Transcription
Hello everybody, and welcome to this introduction to photon counting CT. The way I'm going to frame this talk is by comparing photon counting CT detectors to conventional computer tomography detectors that we're using in most CT scanners today. And I want to do this because I feel like photon counting CT has the potential, at least, to be a stepwise progression in computer tomography technology. There are fundamental differences between these two types of detectors. So, let's look at the diagrams that I've created for the two detectors. This is the conventional detector we're going to look at, and this is the photon counting detector. You can see they differ structurally, and they're going to differ in their mechanism for detecting X-rays.
I want to start this talk by looking at an analogy to show you, or to highlight the differences between the types of detectors. If we think of the detectors as being buckets of water, they have some water in them. I've got the conventional detector on the left that's got a finite bucket that doesn't leak water. All the water stays in the bucket. And the photon counting detector on the right that has maintains a constant level of water. And every time new water is added, that water is going to leak out, and the level is going to remain the same. Now, what we're going to do is fill these buckets with X-rays. We're going to fill them with raindrops in this analogy, and we're going to plot the changes in water within the buckets over time. I want you to think of the conventional detectors as being like a scale. It's weighing the amount of water that is going into the bucket. And think of photon counting detectors as a water level measure. We've got a little gauge on the side of this bucket, and we're going to measure the changes in water here.
So, let's look at the conventional detector. Water, or X-rays, are going to reach the detector over a certain period of time. A conventional detector is then going to say, "What was the weight change in the bucket?" We're using it like a scale and say, "How much water was added to that bucket?" And we can take the integral of the graph that we've generated here. The area under this curve is going to equal the amount of water that has reached the bucket. That's what conventional CTs do. They measure the total incident X-ray energy over a set period of time. And we set that by setting the number of projections that we want to take when we're rotating around the patient. There's a time interval when we're collecting data. It's not looking at individual raindrops; it's looking at the total number, total amount of raindrops that have reached the detector, or the bucket.
In this example, photon counting detectors are different. When the water in this example reaches the bucket, it's going to send out a ripple that is proportional to the size of the raindrop. The sides of the different raindrops in this analogy represent the different energies of the X-ray photons reaching the detectors. So, watch what happens here and see how it differs. As the raindrops reach the buckets, they send out a ripple, and the water level measure on the side here is going to measure the size of that ripple. Notice how different the graph is here. We're able to count the number of water droplets that have hit the bucket, and we're able to see the size of the water droplet, or the energy of the X-ray photon. Photon counting detectors have the ability not only to count the number of photons that have reached the detector but also to separate those photons into the energy of the X-ray photon. You can see here that as we are calculating the area under each one of these curves, the area under the curve represents the X-ray photon energy for an individual photon. And each blip on this graph represents a separate X-ray. You can see how we've been able to separate the X-rays out here, whereas in conventional detectors, we're just getting the total X-rays over a set period of time. We can't say anything about the individual energies. And this is the key difference between the two types of detectors.
So, let's look at conventional CT detectors and see how they differ from photon counting detectors. Now, in conventional detectors, we use what's known as indirect detection. X-ray photons, electromagnetic radiation, is converted into light photons. The same type of electromagnetic radiation, but now they've got longer wavelengths. Those light photons are then measured by what's known as a photodiode. And that light energy is converted into an electrical current that's stored. So, let's look at the components here. First, we've got what's known as the scintillator layer. We've got it's usually made of something like gadolinium oxysulfide or cesium iodide. And what that does, what that's responsible for, is converting X-rays into visible light. Surrounding the scintillator layer is what's known as the reflective septum. We've got these septa between each one of the detectors that prevent the light that's generated from spreading out to adjacent detectors. It keeps it on one single detector. And you'll see one X-ray photon makes hundreds, thousands of light photons. And those light photons spread out in 360 degrees. And we want them to be maintained to a single detector, that detector that the X-ray was incident on. That light energy is then converted into an electrical current by the photodiode layer. The photosensitive area is going to create or free electrons that are going to then run towards what's known as a capacitor and ultimately be discharged with the transistor here. That electrical current for each individual detector is going to be sent towards our computer for processing via a mechanism known as the application-specific integrated circuit. And the type of ASIC that we use in conventional CT detectors is what's known as an integration ASIC. We take the integral of the curve that we generated when the X-ray photons were incident on this detector. And that integral is the area under the curve, the total amount of X-ray energy that has reached our detector for a set period of time.
