Video
library

[Silence]
[00:01:42.252]

Bogdan: Hello, everyone, and welcome to a new Novalis Circle webinar. Hopefully, all of you were able to join today, and we look forward to two great lectures from Dr. Zachary Seymour and Cory Knill from William Beaumont School of Medicine. The agenda for the webinar today revolves around clinical utilization of Elements Multiple Brain Mets SRS. We are going to cover dametric evaluation of the technology. We are going to go through commissioning and clinical implementation of our software on a VersaHD linac. And then we'll go through a series of clinical examples covering the utilization of the technology in the clinical practice.

The two presenters today will highlight both physics as well as clinical topics pertaining to the utilization of Elements Multiple Brain Mets SRS. And as always, the webinar provides CE credits to those who are participating live. Should you have a need for obtaining CAMPEP, MDCB, ASRT credits, please send an email to info@novaliscircle.org following the webinar, and we will provide you with details on how you can obtain credits for today. So, we will start with Cory Knill, who will cover commissioning and technical implementation of Multiple Brain Mets SRS. And then we'll go to Dr. Seymour would cover clinical utilization.

Cory: Thank you, Bogdan, for the introduction. So my topic is going to be mostly focused on everything leading up to treating our first patient. And after that, I'll hand it over to Dr. Seymour, and he'll speak about kind of a similar experience when we started treating patients and what we saw in using the Brainlab system. So just a little bit of background first regarding our hospital, we're located in Dearborn, Michigan, Southeastern Michigan. We treat about 90 patients across 3 machines. We installed a VersaHD with ExacTrac 2018 and treated our first MBMS patient in 2018. And we're currently treating with one to three cranial SRS patients per week using a combination of Brainlab's MBS, Brainlab's cranial VMAT module, and cones as well. So our entire health system, our three major cancer centers are located in Southeastern Michigan. We treat about 230 patients per day at our Royal Oak facility. They have a 4C Gamma Knife SRS system that was treating about 330 patients per year up until last year.

Our Dearborn branch, we're treating about 80 patients per day. And like I said, we installed the VersaHD in 2018 and that's what we started treating patients. One of the major reasons we want to install the system at Beaumont Health is that our Gamma Knife was going down for an upgrade in early 2019. And it was gonna take about four to five months to upgrade the system. And at the time, we wanted to provide some additional method for treating some of these cranial SRS patients within our system outside of the Gamma Knife, kind of the logistical challenge of this, we're going to have to treat patients for five radiation oncologists spread over three different sites. So what we ended up doing was, like I said, we're going to treat about three to four cranial SRS patients per week using combination of these different modalities. And finally, after the Gamma Knife was upgraded and everything was complete, we went live with treating patients on that machine as well. And in our system, we're currently treating with both modalities, both Gamma Knife and linac-based SRS at our site.

So kind of some of the commissioning steps, this is kind of a broad overview of what was done, broken down in the separate sections of creating a machine and beam model, validating that machine and beam model, doing plan comparisons, which is really important for comparing the Gamma Knife cases, validating the treatment system, is what coming out of the Brainlab system accurate, a full end-to-end test of the system, and then finally, a lot of logistical prep work for that first clinical case. So in terms of creating the machine models, it can really be separated into two parts. The first is you have your machine model that is sort of energy independent, and more dependent on the machine type and sort of the coordinate systems, selections that you have made in your clinic. Then you're going to have sort of the beam model, which is more energy-dependent, so you may have multiple beam models for a single machine. For iPlan users, it's really nice. Brainlab support can transfer your iPlan model over into elements and then you just have to validate. We were not iPlan users before. We're using Pinnacle. So instead, what we did is Brainlab provides a series of Excel templates and you basically fill out all of the data in the Excel templates. You send them off to Brainlab, they process it. It only takes a few days, and they send you back your beam model.

One additional feature that came out about a year after we commissioned our system and started treating was a secondary source function. And I wanted to mention that for the existing users who haven't had a chance to get into this secondary source function, it's available in Physics Administration 5.5. What you do is you re-process your transverse data in raw data mode, and you iteratively calculate your radio and source function corrections until you converge on a final solution, which is your secondary source function. And this secondary source function is just going to help a little bit with scatter calculations. And we found it did improve accuracy slightly. So I would recommend using it and trying it out, if you haven't already tried it out. When you go through and you calculate the source function, there's a module in Physics Administration, which will show your calculated beam profiles against the measured beam profiles that you did. It will spit out a lot of statistics in terms of how accurate your profiles are in all these different areas, in penumbras in the high dose region. You can go through and you can adjust specific parameters and then recalculate and see if it improved or degraded. And, of course, that's what we did as a physics group, is we went through and we tried to make it as better as possible. But what we found was that those initial calculations where you just let the software do that recalculation to converge the final solution, those results were the best ones that we found that gave the best average over all of our scans and all of our areas. So we ended up just accepting the ones that were calculated by Brainlab.

So once you've created your machine model, you want to validate. For iPlan users, that means just was all the data transferred correctly, so just checking it versus what was in iPlan. And for non-iPlan users, you want to validate the parameters that you put in with your vendor, or perhaps another physicist who are using a similar system. It's a good idea to just send all your parameters over and say, "Hey, are you using similar values? Is there any big difference?" For both users, what you want to do is you want to make sure to validate the data transfer from Elements throughout your system. So you're going to take, create some test plans in Elements, you're going to send it to your record and verify system, make sure all your parameters came across correctly. For example, if you set 90 for a couch angle, is it 90 in your record and verify system? Is it 90 on your treatment machine? This will typically be done during acceptance. So it's one of the first things that's done during the commissioning process, and it will be done alongside the Brainlab installation engineers while you guys are doing acceptance.

So once you've done validating the machine model, you start validating the beam model. Brainlab provides this really nice beam model verification module, in which you can bring in phantoms from DICOM files and recalculate static fields and arc fields onto these phantoms. So you can compare very, very simple geometries in their calculation module to what you measured. This example right here is showing the stereo phantom we brought in. We also created a water tank, basically, a giant square-full of water to simulate water tank measurements. And then we went through and we calculated an array of commissioning measurements that were done on the water tank in the actual Elements software. We did those calculations when we compared the results of the calculations with the measure. What's really nice about this software is that it gives you some really easy tools to extract that out. You can extract point doses, line doses with these shortcuts and those planes as well. So you can quickly pull out of a lot of data, process it quickly, and compare what you measure during commissioning to what the treatment planning system is calculating to verify its accuracy.

