dose rate. We use flattening filter free for all of our lung patients and thinking of free breathing or breath hold, we do a lot of both of those for our lung patients. So we will come back to gating with that TrueBeam flattening filter free dose rate. If you, this is just a really rough back at the envelope calculation. If you take the 200 milliseconds value from the ESTRO ACRO guidelines as a kind of reference of what latency could or should be and then work that through from the dose rate, you get eight monitor units delivered in that time when the patients moved out of tolerance and the beam should be shutting off. If it was perfect you’d get with zero milliseconds, which is obviously not possible, you’d get zero MU. So for our SABRE patients SBRT so many of them with the prescription schedule we have a thousand MU per day. So the 8 MU is 0.8% and taking the TG142 threshold someone referred to it yesterday as well, just that default, 2%, two millimeters. This would allow for two or three beam holds, 2.5 beam holds before you are starting to kind of exceed that 2% level. And that’s not many is it. We’ve all seen patients, I try and watch some to kind of keep a feel for what’s actually happening on the LINAC and plenty of patients move and have the beam gated that much. So if the latency’s less than 200 milliseconds, say a hundred, that’s then going to raise the ceiling of the number of beam holds you can have.

Josh Naylor, MPE, MSc (14:22):

So for DIBH breast more normal dose rates, not the flattening filter free. I looked at SimRT and kind of tried to assess this breath velocity. I couldn’t think of a better term for it. And how quickly the patient’s breathing out what they’ve got their breath held, how quickly do they, the return the surface returns back to the start position and it’s about four millimeters per second. So if we use again that kind of default TG142 threshold of two millimeters, the positional one this time that’s 500 milliseconds. And at a kind of normal treatment dose rate that would equate to five minus units incorrectly delivered. So for a conventional fractionation that we use 2.6 gray a day, and if a patient’s not terribly compliant, they’re not very good at holding their breath for the treatment, you could have up to 9% being in of the prescription being incorrectly delivered where they’re not at the exact breath hold. So if we take that 200 millisecond tolerance, again, this would allow for about 13 beam holds, which is you could see a patient get reaching that point within one day’s treatment especially if they’re struggling with the breath hold and it, if this is a whole breast PTV, it’s not quite so significant. But definitely if you’ve got a boost or nodal involvement, that’s going to have a bigger impact.

Josh Naylor, MPE, MSc (15:51):

So to come back to gating, we’ve been using the respiratory module from Vision RT since about February, I think this year.

Josh Naylor, MPE, MSc (15:58):

And a colleague of mine very kindly made me this lovely spreadsheet that lets you model the different parameters there and work out the clinical impact of the latency. So it has the inputs of amplitude, breaths per minute latency that you’ve measured for that machine or a tolerance just to simulate it at which phase of the breathing cycle you are interrupting the beam. And we used a few different waveforms. I’ve used Cos^4 here because it’s fairly realistic. A lot of publications have said that’s a good one to kind of model a human breathing.

Josh Naylor, MPE, MSc (16:31):

So it also has this nice pretty page to look at the tolerances that you need and you can put in different amplitudes of the breathing and different breath per minute. So I chose eight millimeters and 19 breaths a minute because this is the kind of worst case. It’s the for 90% of patients in, in the literature it says eight millimeters is the kind of the biggest amplitude you’d be getting. And then 19 breaths per minute is the fastest breathing. So the impact of latency will be worse for these patients. And if you enter those in, it shows you on the, where there’s the highlighted blue bar at which point at what latency level do we reach two millimeters of positional inaccuracy. So as the patient’s breathing when the beam they reach the gate where the beam should shut off, how with this latency of a hundred milliseconds there they will go just under two millimeters further. And the positional, the position won’t be quite right. So yeah, for these worst-case patients, this would suggest that our latency threshold should be less than a hundred milliseconds. So this is, I know I labored the point earlier about a hundred milliseconds from expected, but this is actually an absolute tolerance of a hundred milliseconds. And if we then go to a kind of best case where the amplitude is smaller, five millimeters any less than that and you probably wouldn’t want to bother doing gating, you could just do free breathing treatment. And the breath rate is then down slower at 12. That lets us kind of raise that ceiling of latency tolerance to 250 milliseconds. So you want to set your tolerances based on those worst-case patients of a hundred milliseconds. And I’ve got the results there in the corner from the respiratory module in AlignRT and then the C respiration module for catalyst the C-RAD system. And you’ll see that AlignRT is 84 milliseconds. So within the 100 milliseconds threshold.

Josh Naylor, MPE, MSc (18:51):

So just to kind of sum up and we might have time for a couple of questions. Latency really matters because it’s affecting the accuracy of treatment. So you should measure it and the tolerances, you can do those calculations for yourself and for different cohorts of patients. But I would suggest less than a hundred milliseconds for gated patients, otherwise less than 200 milliseconds for compliant patients. And do your latency measurements across all of your SGRT systems and LINAC and use them to inform your real-time Delta thresholds, because if your latency’s a bit longer you could reduce the RTD threshold and kind of compensate for that. Lower latency is always better. Vision RT has the lowest latency of any SGRT vendor and meets those thresholds that I calculated. Always use the auto beam hold. I said this in New York two years ago. Probably most of you are using that, but if you’re not, just go switch it on. There’s no point having SGRT. That might be a bit of an exaggeration, but I’ll stick with it. No point having it if you’re not got the auto beam, hold on because you’re then relying on the therapist to switch it off at the right moment. And yeah, definite thanks to my colleague who made the latency spreadsheet, Steve, the radiotherapy teams across the whole of Wales. And then Alex at Southampton who did, wrote all the code for analyzing the MV images and my email’s there. If anyone’s got question, you can ask now or during the break too.