So, let's look at this head-on. We can see the photodiode layer here. Only this blue region on the photodiode is sensitive to light. The rest of the components here, if X-rays are incident on them, or light photons more specifically are incident on them, they're not going to contribute to our final signal. That's what's known as the fill factor. The difference between the area that we're actually sensitive to light versus the total area of the detector. The higher the fill factor, the more efficient our detector is because we're able to register more of those light photons. So, this photodiode layer has a silicon photosensitive area that's going to receive light. It has the capacitor here, which is going to store or accumulate the electrical energy or the electrons that have been freed in this silicon layer over time. We store and store and store until the transistor switch here gets a little bit of current. We provide a small amount of current to this transistor switch, and that empties the capacitor. The emptying of the capacitor is giving the total electric charge that we've liberated from the light photons, and that's going to send it towards the ASIC here. If we look underneath our detector here, we can see that the circuitry within the detector is linked to the ASIC that's going to send those signals off to our computer.
So, let's look at a practical example here where we're going to have incident X-rays hitting our detector. An X-ray comes in, and it's converted into hundreds or thousands of light photons. Those light photons hit the photosensitive area here, and a charge is generated. Light energy is converted into an electrical impulse. That charge is then stored in the capacitor. That's the function of a capacitor: to store electrical energy. The number of light photons is proportional to the energy of the incident X-ray photon. And therefore, the stored electrical current that we've sent from the photosensitive area to a capacitor is also going to be proportional to the incident X-ray energy. Now, we set a time interval where we're going to measure all of the X-rays coming in. And during that time interval, multiple X-rays are going to hit the detector and send out light photons because of that scintillator layer. Those light photons are going to be converted into electrical energy and continually fill up our capacitor. Here you can see during this representation here, over time, our capacitor is being filled up with stored electrical energy. At a given point in time, that stored electrical energy is going to be released through the ASIC and sent towards our computer. Now, remember, we're looking at just one detector here. This detector forms part of multiple rows of detectors, or an array of detectors. Each one of these detectors is going to be receiving different X-ray energies, and we're going to ultimately store different levels of electrical impulses. When we read out an entire row, you saw that the ASIC actually runs in between each one of these detectors. When we flip the transistor switch, we can actually read out an entire row of detectors. We're going to get a graph representing the electrical stored energy in the capacitors. And that differs depending on the detector location because of the differing anatomy that the X-rays have traveled through. We can then convert the stored electrical energy into ourogram. We have filled a row of theogram based on the incident X-ray photon energy, the number of light photons that were released, and then ultimately the amount of electrical charge that that generated on the silicon portion of the photodiode layer. That's how conventional computer tomography detectors work.
Now, computer tomography detectors, their main characteristics is that there's an indirect conversion. The scintillator layer produces light before we create an electrical impulse. That's going to differ from photon counting CT detectors. And secondly, there's energy integration. We're adding up all of the energy and then using that total amount to create our image. We're not looking at individual X-ray energies. Now, this leads to a number of shortfalls that photon counting CT detectors are going to try and address. The first being that we've got no complex energy discrimination. We're not able to accurately say what was the energy of the X-ray photons that were incident on our detector in that period of time. We've looked at dual-energy CT where we can see two separate X-ray energy levels, but we can't really go much further than that. Photon counting detectors, because we can look at individual X-ray energies, we're going to have much better spectral discrimination and much better tissue discrimination.
Ultimately, the next problem we have is noise, especially at low doses. If you cast your mind back to the analogy of the bucket filling with water, we're only measuring the total change in the weight of that bucket. We aren't measuring individual droplets that were falling into the bucket. So, if we say the bucket got a kilogram heavier, how much of that signal was due to just electrical background noise? It's hard to differentiate that out because we've just got a total value. We'll see in photon counting detectors, we've got a much better primary way to reduce the noise in our image. And we've looked at noise reduction techniques that apply to conventional CT detectors, but photon counting CT detectors are going to give us an added level for noise reduction and ultimately improve our signal-to-noise ratio.
