Transcription
Hello everybody, and welcome back. Today, we're going to go through a few common artifacts in CT imaging.
Now, when I think about CT artifacts, I like to classify them into three main categories. The first type of artifact is caused by the patient themselves, or what the patient's wearing. The second group, or second category, are called physics-based artifacts, and they're due to some physical principle that are resulting in an artifact in our image. And the last type of artifact are artifacts that are associated with the hardware that we're using to acquire the image.
Now, for each one of these, I'm going to go through some common examples, but remember, there are more examples than that are covered in this talk. Also, patient-based artifacts are not solely due to the patient; they have underlying physical principles, and they relate to the hardware that's acquiring the image. These are all intercalated, but it's useful in our minds to separate them into these three categories.
So, let's start with patient-based artifacts. And the first type of artifact I want to look at is a scan here where we can see there seems to be an interruption in the cortex of the sternum here. It looks like there's a fracture here. Now, there actually isn't a fracture here. This is what's known as motion artifact, where the patient has moved during image acquisition.
Now, motion artifact can either cause the step appearance here, or it can cause streaking in the image, or misregistration of the various different anatomy. And the clue here is to look across the axial slice here, and you can see the interruptions not only in the sternum, but also in the skin here. You can see it looks like a break in the skin that would be very rare, and in the aorta here. So, this is what's known as patient motion, where the patient has breathed during CT acquisition, and that breathing has caused movement in the chest, and ultimately, we're going to register these voxels at different locations based on the patient movement.
You can see another example here, again, where we've got a step in the cortex. You can see we've got doubling up here of the cortex of this sternum here. There's also a break in the skin here. This is not an actual break; this is an artifact due to movement of the patient. You can see we've actually registered twice in the same axial slice because our pitch here is less than one, we rotated around that specific slice more than once when acquiring this image.
Here's another example of patient motion, where the patient really has moved during this CT head. We can see this complete misstep here between the parietal cortex and the temporal cortex here. You will note that the patient doesn't look like this going into the scan; they haven't had a massive trauma. This is because the patient has moved. We can do a 3D reconstruction of this image, and we can see these movement bands here. We get streaking across the image, and we get misregistration of the cortex here. Looks like the patient may have been talking as well during this examination.
So, what can we do in order to reduce motion artifact when acquiring our images? Well, the first thing, and probably the most important thing, is to have clear communication with your patient. Tell them when you're going to take the scan, ask them to hold their breath throughout the scan, try to reduce that chest wall movement. Now, some movement's going to be involuntary; the heart's not going to stop beating, we do need to breathe at some stage, but other movements like breathing and swallowing, that's a common artifact in the neck. A patient swallows during the acquisition; we can ask patients not to do that. You can ask patients also just to hold still, not to move within the CT scanner.
We can also provide immobilization. We can either strap the patient's head down, or place it within a foam mold to help the patient keep their head still. Often in pediatric patients, we need to hold them down with some sort of immobilization. The patient may also require sedation to help with keeping still within the CT scanner.
We can also perform the scan quicker, so the patient doesn't have to stay still for as long. We can increase our rotation speed, we can increase the table speed, basically, we're increasing the pitch. We can decrease the coverage, make the area that we're scanning smaller, and ultimately get the scan done faster. All of these are going to help reduce motion artifact.
Helical scanning is the same. If we're doing axial scanning, where we turn the X-ray machine on, rotate it once around the patient, turn it off, move the patient down the bed, and then repeat that process. That time where we're turning the X-ray machine off, or the X-ray source off, allows for movement to occur, and we can get these step-like artifacts. If we're doing helical scanning, we've got continual scanning of the patient.
Lastly, we can do things like ECG gating, where we acquire the CT scan, we only acquire, or only use data during a specific specific phase within the cardiac cycle. We're taking an ECG, and only at a particular point in that ECG do we trigger image acquisition. Either we can do that prospectively whilst we're acquiring the scan, or we can do it retrospectively, where we acquire the scan, and then we only use the data at a specific point in the cardiac cycle. That's going to help with motion in the heart, especially when we're looking at the coronary arteries.