So here are some of those results. The output factors for square fields, we found that they were all within about 1% for all of our square fields, which is really good. PDDs were within 0.5% past dmax. Our diagonal profiles were less than 0.5% difference between measured and calculated for most depths. Once you did get to sort of the periphery of the calculation module, 20-cm depth with 20-cm off axis, you did start to see some large deviations. For us, that's really not clinically relevant at those depths and of those off axis. So for treating a typical cranial patient, we're not going to be treating targets out in that area. You're probably gonna be treating much closer to the 20-cm depth within about 10-cm of your isocenter. That's sort of the limits of where you treat. So in that region, you're much closer, about 0.5% difference with the profiles.

But one thing we did find during the beam model acceptance was that the pencil beam calculation module uses the same penumbra model for both inplane and your crossplane. And you get a little bit of a discrepancy when looking at the crossplane direction, where the jaw will primarily be the beam loading device that forms that edge of the field. So the penumbra model matches up very well for the MLC, but you do see this discrepancy with the jaw. And it's important when you find these discrepancies during commissioning to sort of understand how they can affect patients and how they can affect actual measurements. Something like this if you're just measuring individual field, you could see some discrepancy in the jaw direction. You can see some failures in terms of pass rates, if you're doing a planar analysis. But once you start to deliver these regular fields to an actual patient, and you have a collimator rotation and multiple non-coplanar beams, it's expected that something like this would basically be washed out as you use all these different angles. That's what's expected, but it's always good to, okay, you now know there's this error, let's go through the commissioning processes, measure it with these multiple non-coplanar angles, and make sure that this effect does go away for clinical patients.

So the next step is to create these test plans once we've validated the beam and machine model. The way the Elements system works is that it creates custom templates outside the actual planning software where you specify those prescriptions and you specify beam arrangements. Some suggested initial templates if you have version 1.5, we've got nine-dose templates, a one, three, five-fraction scheme with different coverage levels, various beam arrangements. The big recommendation that we have is to go through and create one of these, make sure you've got all your parameters correctly. You can go through with the Brainlab installation engineers and set everything how you want to set it, and then copy and paste to just create multiple different protocols. This will save you a lot of time in case you have to go back and edit a specific variable. You just have to do it on one protocol and then you copy and paste versus having to do it on 9 or 16 [inaudible 00:14:24.923] all these.

Some good news, if you have the updated software version 2.0, is they have included this advanced editing feature, which allows you to go through and edit a lot of these parameters actually in the software. So you can adjust dose and angles within the software. And that really limits the need to develop or create all those different protocols initially. So if you've started off in version 2.0, you just need to create three-dose protocols, one-fraction, and three-fraction, and five-fraction, and then about five beam arrangements, so two, three and five-arc. And you can create more if you want. This is just what we recommend as sort of the minimum. Another nice feature in version 2.0 is you have the option to include organ-at-risk doses. We mainly follow a TG-101. So in each of the protocols, you can go through and set the OAR dose limits for the different OARs. And even if you don't use them in your actual protocols, these will show up in your actual plan printout. So for documentation purposes, if the brainstem was far away from your target, it will still show the dose. And so it will have a nice record of what each of these structures got.

One important note on these test plans when you start to create them, the MBMS deliveries are unique in that your targets will typically not be located at isocenter versus your other deliveries. Typically, you find a radiation oncology where you'll put the isocenter right in your target. So you'll probably have to be doing a lot of shifting to move detectors off of isocenter to wherever your target is. This can be somewhat difficult to try to figure out, "Okay, where do I need to move my detector?" particularly if you have multiple systems like you're transferring from Elements to record and verify system, to a linac, then you're using a couch on that linac to move, it can be difficult to figure out how does an X shift in the Elements software translate to a shift on the couch in the actual treatment room. So to solve this problem, what we did is we created two targets of two different sizes so they're distinguishable with all these different shifts and the three cardinal directions which were of different magnitudes to be distinguishable. Again, basically, what you do is you go through, you create a plan for these, and you send it through the entire treatment process. So you plan it, you send it to your record and verify system, you put it on the machine, and then you shift your detector on the couch. And what you'll end up is sort of a Rosetta Stone that translates the different coordinates and the different systems from each of the different components. And it's really helpful to kind of have that template where you can use as a sanity check to make sure that your MLC shapes are correct. There's a reasonability check to make sure that if the target is superior to the right, that the MLC is open in the right spot, and also helps in figuring out which direction to move the couch to shift your detector to make sure that you're on the target. So I would highly recommend you doing this from the start.

For plan comparisons, what we want to do is we want to compare the MBMS to our Gamma Knife, because we're bringing over a lot of those Gamma Knife patients to an individual cranial VMAT, which we were planning in our Pinnacle planning system at the time. We wanted to do single and multi-fraction treatments. It's always good to test the full range of different fractionation schemes and treatment types that you're going to be using. What was really nice during this commissioning process was that Brainlab provides the ability to import in our Gamma Knife MR and CTs and structures as well, so structures that were done in Gamma Knife where it could be brought into the planning system, and we could plan directly on these gamut of cases to sort of do an apple-to-apple comparison of what they've done in Gamma Knife to what we were expected to get in our MBMS system.

So in terms of evaluating these test plans, we use some common cranial SRS metrics, the Paddick gradient, which is basically you need the dose fall off away from the target, and conformity indices is how well does your prescription iso line conform to your target. And then for the brain, we're looking at...so brain managed GTVs, the mean/median dose, V10, V12, and these are sort of some indices that have been correlated with radiation necrosis. So from these cases, we looked at targets of all different sizes. And what we found was that the conformity indices, it didn't have a strong correlation to the target size. It was more dependent on sort of the target shape, so the more spherical your target, the better your conformity indices was. But the gradient indices, as you got to about two times the size of your MLC, you saw this large jump in gradient indices. So for these cases, we needed to know, okay, how small could we go before we saw this jump so we can convey that information to the radiation oncologist, and we can kind of make informed decisions on how we're going to treat these patients prior to actually bringing them in whether we use an MBMS plan or a combination of MBMS and Brainlab by cranial VMAT or potentially introduce cones as well.

