Transcription
Hello everybody and welcome back. Let's have a look today at the physics underlying the process of dual energy computed tomography and show you how that differs from conventional computer tomography.
When we think about creating a conventional CT image, we separate out tissues based on their differing linear attenuation coefficients at a specific incident photon energy. And we said that differing linear attenuation coefficient values will get converted into Hounsfield units, and that will then be laid on a grayscale that will allow us to look at an image and see different tissues based on those different attenuation properties.
Now, what happens if we have two tissues that have the same linear attenuation coefficient value at a specific incident photon energy? In conventional CT, we have no way of separating out those two tissues. Dual energy CT now provides us with a way to differentiate those tissues, and we do this by exposing the patient not to one incident photon energy, but to at least two separate incident photon energies. And I want to show you today how that change in exposure will allow us to then differentiate tissues with the same linear attenuation coefficient value.
So, how does this work? Well, let's take a voxel of tissue here. If we had an incident X-ray beam on this specific voxel of tissues, and we said that the average energy of these X-rays here is 100 keV, we've got a set number or a set intensity of these X-rays. Some of those X-rays are going to be attenuated, and some of them are going to be transmitted through that voxel of tissue. We call the transmitted X-rays N here. Incident X-rays and transmitted X-rays, the proportion of X-rays that are removed from the X-ray beam by this length of tissue is what's known as the linear attenuation coefficient. And we know that the linear attenuation coefficient is the combination of attenuation from the photoelectric effect, Compton scatter, and Rayleigh scatter. This is all things we've looked at before, especially in the Hounsfield unit talk.
Now, we know that these processes are dependent on certain factors. They're dependent on the incident photon energy, they're dependent on the density of the tissue, the atomic number of the tissue, as well as the electron density of the tissue. Now, if we know the length of tissue that the X-rays are traveling through, and we know the intensity of the incident X-ray beam, as well as the transmitted X-rays, we can calculate the linear attenuation coefficient using this formula here: the incident X-ray intensity to the negative exponent of the linear attenuation coefficient of this tissue passing through a set distance here. We know that attenuation is an exponential process, that's why we've got this part of the equation here. And we can isolate the linear, the unknown variable here, which is the linear attenuation coefficient, and calculate it with this voxel of tissue. That's how it's done in computer tomography.
Now, what happens if we take that exact same voxel of tissue? The tissue density, atomic number of the tissue, the electron density of the tissue, that all remains the same. We only change one variable: that's the incident photon energy. Now, we've got the same number of photons here, the same quantity of the X-ray beam, but the quality of the X-ray beam has changed. We've lowered the incident photon energy. A lower incident energy will mean that more X-rays are attenuated, especially via the photoelectric effect. We've seen that the photoelectric effect is inversely proportional, actually, to the power of three to the incident photon energy. The higher the incident photon energy, the much less the photoelectric effect is to occur. So, as we lower that energy, we increase the photoelectric effect, and fewer X-rays pass out or are transmitted through this voxel of tissue.
Now, this is incredibly important because you can see that changing the incident photon energy is going to change variables in this equation. We can keep the incident photon number here the same, but it's going to change our transmitted photon number. The distance that those photons are traveling through the tissue remains exactly the same. That means it's the linear attenuation coefficient of this tissue will change. Now, that's the important part here. Linear attenuation coefficient is dependent on incident photon energy. We can't say that all bone has X linear attenuation coefficient without knowing what the incident photon energy is. The higher the incident photon energy, the lower the linear attenuation coefficient for each tissue is going to be, and that happens for every tissue. We get a drop-off in linear attenuation coefficient as the incident photon energy increases. So, we see that this linear attenuation coefficient is dependent on incident photon energy.
Now, what do we use the linear attenuation coefficient for? We use it to calculate Hounsfield units, and Hounsfield units allows us to standardize linear attenuation coefficients of tissue to water here, and it shows us whether a material is more attenuating than water or less attenuating than water. And we can match those Hounsfield unit values to a grayscale that will ultimately allow us to create an image. So, you can see that water's linear attenuation coefficient, even though it changes with incident photon energy, won't change our Hounsfield unit value. Water should always be about zero Hounsfield units because of this equation here. Other tissues are always going to change. Their Hounsfield units are also going to be dependent on incident photon energy. So that's why when we've got a higher energy exposure, we're going to get differing grayscale values if we were taking the same scan with two different incident photon energies.
