Thank you so much, Dr. Leung, for the invitation to come here. I'd like to express my gratitude to people who have contributed to the material to be presented today, including my colleagues at University of Wisconsin-Madison, and our industrial partner from Verox Imaging, and also Dr. Radim from University at Buffalo, Dr. Schart from the TU Delft in the Netherlands. This educational talk, I will focus on addressing the following question. What is photon counting detector, and why is it beneficial to CT imaging? In my home, there's a fifth grader also asking me a similar question, because she overheard that's what I do for a living. What is photon counting? Because she likes gardening, and then I have the following explanation to her. Imagine all the tomatoes, and then you can harvest using a harvester, like a machine like this, into a bowl, and then if you put all the tomatoes into a bowl, and then you can measure the total weight, and then that's an integration process. Now a problem with that is the total weight of all those tomatoes are heavily weighted by the large tomatoes, even though we know usually the smaller tomatoes, they have a better sales price, they're more valuable. In this case, you can see the unit price of those cherry tomatoes is actually much higher than the large tomatoes, but then if you do the integration, all the weight are dominated by the large tomatoes, so can we do better? Now if you can start to count individual tomatoes, and then you can sort them into different bowls based on their size, and this actually can lead to very favorable outcomes. And in the context of x-ray imaging, the favorable outcome is explained as follows. What's shown here is a contrast, attenuation contrast between two soft tissues as a function of x-ray energy. Now we know there's a similar to tomato picking example, those low-energy photon, they are small, but they are more valuable for CT imaging because they carry much higher image contrast. Those higher-energy tomatoes, they look strong and big, but then they don't have too much of a contribution to the contrast signal, their primary contribute to the noise in the image. Now in energy-integrating detector, it's pretty unfortunate that the total weight, the total signal collected by those type of detectors are heavily dominated by those high-energy photons, and then which decrease the contrast to noise ratio. And then when we harvest a tomato using a machine, and then through an integration process, there's some noise can be collected together with those fruits. Now if we, in x-ray imaging, a similar thing could happen if we're using energy-integrating detector, we'll collect everything, including the signal induced by x-ray photon, as well as those electronic fluctuations from the readout circuits, as well as the sensors. Now if we, in a photon counting detector, we can collect an individual pulse induced by x-ray events. Now if you set a threshold, and then you can reject those undesirable noisy signals. So when we watch a basketball game, we usually do not just looking into the integrated score per game, although that's also valuable. If you can sit into a game and watch through the entire game, actually you can get more information. You can draw a line between the three-pointer and two-pointers, and then you can gain more insight into an individual player's performance and behavior. Now in the CT imaging or x-ray imaging, you can do a similar thing. So if you can count individual x-ray photons and then draw a threshold between low and high-energy photon, and then you can see, in this case, you actually can help you differentiate different type of materials, even though the total integrated photon number is the same. Now in detector with a perfect, if we have a detector with a perfect speed resolution, a photon should be completely absorbed in the location where we initially hit the detector. In reality, however, we know the x-ray photon can be scattered or can emit a k-fluorescence photon after photoelectric events. And then in this case, the scatter photon could be absorbed in a different location, and then that leads to speed resolution loss. Note that this kind of events can happen in both semiconductor and scintillator detectors. In a semiconductor detector, the x-ray-induced charges could expand spatially, and then in the scintillator detector, we know the light can also spread spatially. And in an energy-integrating detector, those kind of effects are known as spatial resolution loss. And then now in a photon-counting detector, sometimes they're referred to as charge sharing. All they're doing is to lose some spatial information or spatial resolution. So how do we, and then now, you can imagine, if we operate all those detectors under the photon-counting mode, and then we actually can thresholding those spatially-spreaded signals. In other words, in the neighboring pixel where you induce some undesirable signals causing spatial resolution loss, now if you draw a threshold, you can reject those spreaded neighboring signals or counts. For example, here we measure the modulation transfer function of our VAREX Cat-Tel photon-counting detector. As you can see, as we increase the energy threshold from 15 to 50 keV, you can see an improved MTF, suggesting that this indeed helped by doing a thresholding to help improve spatial resolution. So if you're a PCD vendor, I suggest you just go back and operate your PCD under energy-integrating mode, and I believe you'll see a change or degradation in spatial resolution by operating under energy-integrating mode. To count individual photons, naturally, a detector must be fast, and because it is fast, and then you have a better temporal resolution. Now your frame rate can be extremely high because you're already being able to count all those million counts per second