The 2nd International Symposium on Integrated PET-MRI

Session II: Basic Technology

SiPM PET/MRI

Hidehiro Iida —I’d like to start. I’m Dr. Iida from the National Cerebral and Cardiovascular

Center nearby Osaka University, and I will be chairing this session. The first speaker is Prof. Jae Sung Lee from Seoul University. He will be talking about Silicon-PM PET/MRI. If you please. Jae Sung Lee

Thank you, Dr. Iida, and good afternoon, everyone. As introduced, my name is Jae Sung Lee, and I am from the Department of Nuclear Medicine at Seoul National University, and actually, it is really my honor to be invited here again to give my presentation on the SiPM PET/MRI.

So let me start with the outline of my presentation. In my talk, I will briefly tell you about the photosensors used in the PET/MRI application. They include the photomultiplier tube (PMT), the avalanche photodiode (APD), and the silicon photomultiplier (SiPM) or Geiger-mode APD (G-APD). Especially, I’d like to focus on the introduction of the G-APD-based PET/MRI developments that happened in my group at Seoul National University.

As you know, there are many advantages to PET/MRI. They may include: a reduced radiation dose, the better soft-tissue contrast in MRI than X-ray CT. And you know that almost unlimited combinations of functional and molecular information are possible with the PET/MRI, and it is also possible to correct the motion in PET images using the MR data.

You know that the most commonly-used photosensor in current PET is the PMT. This is because PMTs have a very high gain of about 106, and you know that it is quite stable against changes of environment effects like temperature or humidity. But you know that it also has a weakness in the PET/MRI application because it has a bulky size and is very sensitive to magnetic fields, as is shown in these examples.

Therefore, the use of the semiconductor photosensor in the PET/MRI application is a common way, so one of the most investigated photosensors in the PET/MRI application would be the APD. This is because the APD is insensitive to the

magnetic field and has a very small size. There are also weaknesses in the avalanche photodiode. It has a very low gain compared to the photomultiplier tube. Another critical thing is that the avalanche photodiode does not have good timing resolution. The timing resolution of the avalanche photodiode would be not sufficient for time of flight measurements.

This is why we have much interest in another type of photosensor, the G-APD. The G-APD is also called the SiPM – the Silicon-PM – or the SSPM or the MPPC. This sensor has a higher gain that is almost equivalent to the PMT, and the fast pulse rise timing, consequently it has good timing resolution for the time of flight (TOF) measurement.

Therefore, though we have worked for several years to develop the G-APD-based MR-compatible PET detectors for PET/MRI application. Last year we finished the development of a small animal PET prototype using the G-APD sensor, and recently we are also working on the MR-compatible G-APD PET system for simultaneous PET/MRI imaging. So I’d like to tell you about the story of these two developments.

If you look at this slide, actually this is the component of the block detector of the first G-APD prototype. Actually, when we started this project, only the single-cell-type G-APD was available, so we used this single-chip type SSPM provided by Photonique SA, to make this 2x6 G-APD array. This array was made by soldering each single chip SSPM onto the PC board like this. And we copied this 2x6 array, we did a 4x13 LGSO crystal array with a size of 1.5x1.5x7mm. And these two components were placed inside of this package here, and each of the SSPMs were connected to the preamplifier board like this.

This is the position encoding board, consisting of resistive charge division to reduce the number of channels from 12 to 4 channels for simple data return. And this right figure shows our first prototype, a scaled ????? prototype scanner that has eight detector blocks and it has a 6cm diameter.

This is the flood map of a typical detector block, and it yielded about 25% energy resolution for a single crystal. And this is an image of the point sources, and when we measured the spatial resolution, the spatial resolution at the center was about 1.0mm at the center of field of view (FOV) when using the MLEM construction.

And I remember that I have shown you these animal imaging studies just last year. Actually these are the images that we have acquired using our G-APD PET, and this is the X-ray CT acquired immediately after the PET scan by moving the animal body into the X-ray CT machine without any movement of the mouse. This is the fusion

image of this X-ray CT and G-APD PET. This upper one is the fluorinate-FDG PET in the mouse with colon cancer, xenografted here. And in this example we have used the Galium-RDG radio tracer for the angiogenesis imaging, and you can find a very high accumulation of this Galium-RDG in the angiogenetic glioma cell line here.

This is also a very interesting image, I think. This upper one is a PET image acquired using our system, and this lower one is the PET image acquired using conventional PMT-based PET of GE Healthcare. So if you compare these two images, I think that you can appreciate that these two images yielded almost equivalent image quality and spatial resolution.

This is another example that demonstrates such a fine spatial resolution of our first system, because this was the myocardial FDG-PET study in the mouse heart, a very small heart. But this also demonstrates some physical limitation of our first system, because if you look at this image, it does not contain all the left ventricular regions because of the limited axial FOV.