So this was the first case that we tested out. This is still all in commissioning, 15-grade times one fraction, nice, easy target, good to start with something simple. For the Gamma Knife, you had good conformity and gradient indices for MBMS. This is in version 1.5. So this was available at the time that we did the commissioning. We got similar conformity in gradient indices, the whole brain mean, and V10 was very similar as well.

This next case was chosen because it had two targets that were far away from each other. This would be a good test of the distance ability of the MBMS system to treat targets at distance, and in two different sizes, a small target and a large target. The idea being is that we expected the dose to the large target to be relatively good based off of our first case. But now we're testing out, okay, how does a small target behave in terms of the dose calculation algorithm in MBMS? In terms of comparisons with Gamma Knife, what you see is the conformity incides are similar, and the gradient indices, you start to see that larger gradient indices from that smaller target as we saw it from the previous slides. And the whole brain dose is very comparable as well.

And finally, we're looking at five-target case. And this is where the multi mets module really shines, to be able to treat a lot of targets at once. Once again, your conformity indices are similar to your gradient indices. You see that jump, where we're looking at that really, really small target now. Your whole brain mean dose is a little bit higher than MBMS, and that's likely due to that smaller target having an increased dose. But one of the biggest differences is the beam on time. For this case, for the Gamma Knife, that 10 shots, 9 runs, 105 minutes of beam on time, with the caveat that this case was planned a few months before the Gamma Knife went down, so it did have an older source, so that treatment time is going to be a little bit longer. But even still with a newer source, you're probably looking at 60 minutes or so of beam on time versus the MBMS plan. It was having a six arc-delivery, five table angles. So, from the patient walking in the door to walking out of the treatment room, you're probably looking at more of like 45 minutes total, with imaging in between each couch angle as well. So that's really where you kind of see some of the advantage of the MBMS, is just with that efficiency and treatment delivery.

So in terms of validating the actual treatment system now that we've gone through and created all these test plans and we've seen how they compare against Gamma Knife, the question is, is it real? You've calculated everything in your planning system, can you actually deliver that on your treatment machine? So it all starts with the plan transfer. So we did that during acceptance, does all the data come over correctly? And then what you're going to do is you're actually going to calculate plans on the phantoms and delivering to phantoms to see, okay, can you actually deliver the dose that you expect from your planning system? What's really nice is that the Elements system provides a phantom calculation module where you can bring in all these different phantoms and recalculate clinical plans onto these phantoms. You can toggle MUs on and off to measure for individual beams. You can move isocenters. You can adjust couch angles. So it's a really nice module for doing all these calculations.

So I mentioned before, all these targets are likely to be located away from isocenter. So one of the most important things you're gonna have to figure out during commissioning is how to move your detector for not only commissioning cases but for clinical cases as well. And there's basically two ways you can do this. If you have the ExacTrac system, it comes with a cranial array, and you can export your QA plan and use the cranial array to automatically position your detector in the correct location. It's really easy to use it that way. It's a really nice workflow. It kind of takes the guesswork out of it, or some of the calculation work out of it, I should say. One of the issues with that, though, is you do have to export a different QA plan for each one of the targets. So if you have a five-target plan, you're exporting five additional plans. For an actual patient case, this can make the patient's chart a little bit messy. And at the end of the day, you are measuring fields that should be equivalent to the fields that you're delivering to the patient but there's separate fields in your record and verify system so you're not measuring the exact same thing that you're gonna deliver to the patient.

So what we wanted to do was we used a different method that required a little bit of more upfront work, but then at the end of the day, you are just measuring the actual fields that you're gonna deliver to the patient. So basically, what we're going to do is we're going to manually calculate all those shifts that the array would automatically create in the actual software while we're generating the plan. And I'll show you guys how we're going to do that.

So here's the workflow for calculating shifts. And all this is just distilled into an Excel spreadsheet that we use for each of the MBMS clinical patients that we treat now. And then it works pretty quickly. So what we do for all of our cases is we set all the couch angles to zero. We measure each individual field at a zero-degree couch angle. During commissioning, we did measure with all of the couch kicks, so that's an option as well. These same rules apply and we just don't reset those angles to zero, you don't reset the angles. But since we're measuring for each individual field, the couch angle zero, we set those to zero for all the fields except for the one that we want to measure. We use our cursor, so we calculate dose, and then we use the cursor to determine the coordinates of the dose distribution cloud. And then instead of shifting the phantom to that location, we shift the isocenter away from that location. So we bring the dose cloud on to the isocenter of the phantom. We calculate dose and we record those shifts. When we go to treat it on the treatment machine, we will then move the phantom onto the isodose cloud. And that's where that sort of Rosetta Stone I spoke about earlier comes in handy, figuring out, okay, this is where the dose cloud is located in this calculation module, and then on the treatment machine you'll know which direction to shift the phantom to get into the dose cloud. We reset the isocenter back to zero. That's really important. And then you just repeat this process for each one of your individual fields.

So in terms of measuring dose, for these cases, you're going to get a lot of interesting situations, which you have potentially really high dose rates from [inaudible 00:26:08.235] to three beams. You're measuring those two small fields for small targets. And you can have unusual situations in terms of couch kicks, so you're using your detectors coming in from trajectories that are not standard, non-axial. So whenever possible, you want to try to limit the uncertainty in your measurement, either by using correction factors. There's TRS 483 that has lots of pre-filled size correction factors for all your detectors. Those rates, whenever possible, you just want to calibrate your detector with the dose rate that you're going to be treating with, and then angular dependence of your detectors with your gantry and your couch. That's why we chose to use it at zero for clinical cases to remove some of that uncertainty. And there's various different measurements. There's various publications and presentations regarding measuring dose for these cases. But it's highly advised to read them heavily before you start taking these measurements or speak with other clinical users as well to figure out how they're tackling some of these problems.