So, let's take a practical example. Let's take a voxel of tissue here and say that that voxel of tissue was completely filled with iodine. So, we've given contrast here. This voxel only has iodine in it. We've got incident photons, and we've got transmitted photons. What we can do then is plot the linear attenuation coefficient through a range or a spectrum of photon energies. If we were to increase the average energy of this incident photon beam, what would happen to the linear attenuation coefficient of this voxel of iodine? We would expect the linear attenuation coefficient to drop off as incident photon energy increases, and that's exactly what we see. We see this drop-off here.
Now, if you've done my X-ray Physics course, you'll notice what this is. This is what's known as the K-edge. We know that for the photoelectric effect to occur, an electron has to be ejected from the atom that it, the X-ray strikes. That's a photoelectron that's giving dose into the tissues. If you remember from our previous talk, we've ionized that atom. In order to ionize the atom, the incident photon energy needs to be higher than the binding energy of that electron. And the inner shell electrons of iodine, I is a large atom, it's got tightly bound inner shell electrons. Their binding energy is about 33 keV. Once our incident photon energy is higher than 33 keV, we've got a whole extra shell of electrons that can then provide an electron to contribute to the photoelectric effect. So, we get this jump up in linear attenuation coefficient before we get the drop down again. Okay, so we see the coefficient dropping off as incident photon energy increases.
Let's dilute the iodine down to 5% iodine. What's going to happen to our attenuation coefficient? Well, it depends what's left in this voxel as well, but say it's all just 5% iodine, the rest is water. We're going to get a drop-off in the linear attenuation coefficient. We've still got the same K-edge here because of those inner shell electrons, but the linear attenuation coefficient across the board is lower still, with the same or similar drop-off pattern. Don't use these as absolute values, these are just guesstimates, graphs that I've made to illustrate a point. Notice again, the Y-axis here is a logarithmic scale as well, so this is really is an exponential decrease in the linear attenuation coefficient.
Let's look at another tissue now. Let's say we're looking at bone here. Bone is going to have a completely different profile to iodine. It doesn't have this K-edge here. The average atomic number of bone is going to be much lower, probably seven or eight. So, if we plot the linear attenuation coefficient for bone, we can see that happens here. Got three very different attenuation coefficient profiles for differing incident photon energies.
Now, let's run a little experiment. If we were to place three voxels within our CT scan here, one of them's iodine, one of them's 5% iodine, and one of them's bone. If we were to have an average energy of our beam here to be 80 keV, here we got 80. Here, notice you can see we're going to run into a little problem. Our 5% iodine and bone has a very similar linear attenuation coefficient. When we convert that to Hounsfield units, they're almost going to be exactly the same. We might not be able to differentiate those two. If we've got iodine that's given to a patient that's in the vessel, it's near bone, and it's at a 5% concentration, not going to be able to tell the bone from the iodine. Maybe calcification from the iodine apart.
What we can do is run a CT scan here and calculate the Hounsfield units or calculate the linear attenuation coefficients of each one of these voxels, and we can plot it on a graph. We can draw an axis here and say at 80 keV, at an average incident photon energy of 80, what do each one of these linear attenuation coefficients come out as? Remember, we've performed some form of reconstruction algorithm, we've calculated Hounsfield units here. The first voxel, we say that has an LAC of three. So, here we can see LAC of three, kind of corresponds to iodine at 80. The second voxel we look at here has a linear attenuation coefficient of 0.1. Here's the problem we're running into. Is this iodine or is this bone? The third one as well has a linear attenuation coefficient of 0.1. So, we can fairly confidently say that this first one, if we knew that these were the only three tissues in our image, we can fairly confidently say that that's iodine. We can't differentiate these two voxels here.
What then if we were to then repeat the CT scan but use a different incident photon energy? You can see now our beam is no longer red, it's green here to represent a differing average incident photon energy here, in this example, 120 keV. We've increased the incident photon energy again. We can calculate the LAC values for each one of these pixels and plot them on the same graph here. So, we take the CT scan here, we calculate those values, and we plot them on the same graph here, using one axis as our LAC values for 80 keV and the second axis for LAC values at 120 keV. LAC being linear attenuation coefficient. This first voxel, when we repeat the scan at a higher average energy, gives us an LAC value of 2.1. It's decreased from three down to 2.1. That corresponds to this drop in LAC with iodine as the incident photon energy increases.