So this is why we have decided to develop our second system. In our second system we used the same LGSO crystal but a different number of crystals per every block. In the second system the number of crystals was increased to 360, and instead of the single-chip SSPM, in this second-generation machine we used a multi-channel 4x4 MPPC that has been shown in the previous presentations as well. So because we have used this multi-channel MPPC, the ring diameter and the axial FOV of the system could be extended to 9cm and 3.2cm respectively.

This slide shows the block detectors. To make the block detector, we combined four 4x4 MPPC arrays that yielded a total of 8x8 signal output and we coupled this 8x8 MPPC array with a 20x18 LGSO crystal. And these two components were placed here in the detector module and we directly connected these MPPCs to the charge division socket in the position encoding board to reduce the number of channels as well. And the amplifiers for these four channels and the temperature sensor were placed in this area. And this is the shielding box to minimize the interference between the PET detectors and the MRI machine.

So let me show you our second prototype scanner. That also consisted of eight detector modules and it had a 9cm diameter, as I told you. And for the data acquisition we used our dedicated FPGA-based acquisition system for the experiment.

This slide shows how we designed the resistive charge division network. Actually, this is one of the essential parts in our PET detector module. That is because the MPPC has a very low output impedance compared to the PMT. So because of this low output impedance, we cannot actually ignore the loading impedance by the resistive

charge division network. So we cannot use the conventional charge division network with the same resistor values for each column. So for the impedance matching from each output of the MPPC scintillator, actually we optimized these resistive values to avoid the attenuation of the MPPC output signal.

This is the SPICE simulation result of this optimized charge division circuit, and I think that you can agree there that we could obtain a good linearity in the determination of the interaction position in the scintillator crystal using this charge division network.

We also implemented a continuous temperature monitoring program. The gain of the MPPC is quite dependent on the temperature changing [=sensitive to temperature changes], so that is why we need a temperature sensor for each of the detector blocks like this. And using this program, we measured and recorded the temperature data every one second like this for the retrospective correction of the gain of the MPPC.

This slide shows such a temperature-dependent gain change that was measured using this data. The gain change was about 3.7%/°C when it was measured at room temperature like this. Anyway, it showed very linear changes of the output gain, so using this information I think we can easily correct such a temperature-dependent gain change.

We also measured some intrinsic physical properties of each block. The left figure shows the global energy spectrum of a typical block after the peak alignment. It yielded about 14% energy resolution; that has been much improved from our first prototype version. The right figure is the intrinsic resolution profile of a detector pair; it yielded 1.45mm intrinsic spatial resolution.

We also performed this MR compatibility test and simultaneous PET/MR imaging using the 3T Clinical MRI of Siemens Healthcare, and to obtain the MR image we used a 4cm loop receiver coil. Actually, this coil was placed inside of the PET detector block like this using a custom-made coil holder and animal bed. The use this coil holder and animal pad was for the reproducible positioning of this RF coil relative to the detector blocks and to prevent the motion artifacts due to the vibration of the coils while we acquired MRI data.

This is the schematics of data flow. The PET signals and temperature, the signal from each detector block was transported to a DAQ board that was located outside of the MRI room, and the digitalized signal from the DAQ board was moved to the master board for coincidence processing.

Okay, let me show you some results of MR comparability tests. This left one is

the signal measured by the oscilloscope without any application of the RF signals, and this right one is the same PET signals, but measured inside of the 3T MRI with the T2 turbo spin echo sequencer with RF shielding. So I think you can agree that basically, there is no change in these two signals. It means the RF shielding that we have used was quite successful.

This is another example that showed MR comparability of our block detector. This is the global energy spectrum of a block detector without peak alignment. The blue one is without RF pulse, and the red one is with RF pulse. Actually, they are the same.

Look at these two images. This is the crystal map measured without RF pulse, and this was measured with RF pulse, so this is another [=further] evidence of the compatibility of our detector module.

We acquired some preliminary simultaneous PET/MR images using this system. Actually, these are uniform phantoms and one capillary tube was located among the cylindrical phantoms here. And the capillary tube was filled with fluorinating solution. So this T2 MRI and the G-APD PET were acquired simultaneously.

And in this example, we filled one of the cylindrical phantoms with the same fluorinating solution.

In this example, we used a cucumber to obtain this nice MR image, and we inserted three capillary tubes here. There was also the fluorinating activity inside the capillary tubes. So these images show the feasibility of our system for simultaneous PET/MR imaging, I believe.

We also performed this experiment to look at the influence of a PET insert on the MRI. This left image is T2 MR data acquired without the PET insert inside the FOV, and this is the MR image with the PET insert and the power was on for this PET insert. Although you know that in this region the SNR [=signal-to-noise ratio] is quite similar, but if you look carefully here, it seems that the SNR with the PET insert is slightly lower than MR images without the PET insert. So we are now trying for the improvement of the SNR of the MRI data.

We are trying several methods, but I’d like