So here are some of our just brief, brief overview of our commissioning results. Those were the target size that we measured. Our average mean difference was 0.31%. The max we had was 2.84%. We did a relative dose analysis with the SRS metric and StereoPHAN using a 2%, 1-millimeter criteria. We were able to achieve the mean pass rate about 99%, with a minimum of 95.5%. So we were really happy with that. In terms of our clinical requirements for our clinical cases, we require that our mean dose is within 3% for our plan, and the absolute dose, we're using a 3%/1-millimeter/10% threshold analysis criteria. And so as you can see here, we made it a little bit tighter during commissioning because we wanted to sort of highlight any issues that we'd run into by using that tighter criteria during commissioning so we could solve them, so that things move smoothly with the cases.

Finally, after we did all those measurements, we performed end-to-end tests. Ours was done using the StereoPHAN. They had different module inserts you can use to kind of progress your simulated patient treatment throughout your entire treatment chain. So we took an MR of this insert that had these targets within it, then took a CT. We fused the two contoured on the MR, checked first the CT to make sure the contours are correct. We planned it. We did QA audit. We set it up using the ExacTrac system to position the StereoPHAN. And then we delivered it. And we got excellent results. There are lots of other options besides this to do an end-to-end test. But it is recommended that you do some type of end-to-end test before treatment.

And finally, once you've commissioned your entire system, now you want to start sort of the logistical prep work for clinical cases. So what we did is we got the entire treatment grouped together, [inaudible 00:29:08.823] responsible from billing, clerical, all the way up through treatment. We created a lot of checklists. I'm a big proponent for checklists. They reduce errors, make them consistent, particularly when you're just starting out. So we have checklists for CT, MR, planning, and then our treatment delivery as well. Billing, it's always recommended to include your billers in this process. There's a lot of billing requirements in terms of documentation. And it's a lot easier just to involve from the start, make sure that you check all the boxes and all the documentation is correct than having to go back afterwards to then provide your documentation. Brainlab support, also highly recommend to have the Brainlab trainers onsite for your first case. They were with us from simulation to planning and delivery. They're extremely helpful. It was very good to know you had an expert with you while you're treating your first cases.

This is just a sort of brief checklist overview that I would be happy to distribute to everyone who wants it. It just kind of summarizes what we did in those slides that I just talked about. It can be used as check a good outline. It's by no means the maximum you can do. You can always do more. But this is like a good place to start if you're looking to commission your system. And with that, I'll pass it over to Dr. Seymour. He is the physician who is primarily responsible for the cranial SRS program here. He was heavily involved in all of that commissioning that we did and consulted every step of the way. And he's going to talk to you a little bit about our clinical experience with the system.

Dr. Seymour: So moving on to the clinical implementation of Brainlab with ExacTrac, we initially set up a workflow so that we can get patients onboarded onto the system without any delays. So basically, we created an Outlook group that as soon as I selected a patient or any other of our stereotactic colleagues treating CNS patients, we would send it to the entire work group. So the administrative assistants would block time for the MRI and the CT simulation, the biller would obtain authorization, and physics and therapists would block time as far as time on the machine and time to plan. We worked it so that we would always have at least three slots per week to handle any patients that would be overflow from Gamma Knife as we are transitioning and also just our own clinical buying here at Dearborn.

So basically, if you look over on the left, there's the sim slot. And basically, each one is color-coded. So red would be one patient, green would be a second patient, and blue would be a third patient. And basically, what it would be basically arranged is so that the first patient would be simulated Friday, their MRI would be done Sunday evening, so Monday morning we would be ready to plan, and the first treatment delivery would be Wednesday. And then we did the same thing for the next patients. We always left Friday as a fail-safe day. There is data on both changes in size and volume of your tumor over time. And there's also data on local control reduction as the delays from MRI to treatment delivery, which I've done quite a bit of. But effectively what it is is that there's probably some small changes that a lot of those papers that it says small changes within the first two to three days are probably picking up differences in MRI, so differences in distortion, differences in registration, differences in contouring a cap or not a cap, things like that based off of just how the tumor was imaged.

But there's reduced quality in the plan, meaning increased recurrences, when you go beyond a week. So we always want to make sure that we are MRI within a week of first fraction delivery. As far as the simulation goes, we use the five-piece mask, the only thing to note is just the little bite guard, if a patient has dentures that they need the dentures to come out. Otherwise, I've had my mask made with this. I've also had mask made, just more standard mask made. And I would say overall that this mask is comfortable for a longer treatment. It's really how it's designed. The more rigid components are the ones that are really right up against the face and then really prevent you from rotational translational issues. And it's relatively comfortable. For at least the tight mask, it really needs to be tight. In addition, if you actually look over here on the right, you'll see, you can probably make out that there's a couple of shim bars in between. Basically, if a mask is a little loose fitting, as it cures, it actually will tighten overnight. But actually, you can actually remove some of those bars in case if the mask is not quite as tight as you would like it. So even if, at the time, because actually often the longest time a patient will be in the mask is their first time in it when the mask is being made because this is about an hour-long visit. We actually block an hour and a half of time on our simulator for these cases.

Once you have your MRI image and your CT simulation, we move on to the element workflow. So initially, you review the images, co-register them. In our clinic, we do use a lot of distortion correction now, which I'll go through extensively, but this takes 5 to 10 minutes. Then moving on to contouring, it does the anatomical mapping where everything is auto-segmented. You can select what you want to have auto-segmented or not. For us, we pretty much auto-segment everything we want to ever monitor dose on. So that is all in there. And that's also in our boarding passes in Mosaic. Then we also have the smart pressure where you actually contour the act, when you contour the target, and then any object manipulation for any additional PTV margin you wish to add. And then after that is all done, you move forward with planning with either the multiple brain metastases SRS element or the cranial VMAT element, which takes 10 to 15 minutes. Each plan is relatively fast. It actually only takes maybe two minutes to generate a plan. But each time you do it, an evaluation, whatnot, and that gives you maybe up to about 10 to 15 minutes.

