 I hope you had a wonderful weekend. Got to see something fun, maybe? Maybe on Sunday, maybe Saturday is a little rainy. Today, yes. Yes, that's right. That's right. OK, so hopefully you see this. Yes, that's right. All right, so today I'm going to skip a lot of some of this because we've had a lot of the same stuff. So actually repetition may be good, so you can really learn it. So some of it will go over very quickly. Basically, Sam did a very good job talking about this. If we have some sort of a target that's funny shaped, if we want to do 3D conformable, we can do a nice uniform dose distribution when we add the beams. But if we want to get a concave shape and shape the dose around some structure, then we really need IMRT, we need some sort of intensity modulation. So for example, in this case, you can modulate intensity with one field. You combine it with other fields. You can have some area in the middle, has low dose, and get high dose around that. Here's an example from a paper. Same thing. They have this C shape, and they have this intensity modulation for multiple beams. You get some nice, beautiful things that you can do that you can't do with just normal 3D conformable. And here's another example showing these are different. This is the intensity map for each beam. And so you have higher intensities, lower intensities. You combine that all the way around with multiple beams. You get this beautiful distribution. All right, so difference between forward and inverse planning. So actually, you can do IMRT with forward planning and inverse planning. And Yakov had some slides there showing kind of this breast plan where he had 110%, 105%, so on and so forth. That would be like a forward plan where you decide everything. You decide which beams you want to use, the geometry, collimators. You decide these MLC positions. And you decide the weighting. You decide everything, and you get your dose. You change it if you want, so on and so forth. Inverse planning, basically, you decide the dose. You say, I want this dose. And now you give me what I need to get that dose. And so you do decide the geometry for those beams, but you do not decide the fluence. You decide what dose you want to give. And it gives you the fluence to get that. All right, so here's some examples of you have one segment, two segment, three segments. You add those together. You get multiple dose levels from multiple segments. And that's in modulated intensities. So you have different intensities in the same beam. So forward-planned IMRT, that can be done by defining the fluence. So you actually have some planning systems where you can draw kind of like a paintbrush, very simply an add fluence, takeaway fluence. And you can do it that way. The other way is you create these multiple subfields. So you have one subfield, two subfields, three subfields, so on and so forth. So these are both ways to do forward-planned IMRT. So inverse-planned IMRT, we already talked about this, basically divided into those beamlets. So one thing to note for inverse planning, for the optimization, so first off, each voxel, the dose to that voxel can be described as the sum of the dose from all of the beamlets. And so each beamlet has some contribution to that voxel. And so you get this kind of linear addition of the beamlet doses to get your dose to that voxel. So often in many of these planning systems, that dose calculation, the dose to this voxel, that is added ahead of time. And then you're just changing the weights so that you can change those rapidly in the optimization algorithm. So it does some calculation ahead of time, and then it just changes the weights to get different doses to try to achieve your dose distribution. So that dose calculation is done up front. And then once again, the criteria that we put in is done in the DVH space. So we have a dose volume histogram, and we define the criteria, and this can be for the target, this can be for the OER. Sam already talked a little bit about this optimization algorithms. There's gradient descent, and there's also simulated annealing. Some methods are faster than others, and then some, for example, simulated annealing is usually maybe a little more robust, especially for when you're looking at the optimizing the machine parameters directly instead of the fluence. Most of those algorithms will use that. So most modern planning systems typically will use the fast optimization algorithms such as gradient descent for like a fluence optimization, for direct machine or direct aperture optimization simulated annealing it's typically used. That's right, yeah. So yeah, what he's saying there is, if you have a gradient descent, you can get caught in this local minima and not make it down to a better solution. And so that can be a problem, especially for this, where you have all these different MLCs. It may find something it thinks is good, but there may be a much better solution. Simulated annealing, it has this kind of statistical noise level that it lowers over time. So it gives you, so even if some, it may find, sorry, it may tell it to come here, which it may be here and it may tell it to come up here and it has some probability of accepting a worse solution. And it may accept that, especially early on in the optimization, so that it can get above these local minima and get down to the global, to the global minimum. Let's see if Sam is a good one to answer that. Monte Carlo. Often you have those separate. That's right, so yeah. So you have a separate