Hello, this is the part instrumentation module 3A of ASNIC PET curriculum, acquisition and processing for cardiovascular PET. My name is Piotr Slomka, I work at Cedars-Sinai Medical Center. Thank you very much. So let's start. These are my disclosures, research grants, NIH Siemens Medical, Amazon Web Services and Software Royalties Cedars-Sinai Medical Center. Our learning objectives today are to describe the components of PET scanners and what they do, explain differences between dedicated and hybrid imaging systems, identify the quality control requirements needed for high quality PET imaging. So let's start first with the PET acquisition. This is the sort of overall outline of this module. We first start with basic description of PET CT systems and the differences between 2D and three dimensional systems. And then we will go through various topics, as you can see here, and we'll show some clinical examples at the end. So here you see different designs of cardiac or general purpose PET systems and PET CT systems. So here, those systems here, they have both PET component and the CT component. And here is an example of system where the cover is actually taken off. And you can see this in the design here where you see a PET modules, and you will see that in the CTs here in the front. And here, this system, for example, this would be just a dedicated system which has only PET module there. And for attenuation correction, we will not have the CT, but we will have the built in transmission sources. So here, this detector modules from PET, you can see them here. This is the example of what this detector module actually looks like with four photomultipliers and a crystal block. And you can see a bunch of them here arranged in a circle. There is a review here of the sort of hardware design in this seminars of nuclear medicine. So in the dedicated PET systems, when there is no PET CT and we acquire images with maybe a little bit different when we don't have a list mode acquisition, we will have to perform separately dynamic scan and a gated scan. And this will be done in the legacy systems, older systems, which are not able to accommodate the list mode data from which you will be able to do both scans at the same time. In this kind of scenario, you would do a transmission scan starting here. You will do, for example, two and a half minute dynamic scan if it's rubidium. So you will only collect the first two and a half minutes of the data. And then you would switch the mode to gating and you would collect gated data from which you can get perfusion and gating data. So you'll have two separate scans here. And then you will repeat it with stress, typically REX scan and then stress scan. So you'll see the scans will be separated here. And there will be typically in this kind of scenarios, the dynamic scan will be acquired in a two dimensional mode and the perfusion scan will be acquired in a three dimensional mode. This two dimensional mode here is performed because we want to avoid the saturation of the scanner. The later generations of the scanners have some saturations. They will not have saturations. The early generations, if they have 2D and 3D mode, may not allow you to do correct dynamic imaging in 3D mode. We will talk about what 2D mode and 3D mode is in a moment. Just pay attention to this parameter. When you acquire images in the protocol, it's an important parameter. Most of the new PET-CT scanners will be only 3D, but the legacy systems maybe allow you to do acquisition in both modes. If you have a PET-CT system, which is typically just a 3D PET-CT scanner, although there are PET-CT scanners, which are both 2D and 3D, but this would be a PET protocol here where we have both modes. We would perform the CT scan and then we'll perform one, for example, six minute scan, which is enough to cover both the first early dynamic phase and then the perfusion phase where you can do your gated scans. And notice you don't have to separate these acquisitions. You don't have two acquisitions. You'll have one combined acquisition, one file, which simply records events. And this file we'll call list mode file from which then different reconstructions can be performed for the dynamic phase, for the gated phase and perfusion phase. And then this will be repeated for rest and then repeated for stress with a pharmacological stress for rubidium. It's also possible to do this with ammonia and notice also the doses here. So there's some debate about what dose we'll use. And then again, it will depend if it's a 2D system or 3D system, but on a 3D systems, perhaps we can lower down the dose because we'll have more counts and more about that in a second. And the ammonia is an example where you spread the timing a little bit, spread these two scans a little bit more further apart, and you will acquire the PET-CT first, the rest PET, then the stress PET. And you maybe separate them by half an hour or so or more to avoid residual activity from rest and stress because ammonia has a longer half-life than rubidium. Now for oxygen, similarly to rubidium, we can do them very quickly, just six minutes. The decay time is very fast and we'll perform it like this. And the whole scan can be done within 30 minutes, stress and rest together. So when we acquire PET data, we will typically have different types of events. So let me quickly explain this events, types of events in this graphs. True events is really what we want to get. It's where there's an inhalation event and there's two of positron, positron annihilates and then emits two photons exactly 180 degrees from each other with a little bit of deviation, but pretty much along this line. And that allows us to detect this event in coincidence with the detector here on the left and detector on the right can detect events exactly at the same time. That tells us that there was some kind of activity on this line. And as you can imagine, if you collect many of these lines, you will then