of photons. Now the frame rate can be extremely high. In fact, many years ago, my colleague had to run a Cat-Tel PCD at 600,000 frames per second in an inverse geometry X-ray system. And then here is a demonstration of the benefit of running a high-resolution, high-temporal resolution PCD. Dr. Rubin's group at the University of Buffalo, they run a VAREX Cat-Tel PCD at 1,000 frames per second for imaging the flow pattern in the aneurysm model. Here are the results acquired with a PEPLI embolization device flow diverter placed inside the aneurysm model. You can see the PCD allows physicians to see those detailed iodine contrast flow pattern and the vortex. The detail and clarity seen on the PCD image on the right are unprecedented and has a potential to impact the outcome of the intervascular interventions. Now photon counting detectors are sometimes confused with direct conversion detectors. Whether a detector is direct conversion or indirect conversion is determined by the radiation sensor, while photon counting is a signal readout mode. In other words, a direct conversion can be an energy integrating detector and an indirect conversion detector can also be a photon counting detector. And semiconductor detector is just a subtype of direct conversion detector and it has the advantage of better speed resolution and signal to noise ratio. Now as shown by this family tree of photon counting detectors, we can also build a photon counting detector using gas, using scintillators, and other materials other than semiconductors. The real deal actually lies at the intersection between semiconductor direct conversion and the photon counting detection mechanism. So a semiconductor based photon counting detector enjoys the benefit of both the semiconductor sensors and also the photon counting mechanism such as electronic noise rejection and then spectrum resolving capability. So why didn't we use the gas or scintillators to implement photon counting detector for CT imaging? One major drawback of the gas or scintillator that they produce a very low signal per x-ray photon. And why does this matter? So when we're looking at the electric pulse readout by electronics, there are background noise superimposed in the radiation-induced pulse. Now if the signal induced by the radiation is too weak, then you can see the precision of the pulse height measurement, namely the energy resolution, or the pulse width measurement, namely the timing resolution, can be really poor. Whereas gas, the scintillator can do better, but neither scintillator nor gas, and anywhere close to semiconductors in terms of the number of secondary quanta induced per input radiation quanta. So now we can, as emphasized by Glennell multiple times, the use of semiconductor materials as radiation detectors can result in a much larger number of information carriers for given radiation events than is possible with any other common detector type. And the more charges, we mean a stronger electric signal, and therefore a more efficient conversion of radiation into electricity. Meanwhile, not all semiconductors can be built for a photon counting detector for medical CT imaging applications at room temperature. And then first of all, their effective atomic number and density affect x-ray's absorption efficiency. And then also the so-called electron and hole mobility also matters, because they determine how strong an x-ray induced the signal is. So what are holes, and why do they matter in a photon counting detector? So imagine there's a two-way divided highway, and then we have a severe traffic jam along one carriageway, fully packed with cars. And then you can treat each car as an electron, and then the engine in each car can consider as an electric force from applied voltage. Now since all the cars are packed in one carriage lane, they cannot move. So they cannot form electricity. Now if we have some kind of excitation from radiation, we can excite one car into the other carriage. Now because that's empty, this car with the engine, they can start to move, namely forming electricity. Now for the rest of the car in the original carriageway, and then you can see they start to move. And then the empty spot actually start to move towards the opposite directions, which can also form electricity. Actually in a scintillator, a very similar thing is happening. One carriageway is fully packed of all those junk cars without any engine, so they cannot move at all. Now if we have a forbidden, we also have a wider forbidden central reservation. Now if a car is receiving energy from external radiation to go across a forbidden gap, it does not move down the road since it does not have engine. What it do is naturally fall back to the original carriageway, during which they emit a light photon due to energy conservation, instead of trying to generate any electricity directly there. Now in a semiconductor PCD, the electric signal is generated through the motion of both electrons and hosts, instead of the physical arrival of charges on each electrode. For a negative charge electron, it can induce a positive charge on the surface of an electrode, especially the anode. When the electron moves closer to the anode, you can see it induces a charge to redistribute on the anode, such that more positive charge goes to the central pixel, therefore the central pixel will experience a positive current flow. The faster the electron moves, the quicker the induced charge redistributes, therefore the stronger the induced current is. Now if we consider a positive charged hole moving away from the anode, it also can induce negative charge on the same anode plane. When the hole moves away from the anode, the induced charge, negative charge, on the anode become more and more uniformly distributed, therefore for the central pixel, it start to lose those negative