SGRT through the Entire Radiation Workflow

Good morning. So as she mentioned, I’m Anton Eagle, a clinical medical physicist at AdventHealth in Colorado. Let me… there we go. And I’m going to be presenting on, as I mentioned, SGRT, our experience doing SGRT throughout the entire radiation workflow. I have no personal disclosures, but our site has PSAs with both Vision RT and Varian Medical Systems.

Okay, so first off, who are we? So who is AdventHealth Parker, and why should you listen to anything we have to say about SGRT? Well, so AdventHealth Parker is a small community hospital in the southeast part of the Denver metro area, and we are the worldwide show site for Vision RT. A lot of you may know this as a site where Mike Thalhammer’s the chief physicist, so I got to give him some credit. And as a result, we get access to pre-production and beta versions, and we help provide feedback for development. So that’s who we are.

So SGRT, you’ve heard the phrase SGRT, every patient, every fraction, and now we can expand that to every step. So we’re really talking about using it in SIM, plan, in the treatment room, and dose verification. So SGRT throughout the entire radiation workflow. So by necessity, since we’re talking about the entire workflow, this is going to be a light touch on all of this. It’s not going to be a deep dive, and inevitably, some of this is going to be redundant with other talks since I’m talking about the entire workflow. But what the goal is here, we’re going to follow the patient or the patient’s data as we go room to room and how we use these technologies. I want to mostly focus on how they work and why we use them and not so much about case studies.

All right, so we’re going to start off in the SIM room. So here we are moving into SIM, and if we’re talking about CT SIM and surface-guided radiation therapy, of course, we’re talking about then SIM RT. Okay, so this is used whenever you need to do something with breath motion with regard to your CT. So of course, that’s going to be 4D CTs or deep inspiration breath holds, so that’s like DIBH breast or… This clicker is not working very well. There we go. Breath hold SBRT.

So how does it work? Well, instead of describing it, I think I’ll just show you. Well, actually, let me describe it first. So we use IR cameras to map the surface, and we select a patch on the surface to monitor, and then we can detect changes in that patch in three dimensions, and we can watch it during breath motion. So what does this look like? Well, so here’s a video of this in action, and I don’t have a mouse, so I can’t run the video. Okay. So that would’ve been a video of this in action.

So why do we use it? Do you want me to back up? Let me try to do that. There we go. Never mind. That would’ve been a video of this in action. So come to the site visit, and you’ll see me using it. So basically, being able to monitor the patch improves tracking. You don’t get any cumbersome blocks to secure, and specifically or significantly faster workflow because of that, and you get excellent 4D results. So get ready with the videos because I’m going to try to show some here, whoever’s running that stuff back there. So instead of just saying it’s good 4Ds, let’s see if we can run these videos that are coming up. So do the first one on the left. Thank you. And as you can see, you don’t get any banding. You don’t get any of that what I call mushroom artifact on the diaphragm. You just get really clean 4Ds. Do the one in the middle, if you don’t mind. Thank you. Yep, and as you can see, very clean, and then go ahead and skip to the third one there, and I think it even shows it better on the sagittal. So this is why we use it, because in the end you just get, I think, the best 4Ds that you can get. So these are about as good as you’re going to expect for 4Ds.

Okay, so let’s move from SIM now to treatment planning. But here we are in Dosimetry, but let’s pause for a minute, and actually, let’s turn back the clock a few years. So let’s go back in time here. So we’re going back in time, and here we go. And now here we are in the ancient and distant past of a few years ago. Okay, so in these ancient times of a few years ago, we actually have an unsolved problem here. We have what I’m going to call a planning gap or an information gap between SIM and treatment. So what do I mean by this information gap?

Well, so if we’re talking about collision detection, we have some missing information here. So we have limited collision modeling because all we can see is the contour of the body, the CT that was acquired. We can’t see anything beyond that. And because we have limited collision modeling, then we’re probably only using limited beam geometry. So most things are coplanar. We have a lot of uncertainty about clearance, and better options probably aren’t even explored. And then lastly, and most importantly, if we do have a problem, we’re often not catching it until treatment or doing a dry run, so at the very end of the workflow. So that’s less than ideal. So all these issues have the same cause. The CTs only see a narrow band of the body, and like for this breast CT, we can’t see the collision risk. So, for example, this arm, we can’t tell, is this arm sticking straight up? Is the elbow sticking out? We don’t know because it’s not in the CT. So how do we solve this?