Thirdly, we've got limited spatial resolution in conventional detectors. I mentioned the fill factor. We've got a lot of dead space here. Our septa, as well, are providing place for incident X-rays to hit and provide no meaningful signal. The detectors, as we'll see later, are also larger than photon counting detectors. We can also have light spreading between adjacent detectors and reducing our spatial resolution. And we looked at the factors that influence spatial resolution in the image quality lecture. When we look at photon counting CT, you'll see how some of those factors are reduced.
We have limited artifact reduction. Again, we can't really tell what energy or photons are coming through. So, an artifact such as beam hardening, where the beam's average energy gets much higher, we aren't able to accurately detect that in conventional CT detectors, and we need to provide post-processing algorithms to say, "Reduce the cupping artifact." We can't address the problem directly at the source. And lastly, the conventional CT detectors have limited dose efficiency, predominantly because of these previous shortfalls. We've got a certain amount of noise that we just have to deal with. We can't improve our signal-to-noise ratio past a specific point. Much of the dose that passes through the patient is going to reach parts of the detector that don't even contribute to our final image. When X-rays are converted to light, we get a loss of energy in that conversion. So, it's not a perfectly efficient process. Many factors here limit the dose efficiency of CT detectors.
So, let's then have a look at photon counting CT detectors and compare and contrast that to conventional detectors. You'll see here photon counting detectors are sandwiched between a cathode layer and an anode layer. And there's a large voltage potential that's applied across this detector here, in the order of hundreds of volts. So, we've got a large voltage potential, and sandwiched between these two electrodes is what's known as the semiconductor crystal layer. In our examples, we're going to use a crystal layer that's made of cadmium telluride, CdTe. You'll see that used often. Here's the chemical structure. And cadmium telluride is perfect for this example because it's got good X-ray stopping power. It's got a high atomic number. It's got lots of valence electrons, which you're going to see becomes important during the generation of signal here. It's stable at room temperature. It provides the perfect material for this photon counting CT detector. Of note, though, is that these crystal layers rely on them being perfect. If there's any defect in these layers, we're going to get a drastic offset in the accuracy of our signal.
Photon counting detectors also have an ASIC. You can see it running under here and it heads out. Now, these ASICs are different to the integration ASIC that we looked at in conventional computed tomography detectors because those detectors only receive the total amount of energy from a transistor that was releasing stored energy from a capacitor. Here, there's no capacitor, no transistor. These ASICs are measuring an electrical impulse that's created when an incident photon interacts with the semiconductor crystal layer. The predominant difference here between conventional CT and photon counting CT detectors is that this is a direct conversion detector. X-ray energy is converted directly into an electrical impulse. There's no light step in the middle. There's no scintillator layer here.
So, let's have a look at how that process works. We're going to show you how we actually go about measuring signal in photon counting detectors. We have an incident X-ray that is going to interact with the semiconductor layer. What happens here is that an electron in that semiconductor layer is going to be released from the valence shell that it's in. We think of the outermost electrons in the outermost orbital that's actually filled with electrons as valence electrons. They fill an energy band known as the valence band. If energy is deposited into that system, those electrons can get enough energy to be released from the valence band and enter what's known as the conduction band. Electrons in the conduction band then have the ability to move along an electrical bias, and we've created an electrical potential here between the cathode and the anode. So, as that electron is released from the semiconductor layer, it's going to move towards the anode here. It's going to move along that electrical field or that electrical potential. When the electron is released from the valence band, it leaves what's known as an electron hole. And the electron and the electron hole together is known as an electron hole pair. The electron hole is a potential space, a space where an electron can fill. And what we know is that in a specific atom, that electron space will be filled by an adjacent or surrounding electron. There's an energy differential that is being created, and a surrounding electron is going to fill that hole. The process of electrons filling that hole means that the electron hole is going to seemingly move towards the cathode. And I'm going to show you exactly how that process works. The movement of the electron holes towards the cathode and the electrons towards the anode is going to create an electrical pulse. And that pulse can be measured by the ASIC.