charge, which is also a positive current flow in the same anode. So you can see both electron and hosts contribute to the signal formation in a PCD. So at our university, to convince our students that output of signal of a PCD is indeed form a non-contact electric induction, we drop a charged puck towards a rubber ball, which serves as an insulator from our bottom metal electrode, which is similar to an anode in a PCD. Now since this is electrically insulated from the anode, the charge in the puck never can directly transfer the charge to the electrode. So everything you will see there should be a charge induction. We also have a light sensor at the bottom of this tubing, so that you know when the puck actually physically arrive at the bottom of this tube. So here is our setup, we charge our puck using a piece of enamel pelt, and then you can see we start to drop it, and then start to bounce it back off the rubber ball, and then the yellow curve is the output of the light sensor telling you when the puck reached near the bottom of this tube, and then the green curve is the output of the electrode, of course after being amplified. You can see when the direction of the charged puck switches, so is the green curve's polarity, namely they induce the voltage signal, they change polarity. You can see, and also, all the screen pulses, they can form well before the puck reach the bottom of this tube, namely you can see a dip in the yellow curve, that indicates when the puck reached the bottom of this electrode. And then a semiconductor detector can be considered as a reciprocal of light-emitting diode, namely an LED. So two experimentalists, Kevin Trapp and then Christian DiCaro, in our lab, showed me that by connecting an LED with a battery, the LED can emit light. And then I said, I know that, but then they told me, wait for a moment, if you switch the polarity of the battery, so now you are reversely biasing the LED, now there's a strong electric field built in, formed within the LED, because there's a high impedance when you're reversely biasing it. Now if you shine radiation, the radiation will start to generate electron holes, and then it'll start to generate a signal, as shown by this oscilloscope. So you can simply build a semiconductor detector using an LED. But to do a photon counting, you need more than that. You need to have further post-processing using pulse shaper amplifiers in order to get a stronger signal, to reduce the dead time, to give you cleaner counting signals. And then here is our in-house setup. And then the yellow curve is the output, the analog output, of our pulse shaper. And then the blue curve is the single-channel analyzer's output. You can see it's already become a digital binary output, feeding into our digital counter. Next we discuss PCD in the context of CT imaging. In 1957, in South Africa, when Alan Cormack started to build his first benchtop CT, he used a photon counting detector, namely a Geiger-Muller counter, to count about half a million gamma-ray photons emitted from a COBO-60 source. And then in this case, I estimate their count rate and the flux rate, which is about 200 counts per second per square millimeter. And then modern CT, you can see it's about way above that flux level. How about Godfrey Hounsfield? Well, if you're looking at his first benchtop setup using a lace bed, he's actually also been using a photon counting detector. However, when Godfrey Hounsfield translated this technology to human imaging, the EMI and Godfrey no longer used this photon counting detector. Instead, they're using the sodium iodide scintillator crystal coupled with a photomolecule tube. And then after that, you have a charge integrator. Namely, this will give you an energy integrating detector. So why didn't they do this photon counting? Well, in the world of nuclear medicine, using almost the same setup, except replacing the charge integrator with the pulse-to-digital count conversion device, that's become a typical photon counting detector used in SPEC imaging or PET imaging. The reason lies in the difference in the input radiation photon flux. So we know in nuclear medicine, the flux, or the number of photon per unit time per unit area, is way lower than what we see in a clinical whole-body clinical CT. The challenge with that is the so-called pulse-pop problem. And then to demonstrate that effect of pulse-pop, now we go back to our benchtop, our prototype in-house photon counting detector CT. What we do is to hook up both a charge integrating detector and a counting detector to demonstrate you the pulse-pop problem for the photon counting detector, and also demonstrate you the benefit of doing energy integrating detection, because it's immune to charge pulse-pop. Here we run the experiments at two different MA. You can see at a higher MA, the count, namely the bottom row of the digital counter, actually drops at a higher MA because of count loss from the pulse-pop, while the top row, the output of our charge integrator, namely our electrometer, increases linearly with the tube current. That explains the success of energy integrating detector over many years in clinical CT detector. And then now, with very high flux, we know we have a pulse-pop problem. Luckily nowadays, the charge, the dead time of modern photon counting detector has been greatly reduced to about 20 nanoseconds or even lower, allowing us to catch majority of photon events with negligible amount of pulse-pop loss. And while we know in PET imaging, the pulse-pop problem is much weaker, because the time interval between two neighboring gamma-ray events is just order of magnitude longer than what we see in a CT imaging. We often heard about PET imaging, their detector has a picosecond, 100 nanosecond, a picosecond time resolution. That is not the time interval between two