reconstruct the image. So you don't need to collimate these events because you collimate by the virtue of the fact that there is this coincidence and you notice which detectors actually picked up this events exactly at the same time. But there could be some events which are not desirable and they can also be occurring. So for example, scatter event is something where there is a true activity, for example, in a myocardium here. And then so the photon will be going out in this direction. So it will be detected by some other crystal here, other detector. But unfortunately, that photon would scatter and will deflect at some angle and will be detected here. So now this orange line tells us that there's some event here. There's some activity here on the orange line where this is really not true. So you want to minimize this, but there will be those kind of things. And there's also random events and these events can occur where two totally independent sources in two different places, for example, somewhere in the lungs and in the myocardium can occur at exactly the same time. And they will mimic coincidence on the orange line. So the two separate events would not give you any information on this line. If they occur exactly at the same time, they will actually trigger this, you know, this will add this line to the reconstruction, to the image. And the last one, which also is detrimental, is that attenuation effect where, you know, one of these photons simply disappears. So it doesn't change in scatter, the change direction just will disappear. And then, of course, there is no coincidence and we don't detect it. So those are the kind of events. And we really want to get true events and we don't want these other events. This is what we have in PET imaging. So now when we go to this 2D and 3D PET, you can see that there is quite a bit of, there's a difference in how the events are collected. If this axis here, like along this axis, this will be, imagine the patient is positioned sort of horizontally here. And in 2D mode, we have this blue lines, which are septa, like lead septa, which will not allow activity to be detected in arbitrary directions. They will be only acquired in particular planes like this, this red lines, for example. And sometimes we do allow, you know, some kind of, you know, it's not exactly in the same direction, could be a few slices off, but it's still considered 2D mode. But the black lines would not be acquired. They will just be blocked by the septa. So the good thing about it is that it reduces the scatter, as you've seen on this previous slide, there could be more chances for scatter. But the bad thing is that it could be many true events which we'll just not collect. So our sensitivity will be low, but maybe we will have less scatter. It's also easier to reconstruct on the old systems that 2D data will have less photons to process, will be easier to reconstruct. In the 3D imaging, we actually can detect many directions. And of course, this can be limited, but the more directions we allow, the more sensitivity we have for the activity inside. But as you can imagine, this can lead to some saturation. So that's why on this old scanner, sometimes we acquire in 2D mode the dynamic phase when there's a lot of activity concentrated in rubidium, for example, to avoid to have too many photons. But generally speaking, this is the sort of a new way of doing things. And we like 3D mode because we can extract more photons from the same acquisition. And then perhaps we can correct for this. So detection in 3D imaging, we detect events across all the detector rings. And of course, the more rings and the longer the detector is, the more we can detect. And you heard probably about the whole body PET imaging where you can really have the very long field of view. You can detect a lot more photons. But there will be more scatter and it will be much more computationally demanding. So the reconstruction will take more time. Most current hybrid PET-CT systems are 3D. So when you acquire data in 2D versus 3D, as we said, sensitivity is lower here. And we will have this kind of a profile of acquisition in terms of sensitivity. So depending where in the image you are, at which slice, you have a little bit of variation, but it will be pretty much the same sensitivity across each slice because you're just detecting photons in each slice. However, if you work in 3D imaging, you can imagine that if you are at the edge of the scanner, your sensitivity will be lower because a lot of photons will be coming out this way and will not be captured. So this mode will be, this sensitivity will vary depending where you are in terms of the scanner. This can be corrected for, of course. So that was an overview of how you do the 2D versus 3D acquisition. So now when we talk about this older scanners and new scanners, we have two different ways of acquiring the data. And the old way, sort of traditional way, was that when we acquire the data, the system doesn't store all the events, but it will actually store, will prepare the events, will already kind of put these lines in the form of sinograms. So there is no information. So this is a little bit like inspect as well. Data is already kind of put into the projection mode and tells you how many, all these events are added together and they just need to be reconstructed. There is no information in such data when the individual photons were detected and there is no possibility of adjusting the reconstruction if something bad happens. So for example, if there is a different gating, there's a gating error, you cannot change the gate and reconstruct again and so forth. So or for example, you couldn't adjust and select certain timing of the acquisition. So but if you do it this way, you have less storage because the list mode data, which we will go in a second, will contain every single event stored on a computer. And then after the acquisition, you can play with it and reconstruct every