So, let's look at that process again. The X-ray hits the semiconductor layer. Electron holes travel towards the cathode, and electrons travel towards the anode. Let's zoom in on the structure of the semiconductor layer. You can see this crystal lattice that's been formed in the semiconductor layer. It's a highly organized, tightly packed structure. And I want to give you an extremely basic diagrammatic representation of electrons sitting within this structured crystal layer. When an X-ray is incident on this crystal layer, we get, say, the photoelectric effect or Compton scatter that causes an electron to be ejected into the conduction band. It's not just one electron; it's a cloud of electrons. We get, in the order of tens of thousands of electrons that are being released within this layer. That release of the electron results in a cascade of electrons being released, and we call it an electron cloud. Notice how that electron cloud is headed towards the anode. Now, we can see the electron holes in this diagram. Which electrons are going to fill these holes? Well, because of the potential that we've created across this semiconductor layer, it's these electrons that are going to move towards the anode and fill the hole. As they move towards the anode, look how the electron holes are effectively moving towards the cathode. Now, the electron holes aren't particles themselves; they're just potential spaces. And you see now those electron holes have moved towards the cathode, ultimately creating that electrical pulse.
Let's look at that more in real time. You'll see that the interactions often happen closer to the cathode. Electron holes move slower than electrons move. And that's why the cathode is on the patient side of our detector, and the anode is on the computer side of our detector. So, watch how this happens. Photoelectric effect or Compton scatter. The hole looks like it's moving towards the cathode, and the electrons actually do move towards the anode. You can watch that again. X-ray coming in, electron cloud being released, and holes moving towards the cathode. That's the foundation for photon counting CT.
Now, we can have X-rays of different energies reaching our detector, and that electron cloud that is released is going to be proportional to those X-ray energies. And ultimately, the electrical impulse is going to be proportional to the incident X-ray energy. And that's how we can not only discern separate X-rays but we can also calculate the energy of those X-rays. And that's the major key to photon counting: we've got much more spectral information. We've got much better information about the energy of the X-rays that have hit our detector. Photon counting detectors not only count X-rays but they bin the X-rays based on the degree of energy that they have. We can separate the X-rays into multiple different energy bins: high energy, medium energy, low energy bins, and actually X-rays maybe that are so low energy that we want to remove from our sample. So, hopefully, you can see that photon counting detectors are going to provide us with much more information than conventional CT detectors. That's going to lead to a range of benefits, and we're going to go through some of those benefits here, or advantages of photon counting CT.
Say X-rays have been incident on a specific detector, just one of our detectors, and we've read out the information with our ASIC here. We can see the number of X-ray photons, and we can see the incident energy of those X-ray photons based on the area under each one of these curves. The first advantage of photon counting CT, and I've mentioned it, is this intrinsic energy discrimination. We looked at dual-energy CT. We said that we could have two separate average energy X-ray sources applied to the patient, whether that's alternating X-ray sources or orthogonal X-ray sources. And we can see that different tissues respond to different X-ray energies based on their tissue composition. Bone is going to respond very differently to a change in X-ray energy, say, to liver. And based on those changes in response, or changes in Hounsfield units, to different X-ray energies, we could separate out those tissues in post-processing. But we only had two X-ray energy sources, and they were actually different X-ray spectrums heading towards the patient. What we can do in photon counting CT is separate each one of these X-rays into individual energy categories. Now, we've not only got dual-energy sources, we've got multiple energy sources depending on how well we can discriminate the different energy levels, and we can use each one of these. We can say, "We're only going to use the data from X-rays in this energy bin here," and see how the tissues respond. Now, let's see how different the image looks, how different the Hounsfield values are if we use X-rays only in this X-ray bin, this X-ray energy bin. And based on those changes in Hounsfield units, we can make assumptions about the tissues within the patient. And we can much better separate out individual tissues. Perhaps we're going to get to a stage where we can really easily separate out tissues with very subtle Hounsfield unit differences based on our ability to have intrinsic energy discrimination in photon counting CT.
Next, we've got a much improved spatial resolution. You'll see here in our example, the detector, the anode here, is much smaller than our conventional CT detector, which is much larger. We've got limiting factors with conventional CT detectors: one, based on the septa; two, based on the electrical components that are needed and the fill factor; and three, based on how large the photosensitive area is. We need to have a certain size of photosensitive area to get enough signal that's going to be able to separate signal from noise. Because we don't need those septa and we don't need electronics on the top of our anode here, we've got much smaller detector elements. And actually reducing the size of the detector elements, reducing the size of each one of those anode pixels, it's actually going to be advantageous for photon counting CT because the total number of photons hitting each anode is going to be reduced, and we're going to better be able to discriminate out those X-rays. And you'll see when I talk about some of the challenges we're facing in photon CT, why that becomes important.