For the image fusion, I do like the image fusion quite a bit. I'm pretty draconian about my fusions in terms of wanting them to not just be accurate but how we evaluate them as I evaluate every single neuroforamina. I evaluate all the chart and the ventricles. So I'm looking at both bony and soft tissue alignment to assess for both the distortion within the MRI that I'm aligning, as well as the alignment of the CT to the MRI. And because there's a rotational selection, it actually allows you to avoid the areas that have the highest risk of distortion for your initial alignment. In addition, there's some additional fine tuning. And there's a number of different ways of evaluating yours, both in basically the color wash overlay, as well as spyglass views. And you can view them in all planes simultaneously so it becomes a very seamless image fusion process.

Following the image fusion, if you have the element, you can apply distortion correction. I'll cave through a few cases because we found this, as soon as we had it available, we started to apply it and we found clear value in this element. This was one of our first cases we had before we had used it. This was a patient that was treated just before the Gamma Knife went down when we had both systems up and running. And this was a 48- year-old female with oligometastatic non-small cell lung cancer. We were able to reach all but one of her targets with Gamma Knife. If you're familiar with Gamma Knife, the issue with the head frame sometimes is reachability of the low cerebellum, particularly if a patient has a relatively short neck. And that was this case. So we were unable to reach the most inferior cerebellar target, and so we transitioned them to ExacTrac to complete the final treatment. We then applied distortion correction. The MRI that we used for Gamma Knife has a lot of quality assurance on it and monitoring of the actual distortion. However, there's only so much that can be done when you then put it the individual patient in the MRI, there's just going to be some degree of distortion any way you cut it.

And in this case, in the low lying-cerebellum, which has a higher area of risk for distortion, we saw almost a millimeter of distortion at the target we were about to treat. And actually, if you actually evaluate it, it actually pulled the dose away from the prior Gamma Knife dose. So it probably actually made the treatment maybe a little bit safer, less likely of crosstalk between those two tumors. Hopefully, she's actually had a complete response to both.

This was another case where this was a patient who was treated with whole brain radiation for asynchronous brain metastases at her initial presentation for cancer. This was about 16 months later, she presented with a pituitary stalk metastasis that was pushing up against the chiasm. She's starting to have vision loss, and we were pushed to treat this target. When we actually evaluated the initial fusion, I had some concerns about distortion within the MRI. So this was before we were utilizing the distortion correction in every case pretty much, and now, we just use this as default, but we evaluate it all obviously to make sure that the distortion correction does not manipulate it in any way that would make it not an ideal image to plan off of, which has only happened in one instance that I can ever think of. It was at one instance where we actually had a patient where they actually had an abnormality in their bone and it actually looked like had the same MRI intensity as brain matter. And it actually created some false distortion correction in that area. However, actually, in the area of the tumor, the distortion correction was correct, it appeared. This was a case where it was clear that there was distortion. It was affecting our ability to align and treat the target we were looking to treat. And actually, if you look at it, so what is drawn on that image is the uncorrected and then what you're actually looking at is the actual corrected MRI. And what you can see is that the chiasm actually pulled back into where our actual treatment dose was. So if we actually did this on corrected, we would have zapped the chiasm. No question we would have violated the TG101 constraints. So we adjusted, we adapted on the corrected MR, and we actually met constraints for both the corrected and uncorrected chiasm. And the patient was able to maintain their vision until they passed from their metastatic disease unrelated to the brain disease.

This was an additional case, another low-lying cerebellar target. And I think this was pretty much the last case before we just effectively adopted it clinically. But you can see in green is the contour for the initial, the corrected GTV, and then you see in the lighter orange the original GTV. And then we see in the DVH, a darker orange, and what we see is that effectively, we were going to miss a little bit. But it was accommodated by the PTV. Okay. So it sort of begs the question of, why are we adding PTV margin in these cases? I think while this is not the classical thinking of why we add planning target volume margin, I think it is undeniable as soon as you start to apply the distortion correction to these cases. And that's effectively what we're doing. And it does sort of call into consideration about our choice of dosing, how do we extrapolate from Gamma Knife doses as we continue to move to a more frameless world.

So, after we evaluated those cases, we retrospectively went back and looked at all the cases which I had treated before we had distortion correction. So we found 20 targets which were treated with the Elements treatment and planning system, 7 were single targets treated with cranial VMAT. There were three additional plans that accounted for the 13 additional targets treated with the multiple brain metastasis stereotactic element. The PTV margins were 0.5 to 2 millimeters. We excluded all resection cavities. And again, I contoured both the initial and the corrected. When we actually evaluated our results, we found the displacement from the target center was a median of 1.12 millimeters, with a maximum deviation of 2.57 linearly speaking. The relative PTV coverage was lower when we applied the corrections. However, still in 18 of the 20 GTV targets, at least 99.5% of the prescription covered the PTV. Conformity indices also increased compared to the uncorrected, which suggested that plan quality was slightly worse.

This is graphically showing all that. Again, PTV minimum doses, much lower as soon as we provided the correction, PTV coverage lower, minimum PTV doses lower, and the higher inverse conformity index. So again, the real question is, how do we extrapolate from Gamma Knife which is basically been the frame systems that sort of driven the data about how we do this? We put on the head frame, we don't add margin, however, with some of those patients, we plan a little loose, or sometimes we'll draw type a plan loose or plan loose but draw tight. But with the distance zoom, it really allows you to measure and gauge how much you really want to treat. So what I actually do, just because I know this question comes up every single time I give a talk about the ExacTrac system, is that how do I prescribe? Effectively in a nutshell, how do I pull all this together is I add margin and I dose reduced to the margin. So let's say if I was going to treat a small target, if I would do 20 to 21 gray with Gamma Knife, I'll treat to the PTV margin with ExacTrac to 18 gray, however, I'll make sure that the target actually gets 20 gray. So it's sort of like I'm having my cake and eat it too, with having a high quality and sharp plan delivered to the margin.

This is some of what Cory alluded to with regards to the templates. We both have templates within Mosaic, templates within our planning system. And anything that we have that will be in our template both for the planning system and for...also the boarding passes we have in Mosaic, we basically auto generate those contours. So everything is there, it is easy to pull off, and this does not take almost any time to auto-segment. You can also import structures drawn elsewhere into Elements. The only time I really do this is if we're ever going to do a cone plan, or peeling off a tumor to do a very small one to do a cone, or if we're doing a resection cavity, because the SmartBrush, if you've seen any of these, they'll go through the SmartBrush. There's nothing that's going to be faster to contour multiple metastases that are usually uniform and spherical. It's also a very good contouring application for well-enhancing benign tumors that don't invade into, like, the bone, for instance. So very good contouring, however, it's just the irregular targets, particularly ones that are non-uniformly enhancing that may be better done with other applications. So I use them for those cases.

