So, my name is Rob DeKemp, I'm an imaging physicist at the University of Ottawa Heart Institute. I've been asked to present this lecture for you as part of the PET workshop on the principles of myocardial blood flow quantification. Here are my disclosures. I do work with Jubilant Draxomash, JDI, and Envia on some technologies used for blood flow imaging that is relevant to this talk. So why are we interested in myocardial blood flow? As we know, we see a polar map here on the right with a kind of picture of normal coronary vessels supplying those regions and we were traditionally looking for regions of reduced uptake. For example, in a blocked right coronary artery, this would produce images like this where we would see a relative reduction in the territory supplied by this vessel. If we have three or multi-vessel disease, what we see is perhaps regions that have maintained uptake relative to each other and this of course is a limitation of relative perfusion imaging in patients with balanced multi-vessel disease or perhaps microvascular disease. The relative perfusion images can look normal or homogeneous despite being perhaps globally reduced. And this is a potential advantage of quantification with absolute blood flow imaging. If we can add absolute units to our color scale, we can then appreciate that blood flow in these territories supplied by these abnormal vessels is in fact globally reduced and this has the potential then for improved diagnosis and management of a wider scale of coronary vascular diseases. Much of the content that I'll show today is also contained in this excellent recent position statement, joint statement from ASNIC and the Society of Nuclear Medicine on clinical quantification of myocardial blood flow led by Dr. Benck-Murthy and Marcello De Carli. I would encourage you to use this as a study guide. So the basic principle of blood flow quantification is based on the microsphere method. This is a method that's been used for many years to measure blood flow in animal models of large and small vessel disease. In this model, we're trying to identify or distinguish regions of tissue that are supplied by relatively normal or obstructed coronary vascular beds. So we need to sample activity from the arterial blood, typically by placing a catheter in the left atrium of our animal and we withdraw blood at a constant rate. We then inject either radioactive or fluorescent labeled microspheres into the bloodstream and they travel through the tissues in proportion to the blood supply. So more microspheres would go to those tissues that are perfused and fewer to those that are not perfused. We actually count the number of microspheres in the tissue and we count the number of microspheres that were withdrawn or were supplied to the tissues in our arterial blood sample. Our blood flow measure is really just a ratio of the number of microspheres in our tissue sample divided by the total number of microspheres that were collected in our arterial blood sample with some correction factors for the rate at which we withdrew that blood and the mass of tissue. But really this is the concept of measuring blood flow is looking at the amount, you know, in this case of microspheres or radioactivity in tissue divided by the total amount of microspheres or tracer that was supplied in the arterial blood. So with PET imaging, instead of injecting radioactive microspheres, we are injecting a tracer and here you see a little schematic of a radium generator, for example, where we may inject activity into the venous bloodstream and we start imaging at the time of the start of tracer injection and follow that tracer over time as it moves through the right side of the heart and lungs and eventually into the systemic circulation where it then clears from the blood finally showing our desired image of the heart and the blood cavity. And these are the images then that we use to sample our myocardial tissue shown in blue here and by placing a region typically in the LV or left atrial cavity used to sample the LV blood time activity curve instead of physically sampling those time activity curves or sampling that activity with a syringe. Similarly to or based on the PET imaging, there are two models that are used to quantify blood flow having measured now the shape of our arterial blood input in our tissue response curve. So the simpler model is a tracer trapping or what we call a retention model and this follows really principles that are almost identical to the extraction or to the microsphere model where we assume that all the tracer delivered by the blood is trapped in the tissue irreversibly. So we have our measured time activity curves in the blood and in the tissues. If we wait for some time for example until the blood activity reaches zero, in this case this rubidium scan was eight minutes long, we take our measurement at the end of the scan. So this is kind of like our tissue sampling in the microsphere model and we divide by again the total amount of activity delivered in the arterial blood. So instead of literally taking blood samples, we integrate the area under our blood curve which is depicted by this equation here. And again concept exactly the same as our microsphere method. Blood flow is based on the ratio of this with some additional technical factors to correct for specifics of PET imaging, partial volume correction, and the retention fraction, basically the behavior of the particular tracer you're using which would be different for rubidium compared to ammonia for example. The second model used is what we call a tracer exchange or a compartment model and in this case as you see on the right here, we have also a rate, still a rate describing transfer of the tracer from blood into tissue but we also recognize that most of the tracers do not accumulate irreversibly, there is usually some washout and in addition our blood curve does not always go down to zero and this process can be described more completely by using a slightly more physiologic model. In this case we use a slightly more complicated equation, basically we fit a model with an exponential clearance so you have a kind of clearance right here as an exponent and we combine that with our blood input to actually fit a model to every point along the whole time course of