events, but rather the precision measurements of each pulse width. Now, I'd like to give credit to European Organization for Nuclear Research, or CERN, for the remarkable reduction of the dead time of the pixelated PCD. In 1980, 1D microstrip semiconductor PCDs was developed to address the needs of high-energy elementary particle tracking. And then later in 1988, the CERN researchers raised the question, is 2D pixelated semiconductor PCD a dream? Well, two years later, a CERN R&D proposal, RD-19, on developing 2D pixelated semiconductor PCD was submitted and approved. And then within five years, you can see they deliver, the RD-19 projects deliver 1.5D semiconductor detector with a garnium arsenide sensor. And soon after, in 1998, you can see they start to build a 2D pixelated semiconductor with very acceptable flux rate. See, in this case, we can count up to 7 times 10 to the 7th CPS per square millimeter. And then later, between 1999 and 2007, different generations of many-pix detectors was developed with truly two-dimensional array of pixels and a much faster counting rate to 8 times 10 to the 8th count per second per square millimeter. So finally, that leads to the, after all those endeavors, the PCD technology, especially its maximum flux level and stability, they reach a critical point where it enable you to translate to human CT scanners. And then you can see, this is amazingly, we can show the first demonstration of a very stable CT number across different MA level at Mayo Clinic. They show, even as clinical CT flux level, it works with acceptable stability of CT number, all those quantitative measurements. And then, of course, we know in 2021, the first FDA-approved whole body photon counting detector CT was introduced. Now, photon counting detector can be built using a phase-on or edge-on geometry. And we know it so does a scintillator EID detector. Now you have all those light sensors in a CT, you have light reflectors separating individual cells of scintillators to reduce the light spreading. And then, but on top of that, we also have this anti-scatter grid. So when we calculate the so-called geometric efficiency of each detector, we need to take into account all those anti-scatter grid. For example, when we do a phase-on geometry semiconductor PCD in CT, we also have this anti-scatter grid. So we have to consider those facts when we calculate the theoretical geometric efficiency. And usually, those sensors are bounded with another semiconductor-based readout chip for pulse processing. And then you can also do, orient the sensors of semiconductors edge-on with respect to the radiation direction, and that can help to increase the actual interaction probability. For example, an edge-on silicon photon detector has been built in a modular approach. And then those electropaths are placed on the other side of each silicon detector row, and then those modules are stacked together, and then along the z-direction, and then separated by a metallic scepter to create a bank of detector modules. And then each electrode is then wired into the counting ASICs placed below. And those module banks of detector modules can then be arranged in an arc along the axial direction for CT applications. Now we revisit the table with all those material properties. We know one of the motivations of using semiconductors to build PCDs is due to there's almost no afterglow problem, such as in the scintillator problem. And then when you have afterglow, that leads to the long tail that will increase the probability of pulse path. Another problem, as I mentioned, is the low conversion gain factor of scintillators. However, we know there's also some rare earths and some new semiconductor scintillator materials that has a really short afterglow time. And then they also have very high atomic number and high density for absorption efficiency. So actually, those are quite attractive features for practical utility of photon counting detector in addition to the fact that scintillator-based detector, they don't require high bias voltage, and there's no polarization issue in those sensor materials. But the problem with semiconductor, with scintillator detector, is that they generate a lower signal per radiation events. But however, there's some advantages, as I mentioned, about using scintillators to build a photon counting detector. In term of the conversion gain, the challenge of low conversion gain, we can actually using, for example, silicon falling multipart tube to amplify those signals, the light signals. And then in some of the cases, the conversion gain can be 10 to the fifth. So that mitigate the limitation of scintillator and give promise of building the scintillator-based photon counting detector, such as by using the lanthanum bromide or the LYSO. So at TU in the Netherlands, there's a very nice study showing the promise of building those scintillator detector, photon counting detectors, where the dead time can be around 60 nanoseconds. And then together with a proper pulse shaping, and then you can get actually very promising counting rate and also dead time performance using scintillators. It's just something we can consider in addition to the established advantages of using semiconductor detectors. So in summary, the essence of photon counting detector is to count individual events and to harvest more information, including temporal, energy, spatial information. And then the basic building blocks of a semiconductor detector is a diode, and also we need a semiconductor readout circuit. And the reduction in the dead time over years leads to the translation of this technology into clinical whole body CT. Semiconductor-based PCD, as I mentioned, are very attractive and have many desirable features, physical properties for CT imaging. However, I think the potential benefit of scintillator-based