When we move them from treatment planning, once you applied any PTV margin you wish to apply, the planning system, particularly this is the cranial VMAT element, it really gives you what you want. It's not like you're having the Jedi mind trick to this system like you do with Pinnacle. It's not like you're gonna have to iteratively adjust one shot and then go back to assess the plan and then adjust another shot and assess the DVH again once you have something that roughly looks good to the eye. So with this, within one to two minutes, it generates...and again, all this stuff is based on the same templates we already have in place. So you are able to generate a good quality plan and iteratively adjust it quickly and actually have direct qualitative measures to evaluate.

This was a case that was done initially. What was done by the physicist is on our left, and what you see there is that the PTV...this was a case where it actually was a combination. There's rapid regrowth within a resection cavity. So it was basically we were treating a large mass basically that was partially resected. So we wanted to get an SIB to the large mass to get that to as high of a dose as we could while still treating with a reasonable dose to the cavity without increasing any risk of necrosis. So in this case, we initially started at 27.5 and 5 fractions as an SIB. And with just tweaking the system a little bit, we were able to actually get even better than that. We were able to achieve 30 gray to the SIB, and we were even able to get lower doses to the closest structure, which was the left optic nerve, because this was in the anterior medial temporal lobe. There's also a number of different back things that you can adjust within this planning system, the modulation, normal tissue sparing, which you see up at the top. At first, we were a little hesitant to adjust that. The physicist was a little worried that we may cause increased QA failures just because we hadn't done it. So, with the first few cases, we've increasingly done more of this modulation. And we really have not seen any changes in our pass rate. Effectively, it's extremely rare for us to not have a plan pass QA. In addition, we were able to actually do the difference between smart and strict restrictions and also pick an organ at risk that is the most important organ at risk. And really, it does just give you what you ask for.

This is another case where we had then used the left optic nerve. Similar location, however, this is actually a bone recurrence in a patient with small cell lung cancer. And she was having headaches related to a tumor in this area, so we treated the recurrent tumor after this patient presented with brain metastases at our initial presentation for small cell lung cancer. And we were able to get a high-quality plan in this area, again, using an SIB to the gross disease within the bone while providing some PTV margin and staying effectively off that entire left optic nerve, so that really sparing her vision. And she had maintained vision until she passed from her systemic disease.

As you're...I'm sure most of you are familiar with, moving to the treatment delivery, just the setup of the ExacTrac system. There's two x-ray tubes which are in the floor and the two panels hanging from the ceiling. In addition, you always have an IR couch which guides the table positioning and allows for the six-degree freedom of couch. The main advantage of this obviously is that you're able to image in almost any couch position and track within that with the IR camera. So this allows us to do, which we take another snapshot, image at least each time we move the couch. Often, at the beginning, we're taking more, I tend to take more, where we take it at the beginning and end just to have a little bit better monitoring to make sure that the patient is comfortable in the mask, not moving around or fidgeting. And we usually, at least institutionally, we limit it so that if it is greater than 0.5 millimeters or 1 degree, then we do shift. Most of my cases, I actually have whittled that down as we become more experienced with the system. I usually use 0.4 millimeters and 0.7 degrees rotation before we would shift.

This is an example of...and we talked a little bit about the cranial VMAT, but this is an example of the multiple brain metastases element. This was one case where we had a couple of smaller tumors and one larger tumor. The physician wanted this case expedited. So, even though we got an MRI done on Tuesday, we still managed to fit first fraction in by Friday. As you can see here, very excellent conformity indices. The gradient index for the smaller tumors goes up. This is exactly what you would expect from a linac-based delivery, basically, which is the opposite of what you would see with a frame-based delivery where you would see instead the conformity get worse with a frame-based delivery, like with Gamma Knife, but the gradient remained good. It's probably similar in the end in terms of effective dose between those types of plans. But ultimately, we ended up with a plan with six arcs and five table positions. On your right, you can see here the arrows showing the front and back with the one arc and all the other arcs just doing unidirectional.

When we then went to the actual QA, we see that the pass rate would be totally greater than 97.5% using relative dose, and it met our 3%/1-millimeter/10% threshold which we use on all of our cases. This is a second case, which is more complicated. This was seven targets, with the largest target being 3.49 ccs and the smallest target being very small, at 0.29 ccs. And the timeline from MRI to treatment is much more longer. Our standard timeline, where MRI done Sundays, Sim done Monday morning, we had a plan done by the end of the morning, and we QA that evening, and that gave us plenty time to be ready for treatment on Wednesday. When we look at the conformity indices, again, conformity indices remain excellent. It's the gradient index that goes up. It's important to note that these were all planned on the 1.5 version of the multiple brain metastases stereotactic element, where we did see more of these higher gradient indexes with the smaller tumors. We then see, again, very good pass rates. Basically, we measure every beam and every target at least once. And the pass rate here was greater than 96.2%, and again the passing rate of 3%/1-millimeter/10% threshold, which we evaluate on all of our cases.

We then evaluated the value added of moving from 1.5 to version 2.0. And what we found is that while the conforming indices were reduced a little bit, the biggest gain was in gradient, where we saw almost half a point improvement in gradient index, and the whole brain reduced by 7%. And you can see here that, yes, there was a little bit of a shift in the conformity where we are shifting a lot of our cases down, or some of them are up more like 1.5, 1.6. And we're really shifting down to the 1.3 inverse Paddick conformity index. However, the gradient is really where we saw large shifts. So we shifted where we had almost no cases where even this really small tumor is getting a great index of six or seven and pretty much everything ended up in the three to five range.