possible way. Whereas this would be already sort of combined and aggregated for you to reconstruct from the sinograms. All systems will not have this mode data. Most current systems always have this mode data. So when you go to the list mode data, basically, it's in a format, it's a different format, it's not an image, but it will store events. So it will tell you when the time, for example, in 50 milliseconds intervals or, you know, like the bin of the time is the smallest time it records and when the pet event occurs. And the pet event will tell you which detectors were activated. And then you could also record the ECG signal, you know, at what point of ECG it was, because you can store it together at this time and you can have, you know, again, pet events. So that's why it's called the list mode, because the list of events. So if you can imagine, if you have a lot of these events, it's going to be a very big file and these files can be easily up to two, three gigabytes. And they are usually not routinely stored. However, if you want to go back and re-reconstruct it, you have to store it somewhere and then bring it back to the scanners. So this list mode data will contain pet events and also will contain true events and random events. So you can also apply certain corrections and optimize your reconstructions after you acquire the data. So as we said, a lot of data is required. I've never seen them less than one gigabyte. I've seen them two or three gigabytes. However, yeah, you can do, for example, if you want to do 16 gates versus four gates reconstruction, you can do that afterwards. You can decide that you want different framing for your dynamics scans. So and also there is a lot of work, and I'll show you something of this, that you can actually do motion correction using this list mode data to sort of find out, for example, respiratory phases from the list mode data itself and then correct for motion after the acquisition. So let's move on to the little bit more details on how we acquire images and the point spread functions which you may have heard about. So when the images are acquired in a PET scanner, there is distortion caused by a circular tomograph because as you can imagine when we talk about these lines of responses, if these lines of responses occur at sort of different angles, there is sort of maybe suboptimal detection by these detectors. So it actually happens that depending where your photon is, and like my red pointer here would be like, let's say the positron, and then if they're coming from this direction and they're coming from this direction, it may produce a different kind of a response. So if the images are in the middle of the scanner, they usually give you like a better response like this, a thinner, we call it a full width at half maximum response, sorry, full width at half maximum will be like the value here in the middle of this peak. And this is called point spread function. So like if you have one point, if you had point source exactly at that point, the actual image would look like that, a cross section of this. And if the point source was here, the image would look like this. So it's not optimal and it can be corrected for. One of the techniques, so many of the reconstructions have built in correction for that. And one way to do this is for a particular scanner model to record at each position, X, Y, and Z, you can record this shape of this curve. And then knowing where it comes in the reconstruction, you can just apply and invert it. So sort of filter it more inversely in this location and less in this direction to do something which we call resolution recovery. And so vendors, for example, in this example from this publication by Panin, they show how you can, using a robot, you can move a line source in different directions and record many of this point spread function positions and then store it as a kind of a model of the scanner, which is then used during the reconstruction to correct for that effect. So effect on clinical data is like this, for example, we typically do use this resolution recovery often and effect is pretty dramatic. So for example, here there are images, you know, when you see with and without resolution recovery and you can see that your contrast to contrast to so this in this case is called HD PET is the resolution recovery. Those are the different two dimensional, three dimensional reconstructions. And you can see the contrast to noise ratio goes up by about, you know, from 10 to 20 for perfusion is from four to seven. And for static images is a little less, maybe a little less effect. But static images and negated images. And here is the contrast. So contrast increases, but contrast to noise ratio is also dramatically increasing. So this is one of the examples of how resolution recovery can improve your images. The other important aspect of PET acquisition is something you may have heard of is time of flight correction. And that is available on some of the new scanners. It's not available on all of the scanners. So you have to always know if you actually have it as a function. But as I told you about this detection in coincidence that these two photons are detected exactly at the same time at these two ends, well, actually light travels at a finite speed. And it's not exactly, you know, when events happen, they will not come exactly at the same time to those two ends. So knowing the difference, you know, when, you know, assuming they come almost at the same time, but the difference, there is still some detectable difference. You can figure out where on the line this event actually occurred. And that can give you, you can't do it very accurately, of course, because photons move very fast, but you can do it depending on your scanner resolution. You can do it within a few centimeters or sometimes even better. And as you can imagine, that reduces the noise because suddenly now in typical reconstruction without time of flight, you just know