algorithm for your optimization and separate algorithm for your dose distribution. Often you can have a planning system, you can select, I want this dose calculation algorithm, it will use the same optimization regardless of what dose calculation you will use. Does that answer that? All right, so once for traditional IMRT where you decide the fluence or the actually the inverse plan, the inverse optimization decides the fluence, then you have to deliver that fluence somehow. So there's a few different ways you can do that. Physical compensators is one way. Multi-leaf collimator motion is another way. And so that can be done with a leaf sequence or this can be created directly. And so in this case, you're actually skipping the fluence step, you are deciding the motion, you can actually calculate what fluence you would get from those parameters, from those segments that was optimized. So this is basically the difference between these two here is which one does the inverse optimization decide? Does it decide the fluence or does it decide the leaf motion? And then you can calculate the other one after the fact. All right, so physical compensators, this is an example, some company that sells these, basically this is something that can be done without having an MLC. You just have some type of a attachment that goes on there and it will basically, you have a primary fluence that comes through and you're attenuating that. So in areas that are thin, you're gonna have a large fluence, areas that are thick, you're gonna have a low fluence. So this is a very simple idea that you can create an intensity modulated field. You would need one of these for every field that you're going to deliver. You can imagine that's a lot of work, that's a lot of overhead just to create this intensity modulated field. You need one of these for every single field, for every patient. And this type, for example, may not be, you can't really reuse this. It's milled for each patient. Here's another example. This is a group that was doing this. They have some, these powder or granule fields. They put in a tin, a little tin powder and they have a styrofoam milled compensator mold. And then so this goes with that and so it creates a different thickness of the compensator as you go across. So this is a, this has some advantages in that this is reusable. You can reuse this powder. You just need a new styrofoam portion and then you can get in modulated field that way. It's something also very unique. Not many places are doing this as very kind of academics, one in kind of an old technique, but it got over that this way you didn't have to have a separate compensator for every field. You just, you could reuse those materials. All right, so some of the advantages of physical compensators, there's no need for the MLCs. They can be delivered static so that you don't have this interplay of the motion of these leaves and your organs if you have, for example, something that's moving and you're moving your leaves, you have some sort of interplay there where some areas can get large dose and others low dose potentially that could potentially be an issue. That's not a problem with this because you have your, your modulated field is delivered all at the same time. Some of the disadvantages is this is not very automated. In each field requires a custom compensator. You need to, you have to go in and out of the room for every patient to put these in, take these out. And then the modulation can be limited for these because these have a thickness, a maximum thickness of your compensator. So you can't attenuate the beam to zero so it's actually a little bit less than that. For, here's an example of that. So max compensator thickness typically is around five centimeters. If you're for a 10 compensator that equates to your minimum fluence you can get through there's about 40%. So that's the minimum, maximum's 100%. So you, so you can't get to zero. Here's an example of a modulated field where this is, this is what the planning system would like. So it would like to get very low fluence here. This is what was, they were able to make physically. So here they were not able to get a zero fluence through here. So there's some fluences that you can't actually deliver using the, using a compensator. Or you, there's gonna be a difference between them, quite a big one in some cases. For some other types of materials, some sort of thicker, higher denser materials that can get a little better. And then, so just some of the ideas behind those, what makes a good compensator, if it gives a good range of modulation like it can attenuate the beam a lot, high spatial resolution, something that's not hazardous, so on and so forth. So here's some, just some of the benefits of different materials for compensators. All right, so most IMRT is done by MLCs. And so as Sam talked about, there's two ways here. One is you have a fluence, you get the leaf sequencing algorithm. The other one is this directly determine the positions of the leaves, so on and so forth. So for the leaf sequencing algorithm, there's many solutions to create the desired fluence. And some of those may or may not be deliverable. And you have to put a lower bound on the transmission that you can deliver. And so when you calculate this leaf sequencing algorithm, it has to take into account the limitations of those leaves, how fast they can move, if they cannot abut or you'll come together or so on and so forth. And so some of these algorithms may optimize the number of