our measured data. We look mainly for this uptake rate K1 and this is the parameter, basically the scale parameter that's related to blood flow, in this case by function which is our extraction fraction of our tracer, again that would change depending on whether you're using for example rubidium or ammonia. There are some other technical factors, blood volume and the washout rate that are measured and used to do partial volume corrections for example, I won't get into those today. For those who are interested I would point you to this review paper that we wrote a few years ago which describes how partial volume correction is performed and why it's important and also an explanation of tissue extraction and retention fractions. Onto the practicalities or the reliability of performing flow measurements, I'm going to talk about six and six of one and half a dozen other points to consider when you're measuring blood flow clinically. The first set is in your protocol setup, a number of things to consider when you're first starting your blood flow measurement program and second are things that you should look at for every patient as part of routine quality assurance. So the first six, you should recognize that not all scanners are created equal. In particular all the commercial systems purchased today are so-called 3D PET systems, they have no interplane septa any longer. These systems are like the biographs or the discovery. Some of the earlier generation systems here and you can see going from the bottom to the top, you're basically going from newer systems to older systems. Some of the first generation 3D PET systems have actually quite limited what we call dynamic range or high count rate capability. So for rubidium perfusion imaging, for example, where we give relatively high doses, these scanners will have some difficulty remaining accurate in that very high count rate range. And so this is also a good reference paper where we looked at the capability of different systems and basically be aware that you need to understand the limitations of your scanner, particularly if you're using some of these early generation 3D or hybrid 2D, 3D systems, not to inject too much activity, either in total or using a weight-based approach. The weight-based dosing is something that can help to avoid this limitation of 3D PET. 3D has higher sensitivity than 2D PET scanners. There are still a fairly large number of 2D PET scanners being used for rubidium and ammonia perfusion imaging in the U.S., and in general, the dose can be reduced if you're using 3D PET. In fact, the dose in many cases has to be reduced to avoid these detector saturation issues. In our center, we use, for example, 10 megabecherels per kilo, and you can see then that for a 3D system, for the average body sizes, you might be giving doses in the range of 20 to 40 millicuries, whereas on an older 2D system, you're probably giving more conventional 30 to 60 millicurie doses. Generally, the camera limitations are not really an issue if you're using ammonia, because the ammonia doses are generally lower, a third to half of those compared to the rubidium tracer. So scanner setup, you need to either have an imaging department with equipment already installed, or if you're looking at a new system, you're going to be getting a 3D PET scanner definitely. On the 2D systems, there are different approaches, you have so-called multi-frame, where we have dynamic imaging performed with short frames, starting at the time of injection and usually gradually increasing in length. There are also, one software vendor has a two-frame approach, where they simply integrate all of that blood, all of the area under that blood curve with a single frame, followed by a single frame to measure the tissue response. This cannot be done on 3D PET scanners, because the count rates are changing so rapidly, and the corrections are changing, that you really must use a multi-frame approach if you have a 3D PET scanner. So this two-frame approach, I've crossed out here to emphasize that distinction. You need to set up your image reconstruction and analysis protocol. So most of the current systems have time-of-flight reconstructions, these will give better image quality than using iterative reconstruction alone, and definitely much better image quality than using filtered back projection, which really has really limited implementation today. Most current systems, even most older 2D systems, should still have good iterative image reconstruction. The number of updates, iterations times subsets, greater than 100 is a good rule of thumb. For dynamic imaging, you really want minimal image smoothing, so that you can minimize the partial volume effect, I would say something like 10 millimeters reconstructed as the max. Attenuation correction, of course, is part of PET imaging inherently, and alignment with your CT images, with your PET images is as important for blood flow quantification as it is for relative perfusion imaging. And of course, since we are using dynamic images, measuring the response of the blood and the tissues over time, looking at your data to evaluate for patient body motion is extremely important. There are a number of analysis software programs available. These are FDA approved and available, however, beware that just because it's approved doesn't mean that it's going to give you accurate numbers every time. And so definitely the garbage in, garbage out scenario definitely applies in these cases. So be familiar with the technical requirements and limitations of the particular software you're using. There are, you know, some examples of the tracers that have been used and fully or partly validated by the different software programs here, they're listed in alphabetical order. All of the programs would be applicable with the retention model using 2D, older 2D systems. Most of the systems will use dynamic framing with the compartment model that you see here. This is the one tissue compartment model, and two of the vendors use the retention model. What are the expected values for blood flow if you're starting a program? And if you've read some of the literature, you'll know, you know, in young healthy subjects, you might expect resting flows, you know, below 1.7, 0.75, increasing to 2.5 to three at stress