Our next voxel, we see that the linear attenuation coefficient for that voxel now gives us a value of 0.09. If we were to look at 0.09 on this graph here, we could see that it may correspond to bone. Let's see what happens to our third voxel when we have repeated the scan. It gives us an LAC value of 0.05. This scan here, when we calculate Hounsfield units, would have been able to differentiate those two tissues.
Now, remember, patients aren't just iodine and bone. They've got multiple different tissues. We've got the spleen, we've got the liver, we've got fat, we've got air in the lungs, we've got soft, multiple different soft tissues. So, all of these linear attenuation graphs are going to cross over in varying different degrees. They all respond, all the tissues respond slightly differently to incident photon energies, and you'll see when I show you some clinical examples later on how we can utilize these then to separate out these tissues. But now that we've done these two differing incident photon energies, we can fairly confidently say that this voxel was bone and this voxel was 5% iodine. And that's exactly what dual energy CT boils down to: applying two different energies.
Dual energy CT allows us to separate tissues based on their response to the differing energies, based on their rate of change of linear attenuation between two energies that we've selected. Tissues are going to respond differently. The rate of change is going to be different for different tissues within the body because of their constitutional makeup, because of the rates of the photoelectric effect versus Compton scatter. In bone, a lot of this attenuation is due to Compton scatter. In iodine, a lot of it is due to the photoelectric effect. Compton scatter is less dependent on incident photon energy as opposed to the photoelectric effect that provides these differences here.
So, how then do we go about obtaining a dual energy CT scan? Surely we've got an X-ray source rotating around the patient. How do we provide two different energies? Well, I like to separate this into four different mechanisms and call them either single source mechanisms or dual source mechanisms. The single source mechanisms, the first one I wanted to show you is what's known as fast kV switching. We have a single source rotating around the patient, but we rapidly alternate between two different KVs. So, for example, in our last one, we have say a kVp of 120, so our average photon energy is going to be lower than that, and a kVp of 80. And we rapidly switch between those two. Now, the filament current running through the cathode is going to remain the same. So, when we increase the kVp, we're going to increase the number of photons because we're increasing X-ray beam quantity and quality. So, we actually want to expose the patient slightly shorter to the increased kVp as opposed to the lower kVp, try and even out those exposures. But what happens is, as that source rotates around the patient, notice how it's rapidly switching between the two KVs that we've selected as operator. For those of you, apologize to your colorblindness, these are red and green switching between each other, showing you that switching between two different KVs. And this is one mechanism that certain manufacturers use. I know that GE used this mechanism. It's a single source mechanism to obtain dual energy CTs.
The second single source mechanism is to have a single source with a single energy, but have two different detectors here. I call this a dual detector layers. The first detector layer is going to respond to lower energy X-rays. The scintillation layer in that detector is going to create light only from lower energy X-rays. High, higher energy X-rays are going to pass through that scintillation layer and be recorded by the second, the dual layer here, that picks up higher energy X-rays. What this allows us, it gives us great temporal resolution because we're acquiring the image in one go. We're not rapidly switching between KVs. We don't get motion artifact like we would here potentially. We're taking this in one shot here. You see it goes around the patient in one fluid motion, and we're acquiring data simultaneously here. But what we get is spectral overlap. These two layers aren't perfectly able to separate out two energies. Some higher energies are going to be picked up by the low energy layer, and some lower energies are going to be picked up by the high energy layer. That's just a function of the technology that we have now. So, those are the two single source mechanisms for acquiring dual energy CT scans.
Then we have dual source mechanisms where we've got two X-ray sources rotating around the patient. This is what's known as a dual source, dual detector system, where we have one X-ray source at a specific energy and a separate X-ray source at a differing energy. They're simultaneously on whilst they're rotating around the patient, and we're simultaneously acquiring the data here. The temporal resolution is going to be a quarter offset though between the two. This can then rotate around the patient, and we can create two separate data sets, each representing the different KVs that we've selected for each of these two sources.
The last option, or one that's not really used in clinical practice, is to just perform two scans. Perform one scan at a particular X-ray energy, and then perform a second scan at a different X-ray energy. We can then take those two data sets and again perform that calculation with the two axes of our graph and separate out the different tissues. Practically, that's not used. These are the three main mechanisms that are used in dual energy CTs.