that the event happens here somewhere on the line of the response, but you have no clue if it happened here or if it happened here. Whereas if you actually include this difference in time of flight, the difference in timing, you can actually pinpoint a little bit closer to say that this event actually occurred somewhere here. That will help you reduce the noise. And sometimes this resolution recovery and time of flight, they go together very well. And here are the example of oncological examples where this time of flight is used for acquisition. And you can see the images, you know, look a little better, less noisy. So in cardiac imaging, we have applied this time of flight and resolution recovery together on coronary imaging. So coronary imaging is the special PET imaging where you look at this tracer optics in coronaries, in the plaques of the coronaries. And here really we're pushing the PET possibilities, the capabilities to the limits because we really want the small plaques in the coronaries, which are also moving, visualize them very well. And you can see the difference how this time of flight allows you to, and resolution recovery allows you to create much better reconstructions with the improved, dramatically improved signal to noise ratio and so forth. So that's like a positive plaque in the right coronary artery here. This is a publication from Journal of Nucleic Archeology, which illustrates this application. Other papers on this for perfusion imaging were demonstrating that this resolution recovery, which is this PSF system and time of flight with this PSF, you can see how much more improvements in the contrast. In the contrast, it's much higher, just combining the both techniques together, it gives you much more contrast in the image. And you can see these examples here, clinical examples where you can see these images are, you know, much clearer, less noise and crisper when you apply these two techniques. This is another publication from Journal of Nucleic Archeology, Tamiyama. So now let's talk a little bit about the gating and during the reconstruction. So when we have gated signal in a cardiac imaging, we often have ECG leads on the patient and we collect the ECG signal and that can allow us to do two things. We can allow us to do, for example, bin the images into systolic volume, systolic phase and diastolic phase. We can derive systolic volumes, diastolic volumes. We can derive ejection fractions and we can also look at the contractile, the synchrony when we have these images. So the images, so typically all our images typically are acquired now with gatings. They're always the signal is there. And if it's a list mode, you can just choose to use it or not use it. So those are the sort of review publications on this. And you can see here in this example that you have the image, sort of moving image, and you see that this image was reconstructed from the list mode into eight gates and then you can follow this gate. So each bin, so we have here eight bins and it's just repeatedly moving. It's not like in echo that this is sort of real time. It's just average over the whole five minutes or so. But you can get this average systolic phase and the diastolic phase and you can create this kind of images using ECG gating. So we can also, but the patient, there is cardiac gating, but the patient is also breathing. So, for example, during the typical PET scan over five, 10 minutes, the patient can actually be also moving the chest due to the breathing motion. And you can see here the liver will be moving and the actual, the myocardial, you know, the myocardium will be moving because of that, because the lungs are contracting and expanding and the differences are not insignificant here. Can be one centimeter or one and a half centimeter. So this is the kind of motion you also have during cardiac scan, which degrades your images. And here you can kind of see it, you know, on the actual images where you will see now from the same Lismo data, we now didn't gate it into the cardiac bins, but we gate it into the respiratory phases. And so for each phase, you have a full cardiac phase, but you actually have the respiratory cycle phase. And you can see the heart is moving up and down and it's moving like that. So this obviously is going to affect your quality. It is possible to correct for this if you have a Lismo data and you can make this kind of reconstructions from the Lismo data in PET on the new scanners. And you do not require necessarily to have any external signal like ECG to do that. You can derive this directly from the data itself. So there's different solutions for that. So when you have this respiratory amplitude, this can be derived externally, but it can be also derived from the Lismo data, as I said. And then you can divide it either by this sort of like just by the equal timing. But as you see here, if you do it by equal timing, you may find that some locations are moving a lot. For example, here, the patient can be moving a lot in this blue frame. So you could do it also by amplitude like that. And then when you do this this way, you will have less motion in each phase. So that would be one possibility. And this can be done, as I said, directly from the Lismo data. Examples of this kind of respiratory motion correction, uncorrected image, corrected image published by Lassen in JNC and another publication called Beach in General Medical Medicine, where you see uncorrected and respiratory corrected. So those are the artifacts due to the respiratory correction, not cardiac motion, but due to respiratory motion. And they can be corrected from the Lismo data. This will be the advantage of acquiring Lismo data where such corrections can be applied after the acquisition. We have come up with the technique where you can correct both cardiac gating and respiratory gating. And we do something called double gating where you can then correct images, double correct for cardiac and respiratory motion