Transcript
Let me begin by acknowledging Mr. Justin Vinci, who actually performed this work as part of his master's research, and also to acknowledge BrainLab for their continued support of this work through our sponsored research agreement.
We all know that radiation oncologists, and especially those that don't come to meetings like this, typically view the treatment plan that they see as the truth, what they see is what they get, that's what they believe. And as physicists, it's part of our job to make sure that that's as close to true as possible. But in reality, we know that the delivered dose is not the same as the planned dose to some extent. And there've been many works, a lot of them presented here today or in this past week, that evaluate the accuracy of ExacTrac, better than 1 millimeter in most cases. But a lot of the ones that we've seen in literature, and some are shown here, did not deliver dose to their phantom, and they looked at the ExacTrac positioning. And really it's the error in the delivered dose is what we really are interested in.
And so we've seen that patient positioning is only one component of accurate dose delivery. And we've seen this week that linac mechanical errors, the planning system modeling calculations, Lutz test, calibration, subjectivity, and ExacTrac all lead to errors that have to be introduced into the delivery process, and a lot of times these aren't accounted for. So the goal of this study was to develop a methodology for an end-to-end test of intracranial radiotherapy. And it uses an anthropomorphic head phantom with a custom dosimetry insert, which I'll show a little more in a second, and we compare the measured and calculated dose distributions within the phantom, and thus we evaluated the quantity of interest that the radiation oncologist would like to see.
So for this, we use the CIRS Model 605 anthropomorphic stereotactic radiosurgery head phantom. We had it customized with a custom cube film insert, it was three pieces, and had some other modifications which I'll describe in a second. And we could measure the sagittal, coronal, and axial anatomic planes by changing the orientation of the insert. So here we have the phantom fully assembled. Here's the phantom disassembled, you can see the insert here closed up. Here's an exploded view of the insert with EDR film inside just to show how everything was situated. And then, obviously, a close-up view of the closed phantom where you can see the film plane passing through. Originally the film contains these fully threaded rods, we didn't use these, we actually had modified rods made, shown in the bottom, which are only threaded on the ends where the locking nuts fit. This allowed in the center section, where these replaced through the film, allowed much more reproducible positioning to hold the film into place.
So the treatment plan we used, we had a simple PTV, 2-centimeter diameter by 2-centimeter long cylinder. We acquired a new CT scan for each block orientation. This way, whenever we did our plan, we can make sure that the film plane we were looking at corresponded exactly to a CT slice. That way when we used the export function for the dose distribution, we were getting a plane to plane comparison. We used seven equally spaced coplanar beams with a delivery of 2 gray to the isocenter.
Initially, the phantom was set up to the room lasers traditionally using marks on the phantom surface, and then we intentionally misaligned the phantom known amounts by using the ExacTrac infrared system. We did nine points, and the offset points corresponded to the center and corners of a 16-millimeter cube about isocenter. So our nominal position was shown as 0 here, which was our nominal setup, where we would normally set up a patient, and then again, the 4 corners, plus or minus 8 millimeters in each anatomic direction.
So we performed ExacTrac alignments until the phantom position was within our acceptance criteria, then we delivered the treatment. This typically required 2 to 3 exposures, 1 to 2 alignment exposures, and then we took a final verification exposure just to make sure everything was at position. We looked at 2 sets of acceptance criteria, 1 millimeter and 1 degree, which is sort of a nominal we expected to be able to achieve, and then 0.4 millimeter and 0.4 degree just to see what we could ask the system to do and whether or not it would give it to us. And then, as I said or mentioned previously, we used the dose export tool in BrainScan to get our planar doses that corresponded to the film plane. And then, looking at the rods that I showed earlier, the solid ones, we actually used these to align the measured and calculated doses. So here on the left, we have the treatment plan and the CT. You can see the easily identifiable rods that we used. Here is one of the film planes we used, and here you can see the holes that we cut in the film. We used some in-house software that allowed us to register a template to the CT scan, get the coordinates off-axis of each of the four points, and then align them in red using their standard alignment software.
So we actually, even though we measured 2D dose distributions, we actually looked at 1D profiles. One of the things this allowed us to do is that the same profile could be measured for multiple independent films. And this kind of...this gave us a way to estimate the measurement uncertainty in the process. And then by doing that, it allowed us then to take this and do some metric comparisons on the dose distributions but based on the 1D profiles. And so here you can see just a sample of the three planes that we took from the film with the overlay of calculated and measured, and then the plans, for example, here left and right profile for this one and this film to compare.