photon counting detector CT should not be overlooked. So with that, I'd like to conclude my talk. OK, so I'm going to talk a little bit about the clinical applications. As we have heard from Ke, there are some challenges with photon counting using CT. But in the last 15 years, we have seen tremendous advance. And that result in your lab systems built preclinical scanners. We have seen phantoms, animals, and patient studies. And then we all know the news in 2021. FDA approved the first commercial photon counting detector, CT, the near-term alpha from Siemens. And that enabled us to use this in clinical routine practice so that we can explore the benefit and how this can help with patient care. Then in 2022, Neurologic has their FDA clearance for a mobile dedicated head CT using photon counting detector technology. So those are the FDA approved devices. I have also seen many vendors develop the prototype systems that can do human imaging with the clinical dose level and dose rate. So we have seen a lot of studies from these prototype systems and also more coming out. So this is a really exciting time as a CT researcher because the reason we're so excited about photon counting is the tremendous benefits that this technology can bring us. We have heard from Dr. Li that the reduced electronic noise, the increase in R, the increased spatial resolution, low radiation dose, and the multi-energy capability. So many of the clinical benefits can get from this new technology. So I want to spend a lot of time discussing each of these benefits and what the potential clinical applications. I want to share with you some of the examples that we have produced in our practice and also a lot of great results from the literature. So first of all, the reduced electronic noise, this can be really beneficial for low dose exam or for large patient or for large body part. For example, the shoulder area has always been challenging given the high attenuation along the lateral direction. This results in pretty strong streak artifacts sometimes we see in the energy integrating detector. With photon counting, with the reduced electronic noise, we can see better in this area. So for example, this is a case that we can see the improved visualization of the bronchial plexus, which is nerve. And as we go lower radiation dose, the CT number actually stability was reduced with EID detector. This study from NIH group, they demonstrate very nicely as dose go lower and lower, you see the CT number changes. This could be a problem if we want to do quantitative imaging. But with the photon counting, that this CT number can be very stable even we go lower radiation dose. As we heard from Dr. Li about the tomatoes, I like those small potatoes. Tomatoes, they really taste great. And the energy integrating really didn't wait on those sweet spot of the low energy photons. So it's not very efficient in sense of creating high contrast noise ratio. Photon counting give the equal weighting, so give the low energy photons a better chance to show a better images increases in R. As demonstrated in this early study we did with different phantom sizes, we even fixed the radiation dose. Between the EID and the PCD, you can see the improvement of the contrast noise ratio. And this can also be visualized in patient exam. This is the same patient, scan the same day. EID, PCD, you can appreciate the enhancement of the contrast signal due to the better weighting scheme of photon counting. We can also use this to reduce image artifacts. So for example, this is a CT scan in the post-VASA area. This is typical beam hardening artifacts that probably see very frequently in the daily practice. This is because of high attenuation in the bony structures. You do see that on the photon counting detector CT2 if we just use every photon. However, one benefit of photon counting is you can set up the threshold. So if I only look at the high energy threshold images, you'll see the removal of the beam hardening artifacts. And this can also be applied to, say, for example, metal structures. Like this is a smart hardware. This is a low energy threshold. This is high energy threshold. You can see the artifact is less obvious on the high energy threshold. But when I look at these images, I was like, it's better, but not great. So can we do anything better than that so we can further improve artifacts? So we did some study. We say, what if we apply an additional filter to the beam and make the spectral even harder? And as you see here, we did some simulations. We can see the improvement of the effective energy towards more high end after we added a thin filter. On top of that, I also calculated the percentage of photons in the high energy threshold. And that's substantially improved. This is important because when we use the high energy threshold, we essentially throw away some of the low energy threshold photons. That's a penalty of that. But we took that penalty with the thought that we can reduce the metal artifacts. By improving the percentage of high energy photons, we actually reduced the penalty we took. So as you can see from here, after we added a thin filter, the metal artifacts are substantially reduced. On top of that, we also noticed the noise is lower than the one without a thin filter. So we then went on, performed some patient studies for the spine hardware. These are the EIDCT images. And these are the PCDCT images scanned at the same day. We can appreciate the metal artifact reduction from the hardware, which enable us to look at the soft tissues, look at structures around the spine and around the hardware. We performed a reader study with three radiologists and looked at a few critical structures. And for all the structures, the radiologists' readers gave higher score, better image quality of the photon counting detector. One