Well, let’s go from Dosimetry. Let’s go back to the CT SIM room. So here we go. Now we’re going back to SIM, and now let’s bring in some AlignRT cameras or VisionRT cameras. So here we go, bringing those in. And now if we can use these cameras to map the surface, and if we could—if we could map the surface from head to toe, the whole thing, then maybe we could use that surface to check for arc clearance.

So how do we make this work? Well, let’s do a little deeper dive. So here’s our CT and there’s our AlignRT cameras, and just like I mentioned, we’re getting a map of the surface, so that’s good, but we still have a problem here. We don’t actually know where that surface is in the CT space, so it could be anywhere along here like that. So how do we solve that? Well, we use the same technique that we use for a plate cal in AlignRT. So we bring in a plate, and we take the cameras and image it, and we get a point in 3D space. So now that we have that point, we can just throw the plate on the floor, bring the patient back in, and now we can line them up. So that happens because camera space equals CT space now. And obviously I’m glossing over some details here, but that’s sort of the cartoon version.

Okay, so we have our 3D CT space with the surface in there. And now we’re going to send it to some kind of 3D mapping software. And, of course, we’re also sending the CT to the treatment planning system to get an isocenter and a plan. But here we are now in the mapping software. And now we go to the treatment planning system. We send the plan and the isocenter also to this mapping software. And because of that correspondence between the surface space, the camera space, and the CT space, we know where this isocenter is on this surface. So we can do things like say, “Hey, what if I look at an arc going laterally like this?” Okay, that’ll clear. That’s good. But what happens if I try an arc like this? Nope, that’s not good, so don’t do that. All right. So of course, what I’m describing here is MapRT.

Okay, so we acquire the surface at sim, and then again, we move back into dosimetry, and now we can use that surface during planning. So what does this look like in practice? Well, MapRT is set up as a web app. So it looks like this. And this is your typical patient loading screen, so you can load your patient and your plan here. And then on this screen, you can see your 3D surface that you acquired. And then, of course, we can bring in a 3D model of the linear accelerator. These models are provided by VisionRT. They have all sorts of different models. They even have models with the arms out instead of retracted. So there we go. And now we can move into the clearance mapping space. So what this is, is a place where you can explore combinations of gantry and couch angles, and you can see what clears. Just click on the square. There you go. And you can see it’s a live view. You can explore, and you can see exactly what will clear and what won’t clear, and you can use this as a tool for planning. All right, let me click. And you can also see it’ll flag transition or problematic transitions from field to field. And of course, you can also set a buffer for the couch and the patient so you can decide how cautious you want to be in this collision detection.

Okay, so this then, of course, leads to more non-coplanar planning. So let’s talk about that for a second. So we know for a fact that it reduces low dose smear. There’s some legitimate debate about what happens with the high dose regions. So, depending on the plan, it can improve conformity. Some might argue that it doesn’t really improve the high dose regions much. But one thing is absolutely sure is it doesn’t make it any worse, right? So, I’m not going to really delve into plan comparisons, because like I said, there’s legitimate debate there. So instead, I want to focus on times when this kind of technology is absolutely critical.

So here is a three target liver SBRT. And you can see right now we’re focused on the lateral target down there in that coronal view, and then there’s the more medial target. So what I can do is make this one transparent and bring the first one back and make it transparent as well, so we can see them both. And now let’s focus on this coronal view. And you can see that if we use coplanar beam arrangements here, we’re going to have a problem. The dose from one is going to bleed over into another, so that’s not good. So what we really want to do here is some kind of non-coplanar. So we want to do arcs like this or and like this for the first target, and then let’s snuggle up some arcs next to that for the second target like this and this. And that’s the kind of approach we want to use. And this, in fact, is exactly what we did. So here’s what the actual plan was. And we really needed to push these couch kicks to their limit. So we used MapRT proactively to explore this space and really push both the couch kicks and the arcs to their limits. And then, of course, we went to treat, everything cleared, and there was no problem because we were able to use a tool like this ahead of time.

Okay, there are some advanced features with MapRT. You don’t really need to get into this, but they’re there if you want to. Comes with an API. You can do ESAPI scripting to pull these 3D surfaces right into Eclipse. So here’s a script I made, sort of proof of concept, and it tried to mimic that same MapRT interface. And also you can write Python. Mike does Python apps with this stuff. So once you use this API and get this service, you can do whatever you want with it. So the sky’s the limit. Again, you don’t need to use these features from MapRT, but they’re there if you want them.