to produce the most optimal quality for cardiac imaging. And as you can see here, in this kind of examples, we find that this dual motion in frozen was increasing contrast to noise ratio and contrast dramatically compared to standard images. Here's an example of this dual motion technique. We call it motion frozen technique, but it's basically motion correction. So as you start gating these images into the various small phases, because you now have eight cardiac phases and four, maybe four respiratory phases, you, each of these phases actually contains only 3% of the counts from the whole image. Ungated image would just add everything together without any consideration for the motion. Then we can perform corrections for the cardiac motion by sort of registering the images from the cardiac reconstruction. This cardiac reconstruction is performed just using ECG, just like we would do for gated images. This image just collects a particular aspect of the respiratory gait. So we'll collect only one respiratory gait. As you see, this is more noisy because it still doesn't include all the counts. This image now does all the corrections, the cardiac correction and respiratory correction. And you can see how progressively the quality improves here. So those kinds of techniques are available and possible for standard cardiac images. So let's talk a little bit about attenuation correction now, because that's a very important effect, which attenuation of FODM is an important effect and we need to correct for that. So in the traditional scanners, there is no CT, we will have some transmission scan, which will make this kind of a poor man's CT here, transmission cynogram. You see this kind of images where they will represent the attenuation through the body. The advantage of this system is that it's together with where the heart is, it's not separately somewhere else. There's less registration issues, but the quality is very poor, there's noise. However, the images after correction, you see, you can still correct it by, there's areas in the heart which are sort of more affected by the attenuation. And here, this is being corrected by this attenuation maps where you just basically apply correction factor to know that there's less photons coming out of that. It's actually a very severe effect in PET, it's much more severe than in SPECT. So it really has to be corrected in PET data. During the PET-CT acquisition, we will have a CT and PET scan and they are not necessarily aligned with each other. And because of that, you have a CT here and you'll have a PET scan here. And because of that, you will have to register it. So there is some known shift, but also patient moves during that process. These images are not acquired at the same time and you will have artifacts. And also because the CT images are very high quality, they can be usually acquired just in one respiratory phase, for example, during the inspiration or expiration. And then they may not match exactly how you acquired your PET scan because that one was acquired over many respiratory cycles. So the advantage is the images are better quality, but disadvantage is they don't match quite the way the PET scan was done. However, these corrections are important and you can see here, those are the images, corrected images, uncorrected images, and you sort of put it all together and have your emission scan corrected using this CT image from which you obtain this attenuation map. However, as you can see, the CT scan can be moving. So if it's moving during and you have your attenuation scan not exactly acquired the same position as your heart, it may be misaligned. So people propose to make this kind of average CT where you move it, during the CT scan, you actually get it for respiration, but then as the dose is not routinely used. But you need to find some ways that your CT scan is perfectly aligned with your pet. If it is misaligned, it causes, there's a bigger error possibly. So this is one publication by Gould from Newcomers in 2007, Journal of Newcomers in 2007 shows examples of this kind of misalignments, which can be then corrected after sort of shifting the CT map where it's supposed to be. This can be also done automatically. There's software for this, but you can see that if this kind of map is used for attenuation, there will be an error because it will suggest that there is no tissue here with different and will apply a different correction factor in this areas, which will under-correct the image. We have done a study in, which was published in Journal of Nuclear Cardiology, 2015. And we've shown that about half of the cases, the half of the cases actually needed alignment on stress and rest. And we found that if you do automatic registration, you will actually, you know, find that there is a very little difference after that, what actually needs to be corrected. But if you don't do it, a lot of cases will be judged, here is judged by experienced radiologists as mild, moderate misalignment or severe misalignment. So any of this can affect, any of this misalignment can affect the perfusion results and particular perfusion results. So techniques are applied, they have been developed by vendors to provide a fully automated registration. And you can see here examples of like what it comes from the scanner and you see this clear misalignment here. And this is using this rigid automatic techniques. The software aligns it, gives the exact translations in X, Y, and Z. This is then fed into the reconstruction after this alignment to correct for this. And as you can see, this effects is pretty significant. It can mimic ischemia because sometimes you can see this as stress, but at rest there will be no misalignment. So it can occur just on, for example, on stress scan. And then you will see images like this, whereas after alignment, if you perform alignment is the same patient, same scan, you see the image is normalized and it doesn't have a defect anymore here as