So the metrics we used were determined, again, from comparison of the calculated measured relative dose profiles and orthogonal axes. We looked at two metrics, position alignment error which we denoted as delta c, and this was the displacement between the center points, which we defined as the midpoint of the 70% dose of the measured and calculated profiles. And we chose this because that was actually the highest dose gradient in all three planes because you can see some of the dose spread out because they were coplanar, this actually gave us the best attempt to position the center point. And then we looked at the difference in position of the 80% dose levels, which we called delta 80. And these were more closely associated with clinically acceptable target coverage. We could just as easily have picked delta 85 or 90, but this is what we chose. And without going into detail, previous measurements using the standard methodology determined our measurement accuracy in profiles compared to another dosimeter of 0.2 millimeters and a precision of 0.15 millimeter.
So this is a sample of what we refer to as delta c. At the 70% dose level, we would look at the midpoint from the 2 profiles, and the difference in those midpoints is what we refer to as delta c. For the delta 80 comparisons...just so you know for later, negative values mean that the 80% dose was inside of the calculated, that we didn't get adequate coverage that we expected, and vice versa. So, when you see this later, think negative's bad, positive is good. For the delta 80-1s, those corresponded to the posterior right and inferior edge of the profile, and then delta 80-2 was anterior left and superior. And this kind of gives you sort of a schematic of delta 80-1 is greater than 0, so we have adequate coverage, as opposed to the other side where it's inside of the expected position. So, looking at the results, so the first two measurement sets correspond to the 1-millimeter 1-degree criteria, and the second two measurement sets correspond to the 0.4 millimeter 0.4 degree. And this table's showing the mean delta c plus or minus the uncertainty in the mean, and then below that is given the spread of the distribution. You can see here that other than the AP direction, in general, the uncertainty or the value of the delta c was reduced in all cases, especially here, you can see a 1 millimeter, and then, you know, 0.05 millimeters. However, in all cases, they were within our acceptance criteria, with the exception of these. And the reason the posterior-anterior were a little bit larger is because of the way the star-shaped pattern of the dose distribution fell, you're not actually taking a profile perpendicular to a dose distribution, so there's a little more play in that. And so we were kind of surprised to see that this value really didn't change drastically between the two acceptance criteria.
So we decided to look at what the Lutz position was for those days. And so, what this table represents is the mean delta c plus and minus its uncertainty shown in diamonds with the error bars, and then the dots show the Winston-Lutz error. And that was the three-day average, so the day before, the day of, and the day after. And so you can see that for the most part, with the exception of this one point over here which was still in criteria, because this was a 1 millimeter, it was less than 1-millimeter offset, that for the most part, our results tracked the Lutz error. And what this was telling us is that ExacTrac was giving us what we asked for, we just weren't sure where we wanted it to be in the first place, because the differences between all of these...or with the exception of this one point, are at most about 0.4 millimeters. So indicating to us that by reducing our acceptance criteria from the Winston-Lutz test, which at our institution's 0.75 millimeters or something lower, we would actually improve the agreement to where we expect the ExacTrac to align the phantom.
If we look at the delta 80-1 mean value plus or minus the spread in that distribution...again, remember negative is bad, meaning we're not getting adequate coverage. And so, you can see for here, again, we're seeing a little more of a spread, although in general, we reduced the value of the negative, it's more positive, so it's less bad, I guess if you want to term it that way, but still not within our...and still not meeting the criteria that we expected. And in this case, these were all positive, so there was not really a lot of...much room to improve here. But then in the superior-inferior direction, we actually improved now, we were getting a positive value, which means we were adequately covering our target volume as we expected from the treatment plan. If we look at the delta 80-2 values, again, we see similar results. We see, in this case, it's an increase in the value. And we'll point out that the positive values, in this case, saying that it's adequate coverage is only in terms of the target volume. I mean, obviously, as we increased this dosage, we may not, were there gonna be some critical structures that may get a little higher dose than we expected from the plan, but we focused mainly on target coverage. And again, in this case, even the ones that were formerly negative, when we asked it for the titer criteria, we actually were within tolerance, and covered our dose...or covered our target as we expected.