segments, the monitor units, the leaf travel time and the tongue and groove effect where you have leaves, they'll try to keep the leaves relatively close together to avoid this tongue and groove effect. The other thing is that very complicated intensities can often lead to more complicated leaf sequences need more segments. So often in your inverse optimization, there'll be an option to smooth that fluence. And so you're not getting some very complicated fluence. And that can help to minimize differences between what the leaf sequencing algorithm gives you and your ideal fluence. So the other thing to keep in mind is that the final dose calculation from the treatment planning system, that may be reported in different ways. That may be the ideal fluence or the final fluence from the leaf sequence. So it's important to know what is your treatment planning system giving you and is that going to be representative? Often, for example, with ARIA, what it gives you is, it gives you during the optimization, it gives you kind of a simplified dose calculation and it's based off the fluence. And then afterwards, it recalculates the dose and it gives you dose based on the leaf sequence and there's gonna be a difference there. So you have to expect that. You may be very close on some criteria and you're very happy you met that. Then you go to recalculate the dose and you say, what happened? And that may be because there's some differences in the dose calculation and in the leaf sequence. All right, so ways to deliver this that we already talked about. Basically, you're just adding up intensities and so this would be a field, different openings of your MLCs as you go along and you get these different intensity levels. So there's two different ways of doing this. One is your leaves can close in, so you have a leaf here, here, deliver this intensity, move your leaf in and then deliver this one. Or this can be a sweeping across the field where your leaves only move in one direction and you can deliver the same fluence either way. So your leaf sequencing algorithm may do one of these two methods and this way obviously works much better if you're doing a dynamic delivery where your leaves are moving the entire time. That way you can deliver those in one motion across going the same direction. So here's just an example from a paper showing the method to create this dose distribution and turn it into a sweeping algorithm where your leaves are just moving in one direction. And one thing to note is this method can work for both a step and shoot method and for a dynamic delivery where it's basically the same process it's just instead of having set positions you have a motion so you're creating a more smooth curve along that path. So here's an example. Here you have this dynamic motion so you have lots of different points and basically you want your leaves to meet these points where here's your beam on time and here's your position of your leaves, the motion's in one direction and then here's your step and shoot. You move the leaves, they stop, they deliver. So basically it's going up and over, up and over whereas this one has just positions that you're meeting as you go across. So Sam already talked about this, we can skip this. Basically here you're specifying a few things, you're disallowing things that can't happen, leaves that, you know, small MU so on and so forth and then you're just getting the positions and then you can calculate the fluence from those leaf positions. So here's the, this is the sum of, this is the fluence that's the sum of the different leaf positions, the different segments. So segments, those can be, these segments, these could be defined by forward planning or inverse planning and if they're from inverse planning those can be derived from a leaf sequence or from DMPO. So you could have these segments and they could be, you know, those could come from many different types of, there's many different ways of getting those. So basically in the treatment planning process we start with simulation so on and so forth. You guys have seen all this, I think, where you basically have this loop where you would, you do your optimization, you calculate your dose and then you can kind of iteratively do this to get your desired dose distribution. I'm gonna skip some of this. Here, this is similar to another one you guys have already seen. Basically this is your optimization window. Here's your dose criteria, your DVH space that has all, this kind of puts these criteria on a DVH graph. This is your objective function and how this changes over time. You have your beam fluence that you can see and your penalty, this is a penalty that you can put in to smooth that fluence. So if this fluence has some hot spots here and you didn't like that, you didn't, you thought that was gonna create a problem with your high monitor units or very complicated leaf sequence. You can put in some, you can increase these numbers and that would penalize any smooth, any, you know, unsmooth dose, or fluence in that field or in those fields. And then also there's an option, this is for ARIA, up here for a normal tissue constraint. So you can say, all right, everything that is not my PTV, the target that I wanna treat, I want to create just, you know, a general criteria to say, I don't want dose there. So this here can help with that where you could just have, in theory you would have, you could have one here for your PTV and you could have one for your normal tissue and you just say just