and the flow reserve around four on average. In the typical patients referred for PET imaging, those with chest pain, you'll see very few of them with flow reserves at three or four on average. We and others see flow reserves in the range of two actually. And to check that, you know, your system is behaving as expected, you know, in regions with transmural infarction, you'd expect to have very low flows, down around 0.2, 0.3 with no flow reserve in those cases. So you know, you've done your due diligence and your protocol setup, you have confidence in your scanner and your software, now what are the things that you should look at in every patient to do reliable flow quantification? So we'll look at the second six of our basket of eggs. PET-CT alignment, as I mentioned, is as important or even more important because blood flow quantification is going to improve contrast in the polar maps compared to perfusion. And so any misalignment artifacts will actually be amplified by the tracer extraction correction. So look at your CT, your CT should be free of artifacts. Here's a good case with correct alignment where you can see the color, the PET images or PET distribution is fitting nicely into the CT soft tissue, resulting in a, you know, in this case, relatively homogeneous tracer uptake. Here's another case where the PET data is not well aligned. You can see here on the transaxial image, some of the lateral wall is spilling into the lung field, lateral wall and partly anterior as well, causing, you know, the characteristic lateral or anterolateral wall defect associated with misalignment. And this should be fixed on the console so that your images are reconstructed with proper aligned attenuation correction. Motion is also important. You should review your data. This is an example of one way to look at your dynamic data moving back and forth in time. The bottom panel, the blue highlights shows which frame you're looking at in the transverse sagittal and coronal views. And we can see here, you look at the coronal, for example, the heart is staying relatively still and fills in at the blood pool at the start of the sequence. Here's another case where this is another patient where there's clearly a lot of motion. This is playing quite quickly, but you can see on the transverse images and on the coronals, the heart jumping around. If we slow it down and start at the last frame, we can see, you know, that motion was reasonable there. But somewhere around three or two to three minutes is when this patient moved by a large amount, really blurring out the images and especially distorting the shape of the time activity curves. So in terms of blood flow quantification, that motion is going to produce artifacts in your estimated polar maps. You can see now, if we look at the shape of our blood time activity curve for a region put in the middle of the cavity, initially the peak appears okay, but later in time, the blood activity actually appears to increase, which is, of course, impossible. We also notice that our model in this case, the compartment model, the solid blue line is not fitting through our measured data points. And this, again, is because this time activity data is corrupted and it's no longer representative of physiology. It's now a mix of physiology and body motion. This can be partially compensated after reconstruction by some of the software vendors. Ideally, it should be compensated during image reconstruction by the scanner software itself. You should make sure, as you do for perfusion imaging, that your tissue contours are properly placed. Here you see the contours, the position shown in red. Make sure that the regions that are placed for your arterial input function are well centered in the LV cavity or in the atrium or in the outflow tract, depending on the software that you're using, and make sure that there should be no overlap or minimal overlap of the arterial input region with the surrounding LV myocardium. Look at your time activity data. It's critical that you can verify that you've captured the whole time course of the blood input. Otherwise, your estimation of the area under that curve is not going to be correct. If you start too late, the area under that curve is going to be underestimated and your flow values are going to be overestimated. So check that you have, ideally, at least one point measured with zero activity near the start. If you're using a compartment model, you can check again that your model fit, the solid blue line, is fitting through your measured data points. And if that's the case, you have the whole blood curve and your model is fitting through your data, then you should have good confidence that you can interpret those results. Some of the programs will give you some goodness of fit metrics, like an R-squared. You can get an R-squared value for a model fit just like you can for a linear fit. These should be uniformly high and close to one. Chi-squared generally have values below three. This is a little bit more difficult to interpret and other parameters looking at convergence or bounds, upper lower bounds, important to know as well. So in summary, I would say some points to consider for your blood flow program. Know your scanner, particularly if you have a 3D scanner using rubidium. Know what the high count rate limitations are and don't inject 60 millicuries of rubidium in a small patient. You will overwhelm almost any scanner. Weight-based dosing can help to maintain consistent quality and to limit those high count rate saturation artifacts. Anything you can do to perform your imaging with consistency using a syringe pump for ammonia or using a consistent shape of the injection profile will help improve the reliability of your flow measurements. Look at your dynamic images and your time activity curves to make sure that there's limited patient motion or that the patient motion has been corrected, that you've measured the entire blood curve and your model is fitting your data and with those things in place, you should be able to interpret those flow values with good clinical confidence. And with that, I will conclude and thank you for your attention.

myocardial blood flow

absolute blood flow imaging

PET imaging

coronary vascular diseases

tracer exchange model

blood flow quantification