And so, for...overall, for our patient population or based on the measurements that we took in this phantom study, we wanna know, does the 80% measured include the 80% planned? And so, again, positive means that it was included, negative means that it was not. And so we took the worst case scenario, the worst case of the mean value of delta 80, and then subtracted 2 sigma to see what the probability is that for the majority of our patients that we're gonna be within or we're gonna get adequate dose coverage. You can see here that we didn't meet our criteria for all the cases. There's still some negative values, even with the titer acceptance criteria, which we had hoped would be completely covered. However, it is good to note that by asking ExacTrac to give us a titer criteria, we actually did improve the coverage. So it's still negative, we're still missing by 0.8 millimeters. However, it was over here at 1.3 millimeters. So we've seen about a 0.5 millimeter...anywhere from a 0.5 millimeter to a 1-millimeter improvement in our coverage.
What this ended up forcing us to do is to look at what we were actually measuring and where this caused...the problem would have come from. So it turns out the differences in the calculated and measured delta 80 values are due to 2 things. It's due to the phantom misalignment, so the delta c value, and the TPS, the planning system does calculations in the penumbra. And so, in the bottom, I have a plot here where the solid line shows our calculation 80% level, and the dashed line shows the measured value, indicating that the planning system is calculating a better coverage than we would actually get, and that's actually what we saw from the measurements. And so this led us to realize that if we have a little bit better beam modeling, especially in the penumbra regions, that will get better dose coverage, and will improve, you know...and especially in these distances we're talking about, they're pretty small, so you're talking only tweaking this a few tenths of a millimeter, but it's worth it.
So in conclusion, we've developed an end-to-end test that uses the modified CIRS phantom with film. Some of the limitations...the present work was limited to the central cranium. Now, we do have plans to expand that to other anatomic locations in the head and neck region, and maybe outside of the head, just to see what the results are. Had a precision of measurement of 2 sigma, it's about 0.3 millimeters, so the measurement precision was very good. And it did require a method within the planning system to register the position of the fiducial rods to the dose matrix, which we wrote again from in-house software. So, this method should be useful for evaluating the accuracy of IGRT delivery in the cranium for a number of situations, either different treatment locations and PTV shapes, or different delivery system, different techniques using the same delivery system, or comparisons between institutions participating in inter-institutional protocols.
So based on the results from this, our institution clinical recommendation were that to achieve the delivery accuracy in the present studies, that the isodose error at 80% less than 1.3 millimeters for the 1 millimeter 1-degree criteria, is that we would need to adjust the mMLC beam model to more accurately predict the beam edges and penumbra, and then to decrease our clinical acceptance criteria from 2.5 millimeters and 2.5 degrees to 1 millimeter and 1 degree. Our physicians currently use the 2.5/2.5 criteria, we're trying to convince them that they should drop it down, but, you know, that's a losing battle for us at this point. And then to improve the delivery accuracy such that the isodose error at the 80% is less than 0.8 millimeters or ideally less than 0.6 millimeters, we would have to use a titer acceptance criteria, the 0.4 millimeter and 0.4 degree, and then also take a look at our Lutz test criteria. Can we decrease the acceptance criteria for the Lutz test to 0.5 millimeter, or maybe 0.2 millimeters if possible? You know, obviously, the more we do on that, it makes our results better, but it's a lot more work, and may or may not be achievable depending on mechanical uncertainties of the accelerator.
And so some of the recommendations we had for the vendor, for BrainLab was maybe looking at a virtual isocenter. So, we know the gantry and couch wobble, and mechanical uncertainty is usually reproducible. So we're thinking the BrainLab 6D couch could include automatic offsets as a function of gantry angle and couch angle to account for this. And this way, we produce a very tight virtual isocenter. Now, this would actually provide increased accuracy of the Lutz test and the delivery of the treatment. And then also, we would ask if the vendor would provide access for an end-to-end test, or tools, or to a service to assess the accuracy of the ExacTrac delivery using an end-to-end test in the form that we presented here today, and used in this research.
This is actually incorrect. This was actually just accepted for publication in Med Phys two days ago, so hopefully, it'll be out soon. If you have any questions, take a look at it, please. And if you're really, really bored and you wanna see the entire thesis based on this work, you can visit this website through the LSU, Louisiana State University, Department of Physics and Astronomy. Thank you.