This is a specific case that was hard. It was one of our first cases we had planned in the version 2.0. Now, there's advanced editing features which allow you to put an optimization structure in. That was not the case before. This was a recurrent glioma, which we use the planning system for. So it's a little bit of an outside of the box thinking, basically, where there was two recurrent targets that we wanted to treat, one anterior, one posterior to what was previously treated. So there was some crosstalk previously. However, when we use that optimization structure, we're able to get that crosstalk to almost totally eliminate and there was a 10% reduction in whole rate V12. And this was a case that was treated to 30 gray and 5 fractions for recurrence [inaudible 00:53:46.32], so a much improved multiple metastasis plan with version 2.0.

So, our overall synopsis of our clinical experience with the Brainlab, with ExacTrac, the Brainlab simulations do take more time but they result in a quality mask. It takes at least an hour, be it block, an hour and a half. As soon as we're moving to the planning system, the automated segmentation works. Contouring is fast and efficient, particularly for well-circumscribed targets. Planning times are also equally fast, and we're able to run multiple plans, tweak things without really losing anything and select the optimal plan. The actual QA, particularly on the multiple brain metastases, it gets faster. Our first QA took over six hours to evaluate the plan, evaluate every target, as well as the Winston-Lutz. And currently, we're down to under an hour for any clinical plan we use.

We use checklists throughout, from simulation, to planning, to treatment. And that makes it sure that we have consistent quality plans regardless of the physician, physicist. So it's a lot of redundancy within the system. We evaluate each plan, and our pass rates are greater than 95%. So it's rare that we ever have anything not pass QA. We are always using the 3%/1 millimeter/10% threshold as our metrics to determine a pass. For the cranial VMAT treatments, usually these treatments are pretty darn fast. We're usually done within 25 minutes. It depends a lot on the number of arcs we're using. As we increase the number of arcs, table positions, a lot more therapists going in and out of the room. So when we use the multiple brain metastases element, particularly when you get more complicated plans, it might end up being 30, at worst case, 45 minutes in the mask, I think, is our worst case scenario. But we always block a full hour on the machine so we're never rushing. And, again, we image it, at least, for every couch movement.

The overall pros and cons. The pros, I'd say, clearly this is a very easy to use system which provides you a lot of information where you can really...it lowers that cost of entry into having a quality stereotactic program. The distortion correction is a unique module, which really allows you to better target your tumors. The organ at risk auto contouring is fast and efficient. There is excellent contouring for brain metastases with the SmartBrush. The planning is fast and provides direct informative metrics for planning quality, particularly now with cranial VMAT compared to things like Pinnacle, things like that. There's clearly improved plan quality there, in addition to now improved plan quality with the multiple brain metastasis version 2.0. In addition, the ExacTrac provides a platform that we use for SBRT, as well as stereotactic brain delivery. We use this for spine, as well as using it for a very robust prostate program here. And the delivery speed with a 6-FF delivery, treating multiple targets with the multiple brain metastasis element, it is hard to beat.

The cons, you have to monitor your workflow no matter what type of frameless system you have, and that is not at all unique to Brainlab. And as crossplane comparisons, again, not unique to Brainlab at all. However, it is still a little cumbersome. It is improving. The fact that you have to...if you do want to add additional margin, when you're treating at non-isocenter with a multiple brain metastases element, you do need to then measure to see where it actually generates the isocenter. We usually have some ideas so we usually can ballpark any additional margin you would want to add, which if you do want to do that, that's about a half of a millimeter for every five centimeters. By the time you're more than about that, you're outside the brain usually anyway. In addition, there's no cone element, but to my understanding, that is in the works. If that was, then there would be a lot less need for these crossplane comparisons in our clinic and it would be more seamless. Okay. That was [crosstalk 00:57:53.024] chance to speak.

Bogdan: Thank you very much for all of this.

Dr. Seymour: Thank you.

Bogdan: Thank you, Dr. Seymour, and thank you, Cory, great and very informative presentations. We have lots of questions. And hopefully, we can address some of these. Let's see if we can bring you back online. All right. Maybe, Dr. Seymour, I'll start with some of the clinical questions as soon as...we have quite a lot of technical questions. So then, Cory, I'll ask you. So Dr. Seymour, first question, how many days between MR images and treatments typically at your institution?

Dr. Seymour: I mean, we never go more than a week. It's usually the earliest we scan where at any week it would be Sunday. And they're all being treated by Friday. Usually it ends up being in 3 days, that standard, 3 days, 72 hours between imaging and first fraction delivery. If it goes any longer than that, then it's up to the physician whether or not to add additional margin or re-image. Usually, in my case, I just re-image.

Bogdan: Okay. Regarding ExacTrac utilization, what is your maximum couch tolerance both in translation and rotational deviations that you allow clinically?

Dr. Seymour: I mean, it's whether or not the couch is able to reach there. You can get up to three degrees. It's just reachability of the couch, as long as the image lines up well. I mean, again, like, if we're outside of our couches, we set up the patient. We're not going to deliver a poor quality delivery just because it's not lining up well. We're not going to just loosen our restrictions. Occasionally, we've used it for non-radiosurgical deliveries. And for those cases, we have moved to doing a...in those cases, we have liberalized tolerances a little bit. Like we have one patient right now who has an extremely large meningioma and had a lot of rotational issues. So we just used it. So we just moved over to using the ExacTrac system, which actually made things a lot better as far as our alignment. But for that case, we did liberalize a little bit about the rotational issues because it really won't make a difference about coverage. But, yeah, it's all within what the couch can reach.

Bogdan: For your typical case...and the question was, actually, specifically, to a six months case, but how long is the treatment time with ExacTrac?

Dr. Seymour: That is an arc thing. It matters on the location, number of arcs, and tail positions. Each time you do a tail position, that's adding a few minutes on your treatment because they have to go in and adjust the table. So that takes longer than anything else we do, the snap verification movement. And then repeat snap verification of the couch position doesn't take almost any time at all. It really is the number of times that they have to go in the room. So that's why the multiple metastases cases sometimes can get longer. But usually, I mean, if it ends up being a relatively simple case, then you're maybe looking at 20 to 25 minutes. Again, if you get more metastases, the time does go up. Sometimes it ends up being that if they're not located right next to each other, you do have one that ends up far away from the others, you peel that one off, you do a cone, that cone will take 20 minutes. And then the multiple metastases, if otherwise they conglomerate close together, then you're probably having reduced number of arcs, table positions, so maybe that's 20 minutes.