conform to it. But then if you care about other organs and certain numbers, that's where these come in handy. All right, and so then we already talked about this, the differences that you can get in the dose distribution. There's great things you can do, you can create a smiley face, you can create a frowny face, you can do all kinds of things with IMRT that you can't really do with 3D. Yes, yeah, that's a good question. All of these are relative because there's this weighting, right? So you have some numbers that are for the PTV, you have numbers that are for the normal tissue structures and you have, for these structures, you also have this number for the normal tissue objective and you have these numbers for the smoothing. In theory, you could double all of them and you'll be doing the same thing. So relative to, it's a sort of thing, it's kind of hard to tell mathematically it's just a number, right? And you just kind of change it and you see what happens. It has a number here that's, it defaults like 40, 40, 50 or 40, 30. So for X and Y, I typically don't change those. I do change that actually. I typically, well, I'll use like 200 here and I'll use 150 here, which is probably the same as what they're doing here, 180, about the same. Yes, there's some numbers there. There's a couple of parameters. And today in the, when we do the, we'll go over that because I don't have a window showing that, but we can go over that there. Oh, that's, yeah, that's a good, good point. So here we have, here we have this resolution, right? Oh wait, no, that's priority. Yeah, so you see this revolution here and you can change those numbers. The number to look at is there's this volume. So for each one you have the, you have the structure, you have a volume that's in cubic centimeters and then you have number of points. So as you change that resolution, if you make that resolution smaller, your number of points is gonna go up. So if it's an organ that you care about, especially if you care about the hotspots, then you want that number to be fairly high so that you can, it will calculate a lot of points within that structure. If it's a, you know, if it's a structure that you just care that the mean dose is low or something or it's low priority, you can keep the resolution pretty high and it's just gonna calculate a few points within that organ. So that's the number of points that it's calculating within that organ and that's calculating dose for, and it's doing its DVH based on that. So it's the resolution of the DVH for that organ. And so what number is good? If it's bad, it may complain. So here like the numbers range between 1,000 to very, very high for the body. So 4,000. This is not the number of iterations. This is the resolution. So it's like the spatial resolution of this organ and how many points it's calculating dose to within that. So it's not actually doing, normally when you do the dose calculation afterwards, you basically have a uniform dose, maybe one millimeter or two and a half millimeters, something five millimeter spacing. This is actually calculating dose for that organ. In each organ has its own resolution because some organs are gonna be small. If you had a very large dose volume, they may only have one, two points, but it's gonna calculate a lot more than that. Here's your iterations. You have a max time, max iterations. I typically have not had a problem with it going to max iterations. You're gonna be sitting there waiting, waiting, and it's not gonna be, it's already gonna have reached a solution usually by the time you get to that number. But you could put that number in. The other thing here is now newer versions of ARIA here have this kind of intermediate dose calculation. So they actually kind of, they do an intermediate dose calculation to kind of improve your accuracy. And then it kind of re-optimizes based on that. Any other questions? So you're talking about the fluence and whether you can see it? You mean here or? So when you do the image, like when you do a verification afterwards? Well, so that high resolution is in the dose space and it's just for the optimization. Can you get that high resolution you're saying in the, okay, okay, sorry. Okay, so yeah, once again, very beautiful distributions that you can do with IMRT. DVH space, this is just some difference between 3D and IMRT, so this is very similar to the 3D plan we did last time. And here's our IMRT and our 3D for the PTB, spinal cord and larynx. And then this is a comparison with Quantek, which is some papers that look at what's the normal tissue complication probability, so on and so forth. For this one, with changing the resolution mean, what it will mean if you change the resolution there is that you will have more or less number of samples for that organ. So if you could imagine, your mean dose may be pretty accurate. If you're looking at a hotspot or a low spot, minimum dose, maximum dose, those numbers may, those might be in a larger volume or something, so maybe those are gonna get averaged out and not be as accurate. But that's gonna be here in your optimization space. When you go to calculate dose afterwards, you're gonna get a uniform dose matrix. So it's not gonna have this resolution here, it's gonna have a very different resolution that's gonna be uniform. So that is, even if you were to have this not be very, if you had two little points here, you should see that after the fact if you have a high enough dose resolution for your entire grid. 4.5, yeah, for the