We all know that radiation oncologists, and especially those that don't come to meetings like this, typically view the treatment plan that they see as the truth, what they see is what they get, that's what they believe. And as physicists, it's part of our job to make sure that that's as close to true as possible. But in reality, we know that the delivered dose is not the same as the planned dose to some extent. And there've been many works, a lot of them presented here today or in this past week, that evaluate the accuracy of ExacTrac, better than 1 millimeter in most cases. But a lot of the ones that we've seen in literature, and some are shown here, did not deliver dose to their phantom, and they looked at the ExacTrac positioning. And really it's the error in the delivered dose is what we really are interested in.
And so we've seen that patient positioning is only one component of accurate dose delivery. And we've seen this week that linac mechanical errors, the planning system modeling calculations, Lutz test, calibration, subjectivity, and ExacTrac all lead to errors that have to be introduced into the delivery process, and a lot of times these aren't accounted for. So the goal of this study was to develop a methodology for an end-to-end test of intracranial radiotherapy. And it uses an anthropomorphic head phantom with a custom dosimetry insert, which I'll show a little more in a second, and we compare the measured and calculated dose distributions within the phantom, and thus we evaluated the quantity of interest that the radiation oncologist would like to see.
So for this, we use the CIRS Model 605 anthropomorphic stereotactic radiosurgery head phantom. We had it customized with a custom cube film insert, it was three pieces, and had some other modifications which I'll describe in a second. And we could measure the sagittal, coronal, and axial anatomic planes by changing the orientation of the insert. So here we have the phantom fully assembled. Here's the phantom disassembled, you can see the insert here closed up. Here's an exploded view of the insert with EDR film inside just to show how everything was situated. And then, obviously, a close-up view of the closed phantom where you can see the film plane passing through. Originally the film contains these fully threaded rods, we didn't use these, we actually had modified rods made, shown in the bottom, which are only threaded on the ends where the locking nuts fit. This allowed in the center section, where these replaced through the film, allowed much more reproducible positioning to hold the film into place.
So the treatment plan we used, we had a simple PTV, 2-centimeter diameter by 2-centimeter long cylinder. We acquired a new CT scan for each block orientation. This way, whenever we did our plan, we can make sure that the film plane we were looking at corresponded exactly to a CT slice. That way when we used the export function for the dose distribution, we were getting a plane to plane comparison. We used seven equally spaced coplanar beams with a delivery of 2 gray to the isocenter.
Initially, the phantom was set up to the room lasers traditionally using marks on the phantom surface, and then we intentionally misaligned the phantom known amounts by using the ExacTrac infrared system. We did nine points, and the offset points corresponded to the center and corners of a 16-millimeter cube about isocenter. So our nominal position was shown as 0 here, which was our nominal setup, where we would normally set up a patient, and then again, the 4 corners, plus or minus 8 millimeters in each anatomic direction.
So we performed ExacTrac alignments until the phantom position was within our acceptance criteria, then we delivered the treatment. This typically required 2 to 3 exposures, 1 to 2 alignment exposures, and then we took a final verification exposure just to make sure everything was at position. We looked at 2 sets of acceptance criteria, 1 millimeter and 1 degree, which is sort of a nominal we expected to be able to achieve, and then 0.4 millimeter and 0.4 degree just to see what we could ask the system to do and whether or not it would give it to us. And then, as I said or mentioned previously, we used the dose export tool in BrainScan to get our planar doses that corresponded to the film plane. And then, looking at the rods that I showed earlier, the solid ones, we actually used these to align the measured and calculated doses. So here on the left, we have the treatment plan and the CT. You can see the easily identifiable rods that we used. Here is one of the film planes we used, and here you can see the holes that we cut in the film. We used some in-house software that allowed us to register a template to the CT scan, get the coordinates off-axis of each of the four points, and then align them in red using their standard alignment software.
So we actually, even though we measured 2D dose distributions, we actually looked at 1D profiles. One of the things this allowed us to do is that the same profile could be measured for multiple independent films. And this kind of...this gave us a way to estimate the measurement uncertainty in the process. And then by doing that, it allowed us then to take this and do some metric comparisons on the dose distributions but based on the 1D profiles. And so here you can see just a sample of the three planes that we took from the film with the overlay of calculated and measured, and then the plans, for example, here left and right profile for this one and this film to compare.