We've also got a much better ability to deal with dose. You see, when the X-rays are converted into light, there's an inefficiency there; there's a loss of energy. So, we're going to need higher exposure to get the same degree of signal as we would in the photon counting CT. Again, I've mentioned this now multiple times, there's dead space where X-ray incident X-ray photons are going to produce light that don't even contribute to our signal. We'll see later how noise is much less in photon counting CT. We don't need as high a dose, and the noise really predominates at lower doses. And if we have a good ability to reduce the amount of noise, we can actually use lower doses. So, it's better dose efficiency in photon counting CT.
So, how then do we go about actually reducing the noise? I said an advantage of photon counting CT is noise reduction. Well, it's a function of this intrinsic energy discrimination. We know that noise occurs at high frequencies and at low X-ray energies. It gets confused with low X-ray energy photons. What we can do is we can apply a threshold and say, "You know what, X-ray energies under a certain point, we're not going to use in our image." That's going to get rid of the noise within our image. We can still manipulate the data that we have here, especially because of the different energy bins, to create an image that has still sharp resolution and high signal, a good signal-to-noise ratio, but we've drastically reduced the amount of noise by not including those lower energy X-ray photons. Conventional CT just has a total sum of X-rays that hit it. It can't take out lower energy photons and include higher energy photons. That's a major point for noise reduction in photon counting CT. The electronics here also have less electronic noise as opposed to photon counting CT because each one of our data points are much smaller. We're not accumulating a whole load of data and then amplifying that signal as we release it out towards our detector.
Not only can we reduce noise, but we can actually reduce many artifacts that we get with conventional computer tomography. Here, what I've tried to represent is a beam hardening artifact where we've got a detector that's getting all of these different X-ray energies. I've just clumped them together, and we've got a predominance of these high X-ray energy photons. And we can know in our algorithm or in our system that our incident X-ray, our average incident X-ray energy is going to be X. We've set the KVP, we've set the MAS. But the average X-ray photons that are hitting our detector are much harder. We can assume that beam hardening has occurred. What we can do, because we're able to separate the X-ray energies, discriminate between X-ray energies, is we can remove the higher X-ray energies and we can amplify those lower X-ray energy photons and compensate directly at the source for beam hardening. We can say, "This area had a lot of beam hardening. Let's remove those higher X-ray energy photons that are going to cause algorithms to falsely calculate the Hounsfield units as lower values." When we got those dark regions between two highly attenuating structures like the petrous bones, we got darker values between them, lower Hounsfield unit values, because of that beam hardening. We can prevent that from occurring by removing those harder X-rays, those higher energy X-ray photons. That's one example of how we can remove artifacts in photon counting CT. And because of the sheer volume of data that we're getting, we have potential to address many other artifacts.
And that brings me to the final advantage in this talk that we're going to talk about is this enhanced post-processing potential or AI integration. Big data that we now have the ability to manipulate more than the data that we had originally received in conventional computer tomography detectors. We saw in conventional CT detectors, we filled a sinogram, and it was this data that we used to generate the image of one slice. Now, we're not getting a single reading per detector in each line of the sinogram. Each detector is outputting multiple different data sets. It's outputting how many X-rays are hitting it, and it's outputting the energy of each one of those X-rays. That extra data, as we've seen with some of these steps, is very powerful in post-processing. It's very powerful in us saying, "What's the tissue composition of each voxel within this image that we're creating?" We've got much more information, and as a result, there's a much higher upside to the potential of the data that we're receiving.
So, why then doesn't every hospital have a photon counting CT? Well, firstly, this is a new technology. It was only approved in 2021 by the FDA, and obviously, new technology is expensive. So, that's always going to be a prohibiting factor. But there are actually certain challenges that photon counting CT detectors face that we still need to overcome or still need to improve on. The first challenge that we face is what's known as pulse pileup, or a count rate limitation. If an X-ray photon hits our detector, we get an X-ray pulse, and another X-ray photon hits the detector, we get a second pulse. And those pulses are proportional to the X-ray energy. What happens when they hit the detector at very similar times or very close to one another? The two photons strike the detector almost simultaneously. Our output is going to look something like this. We've got overlapping of those electrical impulses, and it's very difficult to separate them out. We do have algorithms that can try and separate them out, but those algorithms aren't perfect, especially if you've got more than two X-rays hitting the detector. There's a limitation into the rate at which we can differentiate between different X-rays hitting the CT detector. Here, this is what's known as pulse pileup.