Now, ultimately, each one of these mechanisms is going to give us the same outcome. It's going to create two separate data sets that correspond to Hounsfield units or linear attenuation coefficient values for each of the pixels for differing incident photon energies. And you've seen before how we can use those differing incident photon energies to extrapolate out different tissues based on their differing responses to those two separate energies.
So, let's have a look at some clinical examples where this is used. The first example I'm going to show you is an example of gout. We've got monosodium urate crystals that are deposited into the tissues around the joint. We know the typical response of monosodium urate crystals to differing photon energies, and we can perform dual energy CT to select for pixels that have had that specific response. Here, we've got a 3D rendering where we've selected for bone and we've selected for the monosodium urate crystals, and we've created this virtual 3D image where we've got a virtual light source here. We've looked at before, this is really useful. We can monitor response to treatment, say, and see an actual visual reduction in monosodium urate crystals. You can see that we've allocated a specific color to those pixels here, a green color to separate visually out those tissues. If we're looking at a patient, especially a patient that presents atypically with joint swelling, we're not sure whether this is gout or whether this is perhaps calcium deposition, we'd be able to separate out uric acid crystals from calcium because they've got differing responses to those different incident photon energies.
Another example here is if we've done a CTPA, we can see we've got contrast in the vessels here. We can see a hypoattenuating mass within this vessel here. We suspect it's a PE. We want to make sure that this is in an artery. What we can do is we can provide an iodine map here. We got lower iodine concentration perfusing the lung tissues here. So, we saw how the iodine percentage has a different rate of change as opposed to 100% iodine. These are going to be lower percentages of iodine, and we can see this wedge shape of perfusion here where we get no perfusion to this wedge-shaped part of the lung, confirming a PE here. We can separate out the iodine from the background lung tissue quite easily here, whereas we can't do that in, especially in this window. We can use these perfusions as well to actually look at perfused blood volume. This is a similar example, but we color-coded the concentration of iodine because we know the response of differing concentrations of iodine to increasing incident photon energies. We can then provide color maps. This has got a scale from orange to red, but it can be from green to blue, you may see as well, and we've got typical changes here. Notice the apexes in the bases of the lungs here where we've got limited perfusion as well, because those lungs have collapsed down, those vessels have collapsed down, we don't need to perfuse those areas physiologically. You can also notice this dark region between the lung and the heart here. This is also another artifact that we typically see in these images where the heart's pulsation is causing this artifact here, and we're going to look at artifacts in the next talk.
Another example is we can say we've got a CT here. We got a sagittal CT of the lumbar spine. We see we've got these compression fractures in the lumbar spine. Now, we want to figure out, have we got acute fractures here, or is this all a chronic process? And we can do bone marrow edema mapping. We can take out the contribution of the calcium from the trabeculae here and figure out attenuation changes to see whether that attenuation change is based on normal bone marrow, see this blue signal here, or whether there's potentially acute edema, like we can see in this vertebrae here. We can suggest that there's perhaps an acute on chronic fracture here in this vertebrae because we've got bone marrow edema here, whereas this looks like more a chronic process.
The possibilities for dual energy CT are seemingly endless, but whether they provide enough clinical value to justify their increased cost is an ongoing concern. It costs a lot more. The machines that acquire dual energy CT images cost a lot more than your basic CT machines. You need special training for your radiographers, for your radiologists, to be able to go through the data sets and know what they're looking at. And that's perhaps an explanation for why dual energy CT has taken longer to be globally adopted. We need to prove clinical value. It's all good and well seeing these, but what change does it make to the patient at the end of the day?
Now, I'm not going through all the clinical applications. I'd encourage you to look up some of the clinical applications. Look up bone removal in our CT angiography studies. You can take out the background bone and look specifically at the vessels themselves. Look at virtual non-contrast CT scans. That's a very interesting one. We've done a contrasted CT scan, and you can actually mathematically remove the iodine because we know where the iodine is based on this dual energy CT. Remove the iodine, and we basically got a virtual non-contrasted CT scan without having to expose that patient to first a non-contrasted scan and then a contrasted scan. So, I'd encourage you to look up those further clinical applications.
What I want you to focus on for this talk is really the underlying physics: how we're able to separate out tissues that have similar linear attenuation coefficients based on their response to a change in incident photon energy. So, that's all for today. We're going to move on to our next talk where we're going to focus on some common artifacts seen in CT imaging. So, until then, I'll see you all there. Goodbye everybody.