Okay, so we’re kind of done with dosimetry now. So now let’s move into the Linac Vault. So if we’re in the Linac Vault and talking about SRT, of course, we’re talking about AlignRT. This is the one that started it all. This is sort of the original SRT solution. And as you know, it tracks the patient before and during treatment, assists with localization and mobilization. But what you may not know is it’s now kind of a whole suite of solutions. So instead of focusing on how AlignRT itself works, I really want to focus on these expanded modules. And again, some of this is going to be redundant, but we’re just going to go through them. So we got postural video, respiratory module, OIS module, AutoROI module, and BeamGuide. So I really want to focus on those. So we’ll just go through them one at a time.

Okay, so let’s talk about postural video. This is the oldest one. It’s been around for quite a while. And it assists with patient setup. Basically, it just shows the outlines of all the surfaces, and you can compare and adjust with the patient outlines in real-time. It’s an incredibly useful setup tool. I’m not sure we could operate without it, really. And respiratory module pretty much does what you would expect based on the name. It does phase and amplitude gating, continuous monitoring in six degrees of freedom, device-less delivery. It basically does the same stuff that SimRT does for breath monitoring, but it does it in the linac vault for gating.

And then let’s talk about the OIS module. There’s been quite a few talks on this already, so this will be a little redundant, but this pushes AlignRT reports directly into ARIA. They show up in the offline review timeline, just like an image. And because they show up there, they show up in the doctor’s image list for approval. So they’re really easy to approve, and it’s a pretty cool technology. So here’s a screenshot. This is from our system. This is what our screenshots look like. And you can see it’s just showing up right down here in the offline review timeline. And if we click on the next one, you can see the report that goes with it. So again, pretty cool technology, but it’s not just neat stuff. It actually is useful for supporting some of the billing that we need to do. So using the 77387 as an example, let’s talk about the kind of billing elements that are needed to justify these billing charges.

So some of these elements have always been there, but now we have new access to better support. So let’s talk about clinical rationale. So you have to justify why SGRT was indicated. Well, that was always there. That was in the physician notes or the prescription. And you also have to record the magnitude of the shifts. And so if you have something like offline review, that’s always been there. But you also need to document that SGRT was actually used. And sure, you could do some cumbersome process, but there was no automated way to do this. It was kind of missing. But now it just shows up in offline review. And then lastly, you need to document that the radiation oncologist actually looked at this stuff and approved this. And again, that was sort of missing. But now, again, it’s in offline review, so they just approve it right there. So this kind of tool gives really good support for the SGRT billing that you need to do.

Okay, so let’s move on to the AutoROI module. So this is just protocol-based. You got three to five ROI options per protocol, and it mostly is for introducing consistency and for speeding things up. So I stole these slides from Adee. He did some nice comparisons with left and right-sided breasts, looking at the auto contours compared to manual. And he found that they’re pretty much equivalent, but something like 10 times faster.

Okay, and then lastly, let’s talk about BeamGuide. So BeamGuide overlays the treatment field onto the real-time postural video. So let’s look at an example here. And the white outline is the intended. It’s sort of on like a virtual DICOM surface, so you’re not really seeing the DICOM surface, but the white outline is where the intended is. The orange is projected onto the current surface. So as you can see in this image, they don’t really line up yet, but that’s okay. We don’t expect them to yet because this is a deep inspiration breath hold, and the patient hasn’t held their breath yet. So this is what it looks like before they hold their breath. And now when I click, that’s what it looks like after they hold their breath. And as you can see, they match up much better now. And if I had a mouse, I’d go back and forth a couple of times, but I’m going to just not take the risk. So you can see clearly, though, that this would be a really good assistance in setup.

Instead of using it proactively, you can also use it retrospectively with DoseRT, which I’m going to talk about in a minute. So you can verify the intended dose extent. So this is a DoseRT image for a prone breast. And you can look here and say, “Hey, are we supposed to be irradiating up on the arm like this?” So on the next fraction, you pull up BeamGuide and take a look and see if that was intended. And sure enough, right there, that’s exactly where it’s supposed to go. So that’s actually okay. And then here’s another example showing the same thing, another prone breast, and we can look at the extent in the armpit, and we can see that that’s where it’s supposed to go. So you can use it as a retrospective tool as well.