Now, pulse pileup is subtly different to our second challenge, which is known as adjacent pixel charge sharing. Now, we've got one X-ray photon creating a cloud of electrons that head out towards our detector, but they strike two adjacent anodes, and those anodes are going to share that electrical signal. So, instead of the signal that we were expecting for that specific X-ray energy, we've now separated that signal into two adjacent anodes. So, we are getting a reduction in spatial resolution, and we've got an incorrect signal here. We're going to say that this X-ray photon was half the X-ray energy that it was truly. So, we're not truly representing that X-ray energy.
Our third challenge that we're facing is what's known as detector dead time. If an X-ray strikes the detector and then another X-ray is incident on that detector close to that initial interaction, we may have a point where that second X-ray is not even registered. It's passed through the detector. There's a very small period of time when that detector or that anode is not sensitive to a second pulse, and that's what's known as detector dead time.
In an ideal world, the number of photons that were incident on the X-ray detector and the number of photons that we actually detect would be the same. If a hundred photons hit the detector, we would want to register a hundred photons in our computer. We want a perfect match-up, and that graph would look something like this: a linear graph where the more photons incident would equal the more photons detected. In practice, we've got what's known as limited dynamic range. At high X-ray flux, at a high number of incident X-ray photons, we get a disproportional reduction in the amount of photons that we actually register and send to our computer. And that's a function of the dead time and a function of pileup and count rate limitation. What happens here is our graph ends up looking something like this. When we have high photon rates that arrive too quickly at our detector, we aren't able to process each one of those X-rays individually. And not only do we have the dead time or the pulse pileup, but our ability to process the signal, process those electrical impulses that are heading towards the ASIC is limited. There's a limited bandwidth there. There's a limited sampling frequency that we have there. If we exceed that, we get this drop-off in our dynamic range here. We don't have the ability to truly represent high areas of X-ray flux.
And finally, the last challenge that we face is the complexity of this new technology. We don't quite know exactly all of the things that we can do with this current technology. Cost is always going to be prohibitive. And now we're getting so much more data. And that data is going to require much more advanced signal processing. That individual detector is not giving us one readout value per projection like in conventional computer tomography. It's giving us multiple readouts for each and every projection. It's not just a doubling of the amount of data; it's an exponential increase in the amount of data. Manufacturing these components are also much more expensive currently, and the hospitals need to justify the cost, whether it's beneficial to the hospital or whether it's beneficial to the patient. And because this is a new technology, there aren't that many providers here. We need competition to drive down the cost.
And I've included this graph here. This is what's known as Moore's Law. And Moore predicted that in the '70s that the number of transistors that are used in a chip is going to double every two years. And that's pretty much what has happened over the last 50 years: that the chip size has gotten smaller, and the number of chips and our processing power has gotten exponentially bigger. This is an exponential scale on the y-axis of our graph. Whilst all of the requirement is increasing, we get this drastic reduction in the cost to manufacture these components. And we can assume in CT or any technology that we're using in radiology, as our demands increase and as technology gets better, we're going to drive down cost, and these kinds of technologies are going to become more and more affordable. And hopefully, the applications that we can use photon counting CT is only going to increase. And as I said, this has the potential to be a stepwise increase in the technology from conventional computer tomography to photon counting computer tomography.
So, I hope that's made sense. This is just a broad bird's-eye view introduction to photon counting CT. As the technology develops, I'm sure we're going to get an expansion of the underlying physics that contribute to this technology. So, that brings us to the end of this learning pathway. I really have enjoyed sharing this information with you, and I hope the information has been helpful to you. If you've enjoyed this course, please be sure to let me know. I've linked a question bank below if you want to now test your knowledge in an exam setting where I've taken actual past paper exam questions and I answer those questions in video format, showing you why I've asked the question that way, how the question could be asked differently, why the other options to the questions are wrong. It's a really valuable resource if you're studying for a specific physics exam. Until next time, I'll see you all. Goodbye, everybody.