So the metrics we used were determined, again, from comparison of the calculated measured relative dose profiles and orthogonal axes. We looked at two metrics, position alignment error which we denoted as delta c, and this was the displacement between the center points, which we defined as the midpoint of the 70% dose of the measured and calculated profiles. And we chose this because that was actually the highest dose gradient in all three planes because you can see some of the dose spread out because they were coplanar, this actually gave us the best attempt to position the center point. And then we looked at the difference in position of the 80% dose levels, which we called delta 80. And these were more closely associated with clinically acceptable target coverage. We could just as easily have picked delta 85 or 90, but this is what we chose. And without going into detail, previous measurements using the standard methodology determined our measurement accuracy in profiles compared to another dosimeter of 0.2 millimeters and a precision of 0.15 millimeter.
So this is a sample of what we refer to as delta c. At the 70% dose level, we would look at the midpoint from the 2 profiles, and the difference in those midpoints is what we refer to as delta c. For the delta 80 comparisons...just so you know for later, negative values mean that the 80% dose was inside of the calculated, that we didn't get adequate coverage that we expected, and vice versa. So, when you see this later, think negative's bad, positive is good. For the delta 80-1s, those corresponded to the posterior right and inferior edge of the profile, and then delta 80-2 was anterior left and superior. And this kind of gives you sort of a schematic of delta 80-1 is greater than 0, so we have adequate coverage, as opposed to the other side where it's inside of the expected position. So, looking at the results, so the first two measurement sets correspond to the 1-millimeter 1-degree criteria, and the second two measurement sets correspond to the 0.4 millimeter 0.4 degree. And this table's showing the mean delta c plus or minus the uncertainty in the mean, and then below that is given the spread of the distribution. You can see here that other than the AP direction, in general, the uncertainty or the value of the delta c was reduced in all cases, especially here, you can see a 1 millimeter, and then, you know, 0.05 millimeters. However, in all cases, they were within our acceptance criteria, with the exception of these. And the reason the posterior-anterior were a little bit larger is because of the way the star-shaped pattern of the dose distribution fell, you're not actually taking a profile perpendicular to a dose distribution, so there's a little more play in that. And so we were kind of surprised to see that this value really didn't change drastically between the two acceptance criteria.
So we decided to look at what the Lutz position was for those days. And so, what this table represents is the mean delta c plus and minus its uncertainty shown in diamonds with the error bars, and then the dots show the Winston-Lutz error. And that was the three-day average, so the day before, the day of, and the day after. And so you can see that for the most part, with the exception of this one point over here which was still in criteria, because this was a 1 millimeter, it was less than 1-millimeter offset, that for the most part, our results tracked the Lutz error. And what this was telling us is that ExacTrac was giving us what we asked for, we just weren't sure where we wanted it to be in the first place, because the differences between all of these...or with the exception of this one point, are at most about 0.4 millimeters. So indicating to us that by reducing our acceptance criteria from the Winston-Lutz test, which at our institution's 0.75 millimeters or something lower, we would actually improve the agreement to where we expect the ExacTrac to align the phantom.
If we look at the delta 80-1 mean value plus or minus the spread in that distribution...again, remember negative is bad, meaning we're not getting adequate coverage. And so, you can see for here, again, we're seeing a little more of a spread, although in general, we reduced the value of the negative, it's more positive, so it's less bad, I guess if you want to term it that way, but still not within our...and still not meeting the criteria that we expected. And in this case, these were all positive, so there was not really a lot of...much room to improve here. But then in the superior-inferior direction, we actually improved now, we were getting a positive value, which means we were adequately covering our target volume as we expected from the treatment plan. If we look at the delta 80-2 values, again, we see similar results. We see, in this case, it's an increase in the value. And we'll point out that the positive values, in this case, saying that it's adequate coverage is only in terms of the target volume. I mean, obviously, as we increased this dosage, we may not, were there gonna be some critical structures that may get a little higher dose than we expected from the plan, but we focused mainly on target coverage. And again, in this case, even the ones that were formerly negative, when we asked it for the titer criteria, we actually were within tolerance, and covered our dose...or covered our target as we expected.