Again, we usually don't like going more than 40 minutes in a mask, in general. So usually, we would just treat that on a separate day. So if we treat them on Wednesday for the multiple metastases, then we'd come back and probably treat them for the cone a second day.

Bogdan: What is the maximum number of mets do you treat in one session? And what is the V10 value that you typically allow?

Dr. Seymour: Usually, I mean, we don't necessarily look at V10 that much in our clinic. We're usually looking at V12 and we're looking at mean brain dose. We don't want the mean to approach the mean brain at single fraction. So that's usually what's driving it. It matters a lot on the number of mets, the size, location. I don't have a firm threshold in my head. I think the most we treated is 10 in any given case, which is not so dissimilar from Gamma Knife. In general, for those cases with numerous mets, I tend to actually...like just because we do have both systems up and running now. So those ones are the ones I would tend to push to Gamma Knife, if I have that option. I don't necessarily have like a...it's not like I won't treat. Again, it's like you may actually end up deciding to treat multiple days with the same treatment instead of just doing a single fraction, if you're really worried that your V10 or V12 is going too high. It's all relative. I mean, most of this is driven by the large dominant tumors. That's usually what we're actually considering necrosis risk. So it's driven by the largest tumor. So if you have a bunch of these really small ones, there's some more anecdotal data on...some Japanese physicians I've known have said like the 12 jewels from the Gamma Knife or things like that for a single fraction, which roughly equates the three gray mean brain dose. We just never want to approach that. If that's the case, then we'll break it up over multiple days with either system.

To be honest, I view the systems as being each one has their positives and negatives a little bit. But overall, they're pretty much similar. There isn't a huge difference between these. It's just how you go about skinning the cat and developing a quality stereotactic delivery. If they do have a lot of tumors and I will push them to Gamma Knife, mostly because I never wanted to get the...my concern would always be that what if I saw more tumors than I wanted to treat with what I saw on the planning MR, and then what am I going to do? So if there are already going to be in a head frame, if they have 10 tumors and the number goes to 15, then I'd still feel really comfortable with Gamma Knife. With Brainlab, then I start to go, "How am I going to do this?" and I started to think a little bit more about the workflow. But that's a rare exception. I mean, how many patients do I treat like that a year? One, two. It's not that many. We're usually transitioning the whole brain. There really aren't that many unless they've already had previous radiation.

Bogdan: We have two questions more regarding patient selection here. So what is the minimum tumor volume before you decide to treat a metastasis, and I guess in terms of dose heterogeneity within the tumor, what values do you accept? Obviously, you have a lot of Gamma Knife experience, so it's probably going to [inaudible 01:05:26.324].

Dr. Seymour: No, [inaudible 01:05:26.922] for the heterogeneity. There was one paper that came out from Wash U that actually found that heterogeneity was your friend. I honestly don't believe the paper totally. I think they probably just found something that was co-linear with very small tumor volume. It would be the actual, probably the biggest takeaway from that analysis. But they basically found that 34 gray to at least a third of the tumor improved in a single fraction, improved control. But the only time we ever get that with Gamma Knife is with these super small tumors that you have bad conformity. So I don't actually have a limit. It's all about limiting the actual total V12 if you're gonna extrapolate even from the ABM data where eliminating the ABM didn't improve permanent neurological deficit or necrosis. So it's really about your total V12. Not even the V12 outside of the tumor. That didn't improve models. So your tumor is your toxicity. Once you've made that decision to go forward and treat, I'm not going to change my mind just because of slight differences in V12. I'm going to try to make it the most optimal I can. And really, it's always about comparing gradient to gradient once I pick a prescription. If I really don't like it, maybe I go dose reduce a little. But that's usually much more the framework that I work. And I was trying to remember, what was the other part or the second part?

Bogdan: The other part of the question was tumor size. So is there a volume threshold where you actually choose to treat before continuing to observe?

Dr. Seymour: Again, it matters on location. So if it's a small thing in the brainstem, I'm not going to observe. I'm treating that now because I'm on a dose reduce anyway, which we have done with Brainlab system, and that patient tolerated it fine. I mean, there's no real threshold. If it's an eloquent brain, I'm not gonna observe. If it's an ineloquent brain, really, if it's super small, then we often question if it's a met or if it's artifact. So by the time that it's grown, I generally recommend treatment. So we're not going to pick up something smaller than 2 millimeters. If it grows on the next scan, then you confirmed it, so then maybe it's 4 millimeters. Then, by that point, you might as well just go ahead and treat as long as it makes sense for that patient. A lot of this is just putting everything together. It's the context of the case. If the patient has a massive mediastinal thing that causes them symptoms or there are uncontrolled metastasis disease elsewhere that's going to be the cause of their passing, then the percent rating over a 4-millimeter brain met is neither here nor there.

Bogdan: Right. Cory, I have a bunch of questions for you as well. And I'm going to start with size too. So first question is, what is the smallest field size that you measured and commissioned? And maybe and then a practical extension to this, do you guys have a threshold when you consider a cone treatment versus and an MLC treatment, or do you just start with the MLC and check the dosimetry? And I'll let you answer specifically to your setup, which is [inaudible 01:08:43.924] with Agility. But from the Brainlab side, if you don't mind, we do provide scanning measurements that are standard and all the values required in our documentation need to be measured. So what we always require for an agility MLC for the smallest field is to be a 10 by 10 millimeter square field. For the HD 120, we go down to a 5 by 5 millimeter square field. So, Cory, please?

Cory: So that's what we started out with, was following the Brainlab recommendations and measuring down to 10-millimeter field size. Later, we went back, just try to look at the models until to get some improvement on it, and we went down to 5-millimeter measurements, particularly once you get down that small...you've got to spend a lot of extra time setting up your tank, re-checking everything to make sure that you're located in the center of your field. So the amount of time required once you get smaller, and smaller, and smaller just sort of increased exponentially to make sure you're doing the measurements correctly and you're not measuring on the periphery and ending up with a smaller output factor. In terms of the size of the targets that we treat, most of the time, what we'll do is we'll go through the diagnostic MR. We'll measure all the targets and we'll kind of create like a pre-plan of how we think we're going to treat this patient, whether we're going to use multi mets individually, mult