body. Actually, I have in some cases made that number smaller or actually larger, larger, so less resolution. Just if I run into problems with memory for the computer, if I have a computer problem and some sort of very complicated case, and I am not using a maximum dose constraint for the body, I may make that number smaller, I don't need that high of number. Okay, so if you have, let's say we have a case and they're definitely are cases where you have an IMRT case and a 3D case and they look very similar. Yeah, I would agree, any other comments on that? So he says that because the 3D conformal would be preferred because we're gonna be using less monitoring units, the time of the treatment, and also the work, right, if you're gonna be verifying these fields, everything's more complicated with IMRT. It's a great tool, but it doesn't have to be used every time. There are definitely cases where a 3D case will do just as good. And even if you see, let's say you see a slight difference between IMRT and 3D, is that gonna make a clinical difference? That's a good question. And I think you have to, if you see a clinical difference and a head and neck case is probably as good of an example as any, as where you might, then IMRT makes sense. And if you're not gonna see a clinical difference, then 3D is just fine. It's worked for a long time. And it's a very good technique. Any other questions? Moving on, I like this slide. So this is an IMRT field. You can see the fluence. And this is on the head. So you can see the fluence there, the hair loss, and it's proportional to the fluence. So IMRT works. And you gotta be careful. I mean, this is, if you think about what we're doing here, we're increasing the monitor units a lot. And we have this very complicated shape. And it's a great tool, but it's complicated. And we're increasing the complication to do some things. So it requires a little more safety. It's a tool that just needs to be used strategically. So here's just an example showing, I don't know if prostate's the best example for IMRT, but this is showing how in this case, we have a hotspot for the 3D case. And the hotspot is in the area of the PTB that's overlapping with the organ at risk. So you can, with IMRT, you may be able to shape the areas where your hotspot is to areas that are only within the PTB. So just kind of some subtle changes here that can happen. For example, here's your hotspot, it's going here into the rectum. You can maybe kind of push that hotspot over outside of the organ at risk. So you're still getting the dose, the full dose to the PTB, but the area that's most likely to be near your organ at risk, you can shape that with IMRT. So there's subtle things that you can do with your planning system. And this may require creating a structure, right? Of your, in your optimization algorithm, you can create a structure that is PTB minus rectum, or PTB minus bladder or whatever. And so then you can kind of optimize these and throw some, you know, there's fun things you can play around with to kind of shape your dose distribution the way that you want to. So just some comments on IMRT, it's, you know, this more conformal, I think Yaakov mentioned, that also means it's much better at missing the target if you're not careful. Since you have a conformal dose, you know, if we have a large dose, it doesn't matter how well you set the patient up, you're gonna hit it. But if you have a very, you know, if you have a very conformal dose, you have to be more accurate in setting that patient up. So it's important first to make sure that your margins that you're using when you're, when you're looking at uncertainties, make sure those are correct and then use IMRT. Beam selection is a little non-intuitive in that way we use, you know, more, maybe more beams instead of less and maybe equally spaced. A lot of this is done by the optimization, it's not done by beam selection. So typical monitor units are three to five times higher and there's a tendency to use lower energy to minimize neutrons because we're using higher monitor units for these cases. So there's a tendency to overstress IMRT planning. Just let it do its thing and you don't try to make it, achieve something that it can't do, basically. So some of the advantages, you know, it's able to create these conformal doses. You can escalate the dose if, you know, by shrinking your margins, maybe the dose can go up that you're delivering and you can decrease dose to normal tissues. Disadvantages is it's a little more labor intensive as far as planning goes. Although I think now a lot of modern dose algorithms or treatment planning systems, that's debatable. Sometimes I think a 3D case is actually more work than IMRT because it's doing all the work for you. You know, it does the fluence instead of me manually going through and adding a wedge here, taking one out there. So some of that's debatable, but there is more involved as far as, you know, verification and so on and so forth. The delivery time's a little bit longer. It has the danger of being too conformal potentially if you're not, if you don't adequately adjust your margins. Generally more inhomogeneous dose distribution, although that I think depends on the way you're optimizing. I think you can also get a uniform dose with IMRT if you are not trying to overstress the optimization. And increased, MU equals increased whole body dose and increased room shielding. Here's some references that might be useful. And I think that is everything I have. Thank you.