of the major benefits so far we have paid out fantastically of photon counting is the ultra-high spatial resolution. CT is a high-resolution modality. But we can benefit from further improvement of high resolution in multiple clinical areas. MSK, lung, vascular, imaging. To make high resolution, we have to make the detector smaller. That's one must factor. With a conventional energy integrating detector, to prevent the light spreading, we have to use the SEPTA. And as the detector goes smaller and smaller, the percentage occupied by the SEPTA, which is the data space, will increase and not reduce the field factor and geometric efficiency. With the direct conversion, we don't have that problem in the photon counting detectors. And that can give us a dose-efficient, ultra-high resolution scan mode. Again, as mentioned, this has been a great application in many clinical areas. So what's the benefit? Obviously, better resolution enable us to see things better. So for example, this lung imaging, the lung nodules, the airways, we can see them in the current EIDCT. But with the high resolution, we can see them better. And I hope you can appreciate from these images comparison. Then in addition to that interstitial lung disease, where we see certain radiological features, with the improved resolution, those features will stand out better. So this is a study led by my colleague. And in the study, this is a case with traction bronchiectasis. You can demonstrate that on the EIDCT images and the PCD images on the right, the PCD revealed better visualization of the reticulation and also more clear traction bronchiectasis compared that with the EIDCT. In MSK applications, clearly higher resolution enable us to see better with fractures, like the fine line hairy fractures. But in addition to that, we can also use it to evaluation the healing. So for example, this study, my colleague, Dr. Bafar, this patient with the healing fracture of the glenoid, the early catalyst formation at the base of the fracture can be better visualized with the photon counting. So then he will be very confident this patient is healing. We don't need a further intervention for this patient. And this is a slide made by my colleague, Dr. Dean, who is a neuroradiologist. And so in the head and neck CTA, there are a lot of small vessels. And this, for example, if ophthalmic artery is about 1.3, 1.5 millimeter diameter, this arrow help us to point to the artery what we want to see. So this is EID image. I probably will not see it without the arrow pointing there. But with the PCDs, the same location, I don't think anybody will miss that. It's just so obvious. And then he actually went down to even further smaller artery. This is the smallest clinically important artery in the intracranial area. And here, this is the EID. He pointed the arrow here, but I don't see nothing there. Actually, he told me he couldn't see nothing there. The artery is supposed to be there. But with the EID limited resolution, you don't see that. And here's the PCD. I think I can see it too. So this not only enable us to see things better, they actually enable us to see things we don't see them before. So that's one example previously. And this is another example. These are the blood vessels supply to our eyeballs. We don't see them in the conventional EID and the radiologist don't comment on them. They don't even mention them. But now with the high resolution, they become visible. And I think it still need further investigations what to do with those arteries. So that's something you need to look at in the future. Coronary calcification. Co-cal is a strong indicator of the future major cardiac event. So we know that's a biomarker well established. There's also a theory called power of zero. So patients with zero calcification have much lower risk than those patients even with just a little bit of calcification. And however, the EID with limited resolution, if we have teeny tiny calcifications, we might miss them from the EID CT. As you can see these images. This is using an eggshell phantom. This is very nice work from our colleague, Dr. Chang. And using PCD on the other hand, we can detect all the way down to the 0.4 millimeter calcification. We probably can detect even smaller, but the 0.4 millimeter is the smallest that we can break the eggshell reliably. So that enable us to see the things we might miss on the EID. This can also help us work with the challenging situations. For example, the stent in vascular imaging. This could be very difficult to look inside the stent because stent is metal and has blooming artifacts. You can see the lumen here, it's very narrow. It's hard to evaluate particularly the area close to the stent. But same stent scan with high resolution and you can see them very well inside the stent. And so that can help us working with the challenging situations. This also been seen in vivo study to their multiple vivo study published. This one by Dr. Sam Muhammad, they actually use a prototype system from Philips scanner. And you can see the stent with high resolution. You can see the individual strata very well and you can also evaluate the lumen better. So those are visual effect. This also help us quantitatively. So if we want to quantitative imaging, we want to quantify any measurement. This is also helpful. So for example, coronary calcification, we assign a score to that. We want to estimate the volume. With the blooming artifacts, the current CT scanner with the EID technology usually overestimate the amount of calcification the patient have. As you can see here, it is much large. With the PCT, you see those calcium becomes smaller, but we were wondering which one is true. So we actually scanned them with macro CT for the specimen. And compared to the macro CT, the EID result in 24% overestimation. And this