So, that's our first artifact, motion-based artifacts. The second patient-based artifact that I want to look at is what's known as trans-interruption of contrast. Here, we're doing a CTPA. This is commonly seen in a CTPA. We can see there's vivid contrast enhancement within the ascending and descending aorta here, and we've got contrast filling the superior vena cava, but we've got much less contrast in the pulmonary trunk here, in the pulmonary arteries.
Now, this artifact could lead us to think that maybe there's some sort of blockage, maybe there's an embolism that's preventing contrast from flowing into the pulmonary trunk, or into some of the pulmonary vessels. What's happening here, actually, is that we've got a delay of contrast entering the pulmonary vasculature because we've got uncontrasted blood entering the heart from the inferior vena cava. We've given the bolus of contrast in the arm, say, and that's going to travel up through the superior vena cava. But we've got blood entering the right side of the heart from the inferior vena cava that doesn't have contrast. So, we get mixing with that contrast. And if a patient were to take a very deep breath, drastically decrease the intrathoracic pressure, we're going to get an increase of venous return from the inferior vena cava, and that's going to dilute, or prevent that contrast from entering the pulmonary vasculature when we expect it to. It's also common in pregnant patients, where they've got increased intra-abdominal pressure, and we've got increased venous return through that inferior vena cava.
So, those are the two major patient-based artifacts that I'm going to cover in this talk. Now, I want to move across to some physics-based artifacts. Now, obviously, this is a physics course, so I really want to spend some time really going through these artifacts. The first artifact that I want to look at is what's known as beam hardening artifact. Very common artifact that you'll see in CT imaging. If you've been doing this course, you'll recognize this diagram. Well, these are electrons that are being accelerated from that cathode towards the anode. This is one tungsten atom within the anode here. In this diagram, we are creating, or generating, X-rays at the anode that are going to be accelerated towards our patient. I'm representing the patient here by this gray sheet here. You can think of this as a slice through the patient, and we're going to get an X-ray spectrum reaching our detectors.
Now, the degree of X-ray attenuation depends on the tissue within the patient. Some tissues are going to barely attenuate X-rays, say like tissues like the lung. Some tissues are going to attenuate a lot of X-rays, like dense bone tissue. There's a lot of attenuation happening in bone because it's so dense.
Now, when we are reconstructing an image, there are certain assumptions that are made, and they're actually false assumptions. The first assumption that's made is that the X-ray beam quality, the average energy of that X-ray beam, remains the same as it passes through the patient. And as a result, we make a second assumption that the linear attenuation for each individual tissue remains the same throughout that patient. We've seen that linear attenuation coefficient is energy-dependent. The higher the energy of the X-ray beam, the lower the linear attenuation coefficient for a specific tissue is going to be. And it's these two false assumptions that lead to beam hardening artifact.
Watch what happens if I place a more dense tissue here between the anode and the detectors. As that tissue gets more dense, the average energy of the photon beam, or the average energy of the X-ray beam, increases. Watch how that happens. Less dense, more dense, like that. The quantity of X-rays reaching that detector decreases, but the average energy, the penetration ability of that X-ray beam, gets more. That's what's known as beam hardening. A harder beam has a higher average X-ray energy, and that becomes important if we look at a patient placed in a scan here, and we've got X-rays passing through that patient. As the X-rays pass through the patient, we're going to preferentially attenuate lower energy X-rays, and that X-ray beam is going to get harder. The average energy of that X-ray beam gets more and more. As the energy of the X-ray beam gets more, that X-ray is able to penetrate through tissues more readily than it did initially when it was entering the skin. The beam is getting harder. We know that with this increase in X-ray energy, we're going to get a decrease in linear attenuation. If we had water at the skin surface here, and water at the exit surface here, the linear attenuation of that water at the exit surface is going to be less. And ultimately, the linear attenuation coefficient defines the Hounsfield unit that we're going to calculate for that specific voxel.