it's shown here on the polar map. So you have to be very aware of this. This has to be always part of your quality control when you do PET-CT scanning. We have shown that this, if you don't do any alignment, this is our data, Rubidium PET-CT cases, stress Rubidium PET-CT for detection of obstructive coronary disease. In 171 cases, we've shown if you don't do any alignment, your AUC is 0.81. Manual alignment can get you better result, but automatic alignment gives you better results. And when you actually visually check what needs to be aligned and then align only those cases automatically, then you get the best results. So the moral of this is that you have to check your alignment and you should align. Ideally, you should check with your suppliers for your PET scanner that you have some automatic mode for PET-CT misregistration alignment. If you don't have it, you at least should do a manual alignment. So just going into the calculation of myocardial blood flow now, which is sort of the sort of ultimate thing you can do with cardiac PET from this first two minutes scans. We typically have those isotopes now. And I think right now, most of the labs in the United States will be using Rubidium. Some labs use ammonia and water is technically difficult to do. I don't know if there's many places doing it. In US, there's some places in Europe doing it clinically. And now new tracers coming up, which is not yet FDA approved, but soon will be, which may be easier to acquire. So when you look at this table, a few things are important to notice here is that Rubidium is the most commonly used. It has a certain lower resolution because the images, first of all, the images, the positron range for Rubidium is high. So that means that before it annihilates, it travels a little more in the body. The dose, it's created in a generator. You have to do it during the acquisition to make it because the half-life is very short. It's 1.27 minutes. Ammonia half-life is 10 minutes, more manageable, but you still, then you need to have cyclotron on site. Although in our center, for example, we obtain ammonia from a cyclotron, which was 20 minutes away by driving from our hospital. So it's possible, but it's very difficult. Here, of course, you need to have a cyclotron, but for this new agent, which is not yet available, it may be possible to have it on cyclotron, which is off-site and just delivered like F18 FDG to the hospital. And then you can image the scan, you can have the images acquired for a little longer. So those are the kind of different, this is now phase trial is completed now. So right now you have, pretty much you have this two tracer here and it's mostly Rubidium. So for dynamic flow. So when you do the dynamic flow, you will focus, this will be sort of focusing the acquisition on this first phases where the tracer is getting from the blood into the myocardium. And so we will measure the tracer in the blood, in the arteries, and then we'll try to measure and recalculate the rate at which the tracer is taken up by the myocardium. And then that tells us about certain physiological characteristics of a patient, which is very crucial, which cannot be measured by perfusion. So for this, we typically use something called compartmental modeling, where we try to find, first of all, we want to see what fraction of the activity from the blood gets, is taken by the myocardium. Then we can calculate the uptake rate, we call it K1. But because there is this varying extraction fraction, we have to sort of correct it back and figure out what actual flow is to the myocardium. And then there could be something, like a washout fraction where some of this blood, the tracer goes back from the myocardium back into the tissue. And then, so we will have something called K2, which is the factor at which rate this tracer goes back to the tissue. Different tracers will have different K1 and K2. This is like a very simple model. There are more complicated models. So when you have this myocardial blood flow images, you see here, they will, you know, this is in a dynamic mode. You'll have these different phases being sort of recorded. And this is all again, reconstructed from the least mode, and then reconstructed to the desired timeframes from the entire six minute scan, for example. So when you just go back, when you see there's nothing first, this is just before I could see it, right ventricle, left ventricle, and then slowly it becomes focused in the, it becomes accumulated in the left ventricle here in the myocardium. So first you see the blood pool here, right blood pool, left blood pool, and then it accumulates in the left ventricular myocardium. So that by knowing, by sort of looking, following this dynamic process and measuring activity, both in the, in the blood here with this red region, and in the myocardium, in that white contour, you can, from those two measurements, you can recalculate and find out at what rate you, what is your flow rate? What is your flow of the tracer from the blood to the myocardium? And this is what software for kinetic modeling or flow, absolute flow analysis do. So looking at it a little bit differently here, we have this different phases. So we have this blood phase where you have this blood, you know, in the activities yet still in the blood. It here is in right ventricle and left ventricle. It first gets into the right ventricle, then left ventricle, then it's in both. And then, then it slowly gets into the myocardium. And we call that phase tissue phase. And so, so over time, this images will, you know, will change very quickly over. So typically we reconstruct every 10 seconds or so. So here you see, you have another image of this, where we show how we extract the, the phases. And, and so we can make this curves because if we have this multiple phases, multiple frames, we can then plot the activity. For example, in this red region of interest, we can plot that activity here and we can plot it for