was reduced down to 9% with PCT. What about the morphology and the texture? This can also help us with the high resolution. So for example, lung nodule evaluation is a very important task. With the high resolution, we can provide a more accurate volume estimation. This study we did with a prototype system a few years back. We tried to evaluate the volume of the different type of lung nodules. And the high resolution can provide more accurate estimation particularly when we look at the small lung nodules and especially with irregular shape. Those can be very large error with the EID technology, but PCT improves substantially. In addition to that, we try to look at morphology because the irregular shape speculated lung nodule usually associated with malignancy. To differentiate them from the regular shape is important. We perform a ROC analysis and demonstrate higher error under the curve using the high resolution for the counting. This can also potentially change patient management. This is work by our graduate student, Emily. And in this study, we tried to characterize the degree of stenosis in coronary CTA. This is critical because that guide the clinical decision how to manage the patient. High degree stenosis, we might need to send the patient to a cath lab, might need to stand, but low degree can be managed medically. And clinically, there's a so-called cataract scores you report to the patient based on their stenosis. And as you can see from this particular case at the top of the EID, the bottom is PCD, the stenosis is much less on the PCD. This changed the category which the patient fall into. And this is a group of patients that we did the study, and you see the shift of the histogram distribution with the high resolution. The degree of stenosis shifted to lower end. So clinically, today with the EID, this patient might be overestimated. And what about radiation dose? The reduced electron noise help with radiation dose. The more efficient photon weighting help with the radiation dose. Another benefit with the smaller detector is the high resolution sampling allows stronger filter during the reconstruction. That can also help reduce radiation dose at a given spatial resolution. As shown in this case, we demonstrated a 47% lower noise at matched radiation dose. If we flip the coin, this give us a high potential for dose reduction. In addition to that, we can also play the game of optimize the beam spectrum, like use a external filter that can help us remove the very low energy photons that can not penetrate through the patient, and also the high energy photons don't provide great contrast. So this kind of shape the spectrum to a sweet spot in the middle. And combining all these together can generate us very significant dose reduction, like in our study shown in these applications. And also, this is for a temporal bone. So this is the teeny tiny bony structures here. And this particular patient has a prosthesis as pointed by the arrow. Here you can see, we barely can make it out on the EID, and it seems like it's tilted and screwed, but the reality is it's in good position. Nothing wrong with it. It's just we didn't have the right solution. And on top of that, we see these things with lower radiation dose. So we see more with less. That's the benefit of the photon counting detector CT in clinic practice. As we heard, the photon counting intrinsically has multi-energy capability. So this provide us a simultaneous high-resolution multi-energy capability, which is not feasible with current EID technology. You have to choose, go high-res, or you go multi-energy. But with photon counting, we don't need to choose that any longer. We have both benefit in a single scan. So this could be very helpful in clinical practice. My colleague, Dr. Shamblit, gave a talk earlier. It's week for the cardiac application. So this is same scan of a patient, but different reconstruction. I want to highlight the benefit. This one is ultra-high-resolution, more or less like a single-energy image. So they can help us see the small coronary arteries and see the calcifications with high-resolution, as we talked about. In addition to that, we can generate a slew of multi-energy images. For example, different energy of the VMI. The lower KEV VMI give us a higher contrast. The high KEV VMI give us lower artifact. On top of that, we can perform a tear decomposition. We can remove the calcium and visualize the lumen. Off-leave the crown, we remove the iodine and leave the calcium there, so we can do a calcium quantification. You have contrast, it has the skin. Outside the coronary, we can also look at the myocardium for the infarct, for the ECV. So many things we can do with this capability. MSK application, look at a fracture with high-resolution, look at a bone edema with a multi-energy. On the dual-source photon county detector CT we have, this also has a simultaneous high-pitch mode with multi-energy capability. So this can, in addition to the benefit that we just mentioned, it also provide us a super-fast scan. And this can be used to see cardiac lung imaging using this new mode. This also have some impact on the cardiac motion. So previously, dual-source EID system have a benefit of improved temperature resolution in the single-energy mode. Once it go to dual-energy mode, the two tubes need to operate at different KV, the temperature resolution drop back to single-source. So we lost the benefit there. But with photon counting, the multi-energy coming from the detector, so that two tube can still operate at same KV, have the benefit of temperature resolution. So, for example, in these images, you can appreciate that these are item map. On the EID, lower temperature resolution, RCA got strongly distorted. But look at the PCD with the 66 millisecond temperature resolution. All the RCA's are