Now, we know that patients aren't uniform in width, and the beam traveling through the largest width of the patient is going to be harder compared to these peripheral beams. Not only that, but the CT source rotates around the patient. We've seen that before. And due to this rotation, we're going to get a central region where the beam, on average, is harder than all these peripheral regions. Because in these peripheral regions, we're getting less attenuation, as the beam rotates around, this harder region within the middle of the patient is going to ultimately lead to a calculation of lower Hounsfield units than what's truly represented by the tissue because of that beam hardening. And this effect is what's known as cupping. If in our theoretical example here, all the tissue, all the voxels in that patient had the exact same linear attenuation coefficient for a specific X-ray energy, we would reconstruct that image to look at something like this, where we get a reduction in Hounsfield unit centrally. And that's the first way that we can see beam hardening artifact in our image: cupping in the image, decreased Hounsfield unit centrally in the image.
The second way that we can see beam hardening artifact is when we get two really dense structures next to one another. X-rays are going to pass through the petrous bone, which is dense bone, and the X-ray beam is going to become disproportionately hardened compared to other regions in the scan. That hardened X-ray, a higher average energy, is now going to pass through the medulla or the pons here, and it's going to pass through this region with a higher X-ray energy than what we would expect. Because of that higher energy, we're going to calculate lower linear attenuation coefficient values here, and that's what's happening. See these dark streaks that are coming across the posterior fossa here? These dark streaks. So, what's known as beam hardening, it often occurs between two dense bones because it's not only happening from this way, but it's also happening from the other way as the CT machine rotates around the patient. It's a common place for this to occur.
Here's another example of a patient that's swallowed two screws here. You can see how attenuating the metal is on the scanogram. We can see that in the CT version here, we've got these dark streaks coming away from the screw here. The screw also, in this image, in this window, looks bigger. That's what's known as blooming artifact, where we're unable to properly assign Hounsfield unit values to the voxels surrounding the screw itself. If we change the window here, we get much better resolution of that screw, but we still got beam hardening where we got these dark streaks heading away from this highly attenuating structure.
So, how then do we go about reducing beam hardening in our image? Well, the first thing we can do is to increase the penetrability of the X-ray beam itself, where the beam hardening change is less because we're starting with a higher average energy of the X-ray beam. The second technique we can use is to utilize iterative reconstruction algorithms. Instead of spreading the data back across the patient through multiple different angles, like we use in filtered back projection, and we perpetuate those false assumptions about linear attenuation coefficient of tissues through different regions in the patient, as well as the false assumption about the X-ray beam staying the same average energy through the patient, iterative reconstruction doesn't use those false assumptions, and it allows us to build our image with much less beam hardening within it. If you don't know what I'm talking about here, go back to that iterative reconstruction video within this learning pathway.
Next, we can use filter filtration. We can pre-harden the beam. We can remove those lower energy X-ray photons prior to them reaching the patient. We've seen beam shaping filters as well that will help us reduce that cupping artifact. If we use a bow tie filter, attenuate those X-rays more peripherally on the patient more than we attenuate the central X-rays, and we can just use sheets of metal to take out those lower energy X-rays that aren't contributing to our image but are contributing to patient dose.
We can use dual energy CT, like we looked at in the previous talks, and we can create what's known as a virtual monoenergetic beam. We take X-rays at say, 80 keV and 100 keV. We can take those Hounsfield unit values that we've calculated for each pixel within the image, and we can create a virtual image, say at 90 keV, where we say, what would the Hounsfield unit be in this voxel if the source was a monoenergetic beam at 90 keV? And doing so, because we're creating a virtual monoenergetic beam passing through the patient, we can reduce the beam hardening artifact.
Now, I said that beam hardening artifact can either look like cupping, or we can get dark bands between two dense objects. Cupping, we saw, was falsely giving lower attenuation values to the center of the image. We can place a phantom within the CT machine, create this image, and then normalize these voxels so that they all get the same Hounsfield unit value. That normalization algorithm that we've applied to the phantom, we can then apply to the patient, and we can increase those Hounsfield unit values centrally in the image to more accurately represent the anatomy.
So, that's beam hardening artifact. It occurs because the beam becomes more penetrating as it passes through tissues, and as those lower energy X-rays get preferentially attenuated, and the average energy of the beam increases.
Now, the next physics-based artifact that I want to look at is fairly similar to beam hardening artifact, but the underlying physics is different. This is what's known as photon starvation. You can see a very large patient here that they're extending outside of the field of view. If you look at this image, we've got high Hounsfield unit values around the peripheries of the image, and we've got a very noisy image. And if you look carefully, that noise is coming out almost in streaks like this. It's got very low contrast and high noise.
Now, what's happening here is because the patient is large, we've got a much larger patient than our phantoms that we calibrated the CT machine on. We're using an X-ray fluence, the number of X-rays and average energy of those X-rays, that's too low to penetrate properly through this patient. A large proportion of these X-rays are being attenuated by the patient before they reach the detector, and that's the basis of photon starvation. We get not enough X-rays, not enough intensity, reaching the detectors. We don't have enough data there to accurately calculate or represent the anatomy, and as a result, our signal-to-noise ratio is much reduced. We've got much more noise compared to the signal. There's low signal reaching the detectors.
This can happen not only in large patients where we are attenuating a lot of those X-rays, it can happen in patients where there are very dense structures. It's a common occurrence in the shoulder girdle here. You can see it's very noisy around the shoulders here because we're attenuating a lot of X-rays, especially when we're going across the patient here, we're going through the head of the humerus, we're going through the scapula, the clavicles, all highly attenuating structures, and as a result, we get this noise, low contrast within the image here. We can also see some beam hardening from a metal artifact where we've got fillings here in the patient. It's also quite noisy here where we see at the same level as the petrous bone, not only would there be beam hardening, but we also get photon starvation here, creating this noise.
Now, how do we go about reducing photon starvation within our image? Well, the first thing that we can do, again, is to increase that X-ray fluence, increase the KVP, or increase the current in our filament. That's going to increase the number of X-rays, so more X-rays will penetrate through the patient and reach the detectors, and it's going to increase the average energy of those X-rays, allow the X-ray beam to be more penetrating. The problem with photon starvation is not enough X-rays reaching the detector. Again, iterative reconstruction algorithms allow us to use lower doses because they can account for noise, and they account for noise in various different mechanisms that I covered in the iterative reconstruction talk.
The next thing we can do is to reduce the pitch, especially if we reduce the pitch lower than one. A pitch lower than one means that each slice that we take around the patient is actually going to overlap, and the regions of anatomy are going to get scanned effectively twice. We've got twice as many X-rays then reaching the detectors per slice in our patient. That's going to increase our ability to accurately calculate the anatomy within our field of view.
We can use what's known as automatic tube current modulation, which we've looked at again, where we increase the tube current based on the density of the structures in the Z-axis direction of our patient. Increasing the tube current over those more dense structures like the shoulder girdle, or the pelvis, or the petrous bone, means that we're going to get more X-rays, a higher quantity X-ray beam, ultimately allowing more X-rays to reach our detector and creates a less noisy image, an image with more signal-to-noise ratio. Now, remember, this tube current modulation doesn't only account for density of structures, but it also accounts for the shape of the patient. So, when we're going across a longer region, a wider region of the patient, we're also going to have a higher tube current.
Next, we can use what's known as adaptive filtering. If you look at this graph here, we can say this is the signal that's reaching the detector, and this is our detector here. We've got a very dense structure, say here, and a dense structure here, that's allowing very few X-rays to actually reach the detector itself. Now, this dotted line here is what's known as a threshold. That's the number of X-ray photons that are going to have to reach the detector to give meaningful signal that can be deciphered from background noise. And because these structures are so attenuating, they're below that threshold, we can use adaptive filtering to smooth out the regions that are below the threshold, you see like this. And that means when we back-project that signal, we're going to get less noise in these dark regions on the image, get less noise ultimately propagated into our image because of that photon starvation, those lower number of X-rays reaching the detectors.
Now, the last thing we can use is what's known as metal artifact reduction algorithms, or MAR algorithms, and these help to reduce the noise caused by photon starvation when metal in the patient doesn't allow photons to reach the detector itself. Now, there are multiple different algorithms, and they are beyond the scope of this talk, but know that they exist, and know that when we use them, new artifacts can be introduced into the image that we need to know about.
So, that is photon starvation. Let's go on to our last physics-based artifact known as partial volume artifact. Here are five detectors, and we are trying to scan these two structures here. Say they're lying within our patient. Pretend that these structures are the same density to one another, and every region in the structure is the same density as it has the same Hounsfield unit value, or the same linear attenuation coefficient. We're going to shine our X-ray beam through these structures, and they're going to land onto the detectors. Our X-rays are going to land on the detectors. X-rays are going to be more attenuated through the structure than say, passing through here and here, where X-rays aren't being attenuated. We can calculate the attenuation and represent it on a grayscale here, where we've got very low attenuation being white, and high attenuation being a darker color. Notice here how part of this object is attenuating the X-ray beam and landing on this detector here. We're going to represent that as an average of the attenuation in this entire width of this detector. The average X-ray attenuation has now included some of this object here. When we go about converting these two into Hounsfield units or linear attenuation coefficient values, they're going to represent very different values. But we know that attenuation of this part of this object is the same as attenuation of this part of the object, and we get blurring, especially between dense and less dense objects. This is what's known as partial volume artifact, because only part of this structure is contributing to the attenuation profile of the entire width of this detector.
Now, the way to reduce partial volume artifact is to reduce slice thickness or reduce detector width. This can happen both in the Z-axis and in the XY plane. So, watch what happens here. I'm going to increase the number of detectors, and I'm going to make them much less wide. Now, when we calculate the linear attenuation coefficient, we can see we get a much crisper, much more sharp, much more accurate image here. We've still got some partial volume artifact, but it's much smaller in the final image. So, to reduce partial volume artifact, we want to have a more collimated beam, or we want to use or bin fewer detectors together, so we get a thin slice, and we want to have a smaller width to our slices in the XY plane.
So, that brings us to the end of our physics-based artifacts. I'm just going to look at one hardware-based artifact, and we've actually covered it before when we looked at generations of CT scanners. Got a third-generation CT scanner here with one faulty detector. This detector is either miscalibrated, or it's not picking up any signal at all. When the X-ray beam rotates around this patient, that faulty detector is going to be responsible for a specific line around the isocenter of this patient. That line creates a circle, or a ring, where we've got no useful data to create our CT scan for this slice. We can highlight that ring here with the circle. Every piece of anatomy that falls on this circle is being coded for by that specific faulty detector, and we've got no useful data. And ultimately, our image is going to show us a ring where we've got what looks like an extremely highly attenuating ring. It looks highly attenuating, it looks very bright on our image, because that detector is saying, "I'm getting no signal, there must be full, full attenuation of the X-ray beam in the path heading towards that detector." And notice how that spiral alternates as we're scrolling through the scan here, because we're using a helical acquisition. You can see that in our coronal plane, how that spiral is, it looks like it's being screwed into the scan because of the helical acquisition.
Now, the way we can get around this is either to recalibrate that detector, or we can replace that detector. Or we're going to need to use another generation of CT scanner. Fourth-generation CT scanners don't have this issue because they've got all of the detectors around the patient that aren't rotating, and only the X-ray source is rotating. Or we're going to just look at that scan and know that that's a ring artifact, know that that's not pathology. And if it's not covering vital structures, we can get away with not having to repeat that specific scan.
So, that brings us to the end of this talk. I hope you found it useful to go through some of these common artifacts. Now, know that there are a plethora of other artifacts that occur in CT imaging, and some of those do come up in exams. And what I've done is, in the question bank I've linked below, I've asked questions about some of those artifacts, and then we'll go through those answers in video format. And over time, through that question bank, we will cover some of those more obscure artifacts. But for the sake of this talk, and for the sake of time, we're not going to go through all those minor artifacts.
So, that being said, that brings us to the end of this talk. I hope you've enjoyed it, and I'll see you all in the next one. Goodbye, everybody.