And so, for...overall, for our patient population or based on the measurements that we took in this phantom study, we wanna know, does the 80% measured include the 80% planned? And so, again, positive means that it was included, negative means that it was not. And so we took the worst case scenario, the worst case of the mean value of delta 80, and then subtracted 2 sigma to see what the probability is that for the majority of our patients that we're gonna be within or we're gonna get adequate dose coverage. You can see here that we didn't meet our criteria for all the cases. There's still some negative values, even with the titer acceptance criteria, which we had hoped would be completely covered. However, it is good to note that by asking ExacTrac to give us a titer criteria, we actually did improve the coverage. So it's still negative, we're still missing by 0.8 millimeters. However, it was over here at 1.3 millimeters. So we've seen about a 0.5 millimeter...anywhere from a 0.5 millimeter to a 1-millimeter improvement in our coverage.
What this ended up forcing us to do is to look at what we were actually measuring and where this caused...the problem would have come from. So it turns out the differences in the calculated and measured delta 80 values are due to 2 things. It's due to the phantom misalignment, so the delta c value, and the TPS, the planning system does calculations in the penumbra. And so, in the bottom, I have a plot here where the solid line shows our calculation 80% level, and the dashed line shows the measured value, indicating that the planning system is calculating a better coverage than we would actually get, and that's actually what we saw from the measurements. And so this led us to realize that if we have a little bit better beam modeling, especially in the penumbra regions, that will get better dose coverage, and will improve, you know...and especially in these distances we're talking about, they're pretty small, so you're talking only tweaking this a few tenths of a millimeter, but it's worth it.
So in conclusion, we've developed an end-to-end test that uses the modified CIRS phantom with film. Some of the limitations...the present work was limited to the central cranium. Now, we do have plans to expand that to other anatomic locations in the head and neck region, and maybe outside of the head, just to see what the results are. Had a precision of measurement of 2 sigma, it's about 0.3 millimeters, so the measurement precision was very good. And it did require a method within the planning system to register the position of the fiducial rods to the dose matrix, which we wrote again from in-house software. So, this method should be useful for evaluating the accuracy of IGRT delivery in the cranium for a number of situations, either different treatment locations and PTV shapes, or different delivery system, different techniques using the same delivery system, or comparisons between institutions participating in inter-institutional protocols.
So based on the results from this, our institution clinical recommendation were that to achieve the delivery accuracy in the present studies, that the isodose error at 80% less than 1.3 millimeters for the 1 millimeter 1-degree criteria, is that we would need to adjust the mMLC beam model to more accurately predict the beam edges and penumbra, and then to decrease our clinical acceptance criteria from 2.5 millimeters and 2.5 degrees to 1 millimeter and 1 degree. Our physicians currently use the 2.5/2.5 criteria, we're trying to convince them that they should drop it down, but, you know, that's a losing battle for us at this point. And then to improve the delivery accuracy such that the isodose error at the 80% is less than 0.8 millimeters or ideally less than 0.6 millimeters, we would have to use a titer acceptance criteria, the 0.4 millimeter and 0.4 degree, and then also take a look at our Lutz test criteria. Can we decrease the acceptance criteria for the Lutz test to 0.5 millimeter, or maybe 0.2 millimeters if possible? You know, obviously, the more we do on that, it makes our results better, but it's a lot more work, and may or may not be achievable depending on mechanical uncertainties of the accelerator.
And so some of the recommendations we had for the vendor, for BrainLab was maybe looking at a virtual isocenter. So, we know the gantry and couch wobble, and mechanical uncertainty is usually reproducible. So we're thinking the BrainLab 6D couch could include automatic offsets as a function of gantry angle and couch angle to account for this. And this way, we produce a very tight virtual isocenter. Now, this would actually provide increased accuracy of the Lutz test and the delivery of the treatment. And then also, we would ask if the vendor would provide access for an end-to-end test, or tools, or to a service to assess the accuracy of the ExacTrac delivery using an end-to-end test in the form that we presented here today, and used in this research.
This is actually incorrect. This was actually just accepted for publication in Med Phys two days ago, so hopefully, it'll be out soon. If you have any questions, take a look at it, please. And if you're really, really bored and you wanna see the entire thesis based on this work, you can visit this website through the LSU, Louisiana State University, Department of Physics and Astronomy. Thank you.