well-characterized. One benefit of the photon counting is the K-edge imaging. Because we can select the threshold, we can set it up right around the K-edge of certain materials and enable us to do K-edge imaging. This is a study from Darkroma back in 2010, and they did golden nanoparticles in mouse models. And using the photon counting prototype, they can accurately detect and quantify the golden nanoparticles. This also allow us to potentially imaging multiple contrasts at the same time, especially if one or more of the contrasts have K-edge, as have been demonstrated in animal studies in multiple publications. Photon counting, it's not perfect. There is challenges, too. So, for example, the spectral separation. When we first got working on photon counting, we were hoping it can give us multi-energy bins, clear cut boundaries between multi-energy bin, give us perfect spectral separation. That's our dream, but we quickly came to reality because of many factors like charge sharing, K-scape, et cetera. The spectrum we really see is on the right, strong overlap. So there are multiple techniques, multiple methods try to improve the spectral separation. One thing we investigated is try to use the dual source technology, operate a different KV combined with the photon counting. And this is some spectral simulation. On the right, you can see this is much better spectral separation compared to the left situation. And we also calculated the effective energy and how much different between the different energy bins. Then we move on to some of the experiments. So we scanned with some phantoms. We look at the Iden water map. So that's the character of the multi-energy performance. We started with a single source PCD with two energy bins. Then we move on to using dual source and two different KV with just one energy bin from each of them. So this is still dual energy compared to dual energy. We already see some improvement. And further on, we can use two energy bins from each of the tube and then put together, you see further improvement. On top of that, apply the tin filter to the high energy bin, and that give us the best results. Putting all this together, we actually see 40% reduction of lumen square error by using this technology together. This have significant impact when we try to do multi-contrast imaging because the more contrast we want to differentiate, the higher magnification of the image noise. In addition to that, we want to extract the most information out from the high-resolution multi-energy data. One challenge of the high-resolution is the image noise. On top of that, the multi-energy analysis usually increase the noise. So how can we take full advantage of the system that we need some software to help us? Otherwise, the noise will bury the signal no matter how high-resolution you got, we don't see them. So what we have developed a whole bunch of denoising techniques. There are a lot of denoising techniques in the literature. Just want to show you some example, use some prior information to do iterative denoising technique. And then you can see the images on the right. It's much better than the middle. So now really the high-resolution benefit pop up. So the high-resolution low noise have to go hand by hand. This is some AI-based denoising dedicated for ultra-high-resolution photo counting. And we also did some comparison to the iterative reconstruction. So this one has the benefit of reducing the noise in the ultra-high-resolution while maintaining the structure integrity. Going back and forth, you can appreciate the difference of the images. In addition to that, the multi-energy analysis, particular material decomposition, we needed to extract the information embedded in the multi-energy capability. So this is one algorithm called A-Gate using some AI technology. And we look at the routine images. This is a multi-myeloma case. So the bone signal will bury like the lesion behind the bone. The big one, we can see them, no problem. The other one behind the bone, we don't see them. With the multi-energy, we can generate virtual non-calcium images, remove the calcium, so we can see some of the healing. But image quality is not great, artifacts, noise. With A-Gate, however, we can see them much better image quality. But not just that, look at the small lesions that we didn't see them before, no matter single energy or the virtual non-calcium from multi-energy. So we have to do a good job, extract the information out. Can also apply to VMI, extract the information, reduce artifacts and noise. We can also potentially push the envelope. Some of the imaging not considered today with routine CT, like breast imaging. CT is not generally considered when I'm talking about diagnostic CT, not dedicated breast CT. But with high resolution capability for counting, in combination with the denoising, we actually can see those small calcifications that we don't see today. And this not just give us better images, this could be life-changing. So my colleague did quite a lot of work on CSF leak. This patient usually have deliberating headaches that takes a long time, sometimes could be years for definite diagnosis. But with high resolution multi-energy, he can pin down the leak location so that they can do ambulation at the right place. This totally changed the life of the patient. So with that, I want to quickly summarize. Photon counting offers multiple benefits over the conventional technology. There's a lot of clinical benefits and we have seen commercial available systems and multiple prototypes available. So I think we are truly in a new era of CT. With that, I would thank my colleagues and thank you for your attention.

Photon Counting Detector

CT Imaging

Image Quality

Spatial Resolution

Medical Imaging

FDA Approved

Lower Radiation

K-edge Imaging

Multi-contrast Imaging