both for the right ventricle in gray and for the left ventricle in blue. And then if you have this, this curve, so this is how you, you look at the, you know, the sort of the, how the, you know, how the activity over time travels through this, through this chamber. And then you can also then do the same for the myocardium here and look at this activity in the myocardium. And so, so we call those curves time activity curves. And we can have it for the input function for the, you know, for the blood and also for the, for the myocardium here. So now you see here in the myocardium, there will be some spillover because some of this, this effect of obviously this is coming from the blood. So it's because the PET resolution is not perfect. There will be this kind of a peak here, which is truly not just uptaken by the myocardium, but it's just an effect of the activity in the blood. So when you calculate this, you will then calculate from input function and stress and rest and the, and the, and the whole, and the tissue activity, you will calculate your flow values at stress and rest. And then by dividing the polar maps, for example, of stress by rest, you can create, you can create a myocardial flow reserve polar map, where you see the ratio of the stress and rest flow, which tells you of the functional capacity of the heart. Different models are used, as I explained before, we have this rate of uptake into interstitial phase, and then K2 is the washout. Sometimes some models, most models don't have it, but some models have some additional factors, K3 and K4 for the intracellular space. So it gets very complicated, but for the rubidium, we'll just use this K1 and the K2 model of, of uptake and washout. So if you look at it here, you will see that different tracers have a different extraction fraction. The water oxygen, water is the best. It basically extracts all the tracer, for ammonia is a little bit, you know, falling off, but rubidium has this bad characteristics that the more flow there is, the more, the less of the tracer is extracted. So obviously this has to be corrected. So we correct for this to make this, by knowing what flow it is, we can sort of map it back to that straight line. And that way we can get actual flow value, correct flow value. But obviously when you do this, you introduce some error. So ideally we would like to have tracer, which is like the red one here, but it can be corrected. This is explained in this publication by Nakazato, imaging, in imaging, medical imaging. Different packages are used for rubidium flow. And this, in this publication, we show different software models and using the same kinetic model with three different implementation of, the three different types of software, but using exact same modeling approach, is very clear, good agreement can be reached. So those are absolute values, which may, should not depend on the tool used, but it should just be, depend on the, if there's a particular method applied for the flow, it should be reproducible across different implementations. So what is a normal flow? There's many papers on this, and typically, as you see here, the normal values are like 0.7, from 0.7 to one on the rest, and from two and a half to three at stress. And then you will have like a normal, micro flow ratio would be like three to four, for example. So those are the typical normal flow values. There's many papers, there's new papers also published. This is from Murthy, a very good publication in German, Nuclear Medicine 2008, summarizing the blood flow, blood flow quality control and implementation, clinical implementation. So in last a bit, I just want to talk a little bit about the impact of patient motion on the blood flow. This can be severe, as you see here, because the patient can move between these phases, the blood can, you know, the tissue can be, the blood region can actually be seen in the myocardium, is because of the shift between the, between this early phases and the late phases. And when this is corrected, you see this, this peak here on the tissue, on the tissue curve disappeared, because that was caused by this incorrect positioning here. And as you see here in this red arrows, now this is corrected. And you can see that this can dramatically affect your flow values. So for example, in this case, the flow values were exaggerated because this blood activity was looking like as if this is actual tissue activity in the myocardium, which was sort of similar to higher flow. And here it shows that when you correct for it, it's actually much slower. So when you apply this motion correction during flow, it does increase your reproducibility dramatically, because you see that you have, you know, you have no motion correction. You have 16% variability. When you apply motion correction, you go down, maybe almost half your reproducibility. And this is the latest publication on that from Otaki in Journal of Nuclear Cardiology. So in summary, I'd like to say that the PET is increasingly used for cardiac imaging. 3D PET with CT is the primary modality right now, but the traditional systems are still used. Resolution recovery and time of flight is offered by vendors. It improves quality. Cardiac and respiratory motion correction improve quality, and it can be applied to LISMO data retrospectively. Rubidium and ammonia are FDA approved for imaging, and there's several quantitative flow packages developed, and they agree, these packages agree, if the same exact model for the kinetic model is actually implemented in these packages. Normal flow ranges are established for rubidium PET for various stress agents and for various stress agents. And PET motion during dynamic scan and PET-CT and PET-CT misregistration remain important technical challenges during cardiac PET-CT. Thank you very much.

cardiovascular PET

acquisition techniques

dynamic scans

attenuation correction

myocardial blood flow

motion correction techniques

CT scans

extraction fractions