Okay, so that makes a nice segue into talking about DoseRT. So now we’re on the dose delivery part of the workflow. So how do we see dose on the surface? Well, to do this, we need to use Cherenkov emissions. Okay, so what are those? What are Cherenkov emissions? Well, Cherenkov is the blue glow that you see in the bottom of a pool in a nuclear power plant. We’ve all seen pictures like this. So what causes that? Please click on that video if you don’t… or no, actually, it’s started. Never mind. Just let it go. That’s good. So it’s created when a charged particle transits some medium. The particles often exceed the speed of light in that medium, and so they have to dispense with the extra energy. You can think of it as like a photonic sonic boom.

So Cherenkov emissions also occur whenever human tissue gets radiated during radiation therapy. The problem is the light is pretty dim, so you need a really, really good camera, a really sensitive camera to see these. But we’ve got cameras like that now. So now you can get images like this, where we’re seeing the Cherenkov emissions off the skin in real-time as it’s being delivered. And so if you have really fancy software, you can collect all that imaging data and combine it, and you can make a dose map of what that looks like. And so, of course, this is DoseRT. So this is a cumulative dose map for DoseRT, and it’s pretty cool. You can even see subdermal features. You can see the vasculature and cerumas, that kind of stuff.

All right, so DoseRT. So why do we use this? Well, these are mostly used for detecting problems with setup and planning. Our experience suggests that about 5% to 10% of patients have problems like this, maybe problems with compliance, they’re not doing what you ask them to do, or actual setup problems, or maybe even actual planning issues. The literature, of course, shows similar things. You can do the searches yourself. You’ll find lots of examples. But let’s look at a couple examples here. Some of these will be obvious and maybe some more subtle.

So let’s look at this one to start with. So this is a right breast… I’m sorry, it’s a left breast. And at first glance, it looks pretty darn normal. A little bit down in the arm, but that sometimes happens with the geometry that you need. And then if you go in and play the video, watch carefully at the beginning here. And there we go. See that big flash? So there’s the delivery. So back it up if you don’t mind. Just grab the little slider and back to the beginning and just let it go again. There you go. Just grab it. Just play it again. Don’t worry about it. Just leave it. So did you see—hopefully, you saw the flash at the beginning. And that flash was from an incorrect port film setting. So every time we were taking a port film, we were actually delivering a couple monitor units to the contralateral breast. So all we had to do was go in and correct those port film settings, and then that fixed that right up. So let’s look at another example.

Thankfully, this time not a video. So here is another left-sided breast, and at the beginning here, we can see, okay, we got a little bit of a problem. We’re clipping the chin. Maybe there’s some compliance issues there. But look at, this is most importantly, take a look down here. We’re actually missing breast tissue. The breast is actually sagging outside of the field. So we actually had to go back and redo the plan to account for the fact that the patient wasn’t really laying in the same way that they were during sim. And then here’s another example. So this is a right-sided breast. So this is the lateral field, and okay, a little bit of a problem here. We’re kind of projecting up to the arm, but that sometimes happens. But that’s not the most important one. It’s going through the arm and actually hitting the face. And then if we continue along on the control points, you can see it’s even hitting more of the face there. So that’s a big problem and needs replanning.

Okay, so what this really illustrates is that prior to something like DoseRT, we had an open-loop delivery system. Open loop meaning that we would plan, we would deliver, and then we would just assume that what we were delivering was pretty much what we intended, and we had a blind spot here. Instead, now we have a closed-loop delivery system so we can plan, we can deliver, and now we can verify with Cherenkov imaging, and that gives us the opportunity to adapt. So we can either adapt the setup or we can even adapt all the way back to the plan, and this kind of closes that loop.

So treatment issues. Yep, we can detect these clinically relevant tissue treatment issues with this. But I do want to point out that in order to do that, we need to have some kind of scrutiny. So if you’re going to use a technology like this, you need to set up some kind of daily or weekly review process. So you need to be monitoring this stuff. And then also, as we use this more and more and this becomes more in clinical use, we need to start developing QA systems for this technology as well. The more we ask it to do, the more rigorous QA is going to be needed. We’re going to need things like linearity tests and constancy tests and that kind of thing. Okay, so that is the delivery part of the workflow, and that brings us to the end here. And now we’ve looked at the entire workflow, and we’ve done SGRT, every patient, every fraction, and every step.