Transcription
Hello everybody, and welcome back. Today, we're going to be looking at the hardware of the CT machine and how it's changed and developed over time. And you'll notice that each major improvement in the CT machine itself results in a drastic improvement in image resolution and a reduction in the amount of time it actually takes to create an image.
Now, each step-wise improvement is marked as what's known as a generation of CT scanner, and we're going to look at the four major generations here today. Now, you may have heard of fifth, sixth, even seventh generation CT scanners, and those more complex CT scans, we're going to look at in future talks when we look at dual-energy CT or electron beam CT imaging. But for now, we're going to focus on these four. You can see the four generations here, and to date, we've been focusing on this generation, the third generation of CT scanners. This is the most commonly used generation in current CT imaging.
Now, when we go about defining these generations, we look at three major characteristics. The first thing we look at is the X-ray beam geometry. What's the shape of the X-ray beam? You can see we start with a pencil beam, and that beam gets wider and wider. We get to what's known as a fan beam here. The next thing we look at is how the X-ray source and the detectors move in relation to one another, as well as in relation to the patient. So, beam-detector movement here. And the third and final aspect that we look at when classifying CT generations is how the detector is configured. Do we have a single detector, or do we have multiple detectors that span all 360 degrees around the patient? And it's changes in these three components that are going to define what generation a CT scan falls into.
So, let's start at the very beginning with first-generation CT scanners. This was used by Sir Godfrey Hounsfield when creating the first CT scanner. At the time, it was called an Emi scanner. We've got what's known as a pencil beam. This is pencil beam X-ray geometry, a very narrow beam of X-rays that's been highly collimated, and those X-rays are practically parallel to one another. And as the beam moves across the patient, all the beams that we create are parallel to each other as we scan across the patient.
Now, that movement across the patient is what's known as translation, and the X-ray beam translates across the patient and then rotates slightly and translates again across the patient. So, its movement is what's known as translate-rotate movement, and I'm going to show you what that looks like in the next slide. In terms of the detector, we've got a single detector here detecting this pencil beam here. In the first couple of scanners, it was either a single detector or two detectors creating images from different slices, but there were very few, only one or two detectors in the entire CT machine.
So, let's have a closer look at the X-ray beam itself. You can see the collimator here has both decreased the X-ray beam in the Z direction as well as in the XY direction, and we've got this narrow pencil beam that focuses down on one detector here. Now, if we were to place the CT machine next to a patient here, we've got an abdomen and we've got two arms on either side. Again, a very simple patient because they've got the same attenuation throughout the body. We haven't got different organs in this example creating differential attenuation.
Now, I've said what's going to happen is this beam is first going to translate across the patient. As it translates across the patient, we're going to fill one line of the sinogram. Let's have a look at what that looks like. Now, we're moving across the patient in the XY direction, and we're gathering that data, digitizing it, and placing it in the first line of the sinogram. This represents 0 degrees to the patient. That's what we're going to start at. We can see the analog data here that would have been digitized for this line in the sinogram.
Now, this analog data across the width of a patient, giving us one line in the sinogram, is what's known as a projection. We need multiple projections to generate all those simultaneous equations we were talking about to be able to calculate or reconstruct this image. So, what's going to happen is after this beam has translated across the patient, it's going to rotate by one degree and then translate again across the patient. That's why it's called translate-rotate geometry. Now, I'm going to show you this after multiple iterations.
Now, we've rotated 45 degrees. We've done 45 iterations since the first one. Again, the beam is going to translate across the patient and fill the sinogram that corresponds to 45 degrees. Notice how our sinogram has changed. Our analog data has shown those arms getting much closer to the body here because of the angle that we're cutting through this slice. Again, this is repeated multiple different times. It's repeated for 180 degrees of the patient. Here's an example at 90 degrees. Notice how the sinogram is changing. We get very high attenuation values in the middle here because we're passing through both the arms and the abdomen in our example here, and look how our sinogram has changed.
Now, in first-generation CT scanners, this is done for 180 degrees, moving at one-degree increments, and we generate a sinogram that looks like this. In the first generation of CT scans, we were taking 80 readings across the slice for the patient, and ultimately, we were creating an image that had a resolution of 80x80 pixels. So, if we were to place this sinogram into our reconstruction algorithm, this is the image that we would generate. You see, with these first-generation CT scanners, we're not getting a very high-resolution image, but you can see what the initial image looked like.
If we were to take a CT scan of an actual patient here, say of the head, this is what the CT scan would look like. And if you look back at the first ever CT's image, it looks something very similar to this, a very low-resolution image, but it's got enough information. We can see the midline, we can see that there's a mass that's enhancing, we can see the shape of the head, we can see the bones here, and we can see that there's potentially edema or fluid around this mass here. So, there's not no information, but obviously way below what we expect in current terms.
Now, it was very soon after first-generation CT scanners that we realized we could allow the beam to fan out a little bit more and cover more area within the patient, so long as we could get multiple detectors in a row next to one another. Now, the second-generation CT scanners created a single array of detectors, generally had 30 detectors in a row here. The collimation in the Z plane was exactly the same. It was still the width of one detector, but now we've allowed for less collimation in the XY plane, allowing that beam to fan out. So, we've no longer got a pencil beam. We've got what's known as a narrow fan beam. It's narrow because we've collimated it both in the Z direction, just to the width of a detector, and we haven't allowed that fan beam to go out to what current CT scanners have, say, 60 degrees. It's still quite narrow, just to fit on those 30 detectors.
The movement in second-generation CT scanners is exactly the same as first generation. We translate across the patient, rotate slightly, and translate again. However, that rotation no longer needs to be only one degree. We can afford to have a greater rotation and fewer projections around the patient because we've got more detectors. If you cast your mind back to our introductory video in this course, we said it was the number of simultaneous equations that we can generate that will determine the resolution in our image. Because we've got now 30 detectors instead of one detector for each projection that we're creating through the patient, we're getting 30 equations, no longer one equation. And it's those extra equations that allow us to do more rotation around the patient, not having to do smaller increments like in first generation. And we've talked about our detector configuration. We've now got a single-row detector array with up to 30 detectors in this array.
So, if we look closely now, we've taken out those lateral collimators that made our pencil beam, but we've still kept these collimators just to the width of our detector. And as we head down towards the detector, you can see we've got multiple detector elements here. I've got 17 so that you can see the discrete elements. This can go up to 30 elements. Now, if we were to place the second-generation CT scanner towards our patient, you would think that now that we've got 30 detectors, isn't this translation time going to be 30 times faster than our first-generation CT scanner? If we translate across the patient like this, we've got 30 detectors, 30 detectors, 30 detectors. We only need three separate images across this entire slice. As before, we would have needed 90 different projections with the single detector.
But that's not the case. If you look closely at where this fan beam is going, we've got large regions of anatomy that aren't being covered by this beam. This beam is no longer a parallel beam. We can't place the detectors on end like this to each other. We need to overlap the detectors. We need to squish down our translation times so that all of our X-rays are covering all of the anatomy in the patient. We're still getting a market increase in the speed that we can translate across the patient. We're still getting overlap that's more than one detector with. But because of this fan beam geometry, we have something extra to account for. We get magnification as we're passing through the patient, and we're going to look at various different geometries in CT scanning in the next talk.
This fan beam becomes incredibly important. But again, we translate across the patient like this, collecting data, and as we've translated across one entire slice, we get what's known as a projection that can fill a single line of the sinogram. We then rotate the beam, now more than one degree in second-generation CT scanners, and simply translate again, filling another line of the sinogram. Now, we get a rapid improvement in speed here. In the first-generation CT scanners, we were looking at a scan taking anywhere between 5 minutes and 30 minutes. In this CT scan, we're looking at 30 seconds to 4 minutes. We've got a rapid increase in the speed that we can take these images, one because of those extra detectors and our ability to translate faster, but two because we can rotate by more angles in one degree. We need fewer projections around the patient to reconstruct this image.
Now, our first-generation CT scanner, we had 180-degree sinogram, creating an image that looked like this, 80 pixels by 80 pixels. Because of those improvements in second-generation CT scanners, look how our sinogram has changed, and our image now has doubled in resolution. We've got a 160x160 pixel image. Notice the improvement. We can start to see some gray-white matter differentiation. We can better classify this mass. We can see the midline shift even better, and our bone resolution has got better. All of this resolution improvement has happened, and we've acquired the scan much quicker. We've got a reduction in potential for motion artifact and things like that. We can see this is a stepwise improvement.
We now move on to third-generation CT scanners, which we're a little bit more familiar with because we've been looking at this in previous talks. Our beam now is both fanning out in the XY plane much more than it did in second-generation CT scanners, and we've increased the width in the Z plane in the longitudinal axis of our patient because now we've got multiple rows of detectors that lie in this array here. So, our beam is what's known as a fan beam.
Now, in terms of the beam-detector movement, we've seen that the X-ray source and the X-ray detectors lie on the same gantry, and they rotate together. There's a fixed geometry between the two, and slip-ring technology allows for continual rotation and for continual production of X-rays. This is what's known as rotate-rotate geometry. So, now we've got a change in the movement of our X-ray machines in third-generation CT scanners. And our detectors now have a curved array. They've not got a linear flat array. They've got a curved array, meaning that the X-rays lie perpendicular to the detectors, and the distance between each detector and the X-ray source remains constant no matter where they lie on the XY plane. And that helps with some of our geometry. Our geometry is no longer parallel beam, and it's no longer the geometry that we saw in second-generation CT scanners, which had a flat detector where we were having to calculate the hypotenuse of triangles to get the distance between the X-ray source and the detectors. Now, we've got radial geometry, and our radius of the X-ray beam is remaining constant throughout the detectors here. So, this is what's known as a multi-row curved detector array. May have heard of MDCT, a multi-detector CT, that's what this is.
So, let's have a closer look here. We've increased the distance between those collimators, now allowing for more detectors to receive X-rays, and we've got these multi-detector CT arrays here, multiple rows within our detectors. You can see how the width of the detectors here means that we've got a larger coverage per slice of the patient, and that's going to lead to a rapid increase in the speed that we can acquire these CT images because our slices are much thicker here.
Now, let's quickly review again how the data is acquired in third-generation CT scanners. We've seen this before. One single X-ray beam is going to fill a whole line of the sinogram. We've got an entire projection here. We're covering all of the anatomy. Not only are we filling one line of a sinogram, we're filling one line of a sinogram for multiple different rows for multiple different Z-axis locations. So, we're filling multiple different sinograms that correlate to different axial slices within our patient.
Now, beforehand, each translation created a single projection. In third-generation CT scanners, the projections, the lines of the sinograms, are determined by how rapidly we can sample our detectors. Every time we can take information from the detectors, sample it, and digitize it, and place it in the sinogram, that creates one projection. In that time, the X-ray machine is going to rotate slightly. The time that we can then sample those detectors again determines how many projections we can get in the 360 degrees around the patient. And in modern CT scanners, that's between 1,000 and 3,000 different projections. As we're rotating around the patient, we're generating loads more data than we were previously. And if you think about 1,000 or 3,000 projections per slice, the amount of data that we're generating when we looked at first and second-generation CT scanners, those were developed before the CD was even invented.
Now, you can see how the advances in software, our computational ability, has allowed us to get all of this extra data and allowed us to improve our resolution drastically whilst drastically decreasing the time it takes to generate one of these images. So, let's have a look at that improvement in resolution. And remember, this resolution improvement comes at the drastic reduction in time. When we're looking at second-generation CT scanners, as I said, anywhere between 30 seconds and 4 minutes. Now, we can rotate around the patient, one slice of the patient, in 0.3 seconds. We can get an entire CT scan in just 5 seconds. We saw this improvement from first generation to second-generation CT scans, and it was a marked improvement, a doubling or bettering of resolution. Now, from this 160x160 pixel image, we can go to a 512x512 image here. Clinically, this improvement is massive. We can see how much resolution we've got in the bones here. You can imagine if we're looking at fractures, especially once we've windowed this properly, how much better we'd be able to identify fine fractures in this image as opposed to this image. We can still see these stepwise pixels on the edges of the bones themselves here. We can see the gray-white matter differentiation much better than we could in previous images. We can see individual vessels. Our improvement has been marked between second and third-generation CT scans.
Now, there is an issue with third-generation CT scans. We've got a reduction in motion artifact, but we've got what's known as a ring artifact that can form, and that's unique to third-generation CT scanners. Say, one of our detectors was faulty here. It was giving us either incorrect information or giving us no information at all. Notice as the CT machine rotates around this patient here, where this black line is, where the faulty detector is, I'm going to show you where that's at with a white line that appears as this rotates. We see the CT machine rotating here, and this white line is showing where the faulty information is coming from. I've spaced it out a little bit so you can see, but this would be a continual white line as the CT machine rotates. Notice how we're getting no information, especially in this distinct circle here. I can actually draw a circle there. This is what's known as a ring artifact. Anatomy that lies on this circle is only encoded for by this single detector, and it's a faulty detector, so we're going to get no useful information on this ring here. We're also going to get artifact in the peripheries here because of the projection data created by this single detector.
Now, this was a major issue, especially in the past when detectors weren't as stable as they are now, and we didn't have computer algorithms to account for this ring artifact. I'll show you what it looks like clinically. We can reduce the size of this and see this image here on our right. Notice this ring being formed here, what looks like a highly attenuating ring going through the axial slices of the patient. The ring is normally near the isocenter because of the geometry of the X-ray beam, and it's at a fixed distance from the isocenter. Now, this is a classic appearance of a ring artifact. If we were to reconstruct this image with a multiplanar reformat and create the coronal image here, you can see what the ring artifact looks like. Also, notice as we're scrolling through these axial slices, how the ring seems to only appear as a semicircle that's winding through the patient. That's a function of how the patient is moving through the CT scanner as we're acquiring the images, and it should be a clue to us that a single rotation of the CT machine doesn't account for a perfect axial slice through the patient. It accounts for a spiral going through the patient. That's because the mode of acquisition we're using here is helical or spiral, and that's something we're going to look at in a lot of depth in two talks' time when we look at modes of acquisition.
Now, the next generation of CT scanner actually accounts for this ring artifact, and that's why ring artifact is unique specifically to third-generation CTs. Obviously, in first and second-generation CT, we're not going to get ring artifact because we're translating across the patient. That detector is moving across different regions of anatomy, not accounting for just one ring of anatomy.
In fourth-generation CT scanners, we've got the same beam geometry. We've got a fan beam, the same setup here, but now only the X-ray source rotates, and the detectors remain stationary. Our movement is what's known as rotate-stationary. You can see we no longer have an arc of detectors. We've got a full 360-degree coverage of our detectors. So, this is a multi-row 360-degree detector array. We've still got multiple rows. We're still covering a thicker slice like we were in third-generation CT scanning.
Let's have a look at how this works. Now, again, one projection is acquired by the single fan beam. Not much has changed here between third and fourth-generation CT scanners. We're still acquiring the data, but now that data acquisition is happening from different detectors, not from the same detectors rotating at the same time as the X-ray beam. We still get 360 degrees of data filling our sinogram here, again, 1,000 to 3,000 different projections. We can often have up to 2,000 different detectors around the patient here.
Now, the problem with fourth-generation CT scanners is that detectors themselves are extremely expensive, and they were actually created because detectors were extremely expensive. And if in a third-generation CT scan, we had one detector that was faulty, we had to replace that detector because of the ring artifact here. In fourth-generation CT scanner, if we have one detector that's faulty, again, we're going to create a single line of incorrect data. Now, watch how that single line, though, as the X-ray source rotates, that single line fans through the patient, and it doesn't create a ring like it did previously. I'll show you what that looks like now. It's the faulty detector that's creating that projection data. All of this data that's being acquired now doesn't include that faulty detector at all. It was only that initial pass through the faulty detector that had those specific lines of faulty information. That faulty information didn't create a ring either. It wasn't responsible for one plane within this patient. It fanned across the patient, and when we look at reconstruction algorithms later and how we go about generating an image, processing an image, you'll see that the mathematics involved here would mean that this faulty detector was averaged out. It became inconsequential, and we would have got rid of the ring artifact.
Now, fourth-generation CT scanners, I've mentioned, were incredibly expensive. And nowadays, we've got computer algorithms that can account for a faulty detector, can account for that difference that was causing the ring artifact, and detectors themselves now are much more stable than they used to be. And that's why third-generation CT scanners remain the mainstay. There are hospitals that use fourth-generation CT scanners, but because of their cost, now we've got 2,000 detectors, the rate at which detectors are going to become faulty is exactly the same, but now we've got more detectors to become faulty. It ends up being more expensive to acquire and more expensive to maintain.
So, that's the four generations of CT scanners from 1 to fourth. Here, we can see that it was initially taking us from 4 to 30 minutes to acquire an image. Not only that, but the image had to be processed overnight. The computational power just wasn't there, so nothing was happening quickly. In first generation, that acquisition time decreased drastically as we move through the scanners, and we can see that represented by this graph here. Not only was acquisition time decreasing, but our resolution was also increasing rapidly. And that's how CT scanners have advanced in actually a relatively short period of time. This happened in just 50 years here. And these advances weren't only in just the hardware, but as I mentioned, the advances happened in the software, not only being able to process all that data but also being able to account for the varying geometry. Here, the geometry was fairly simple. First up, when the first Emi scanner, Godfrey Hounsfield, did that first scan, the geometry was parallel. And in fact, the mathematics for that was done by a gentleman called Allan Cormack, who studied just 100 meters down the road from me here. Both of those people winning the Nobel Prize for their work in CT scanners. The ability to account for the changing angles of the X-ray beams and how the magnification and the projection angles change as they pass through the patient was a major advance in the CT machine itself. And that's what I want to touch on in the next talk is looking at some of the geometry of the fan beam passing through the patient and how that's going to influence the data that we're receiving. That becomes important later on when we look at the modes of acquisition, both axial and spiral acquisition, in the next talk. And it becomes even more important when we look at image reconstruction algorithms, when we look at back projection, when we look at Fourier reconstruction, and when we look at iterative reconstruction methods.
Now, it's been my experience that comparing the different generations of CT scanners is a classic question that comes up in exams. It's a really easy question to ask and a really easy question to mark. I can see why examiners put it in there. And for me, not only is it an easy question to ask, it's an easy question to answer. It should be marks that you just know you're going to get right in the exam. It's something that I will cover in the question bank that I link below this talk. Otherwise, I'll see you all in the next talk where we really take a deep dive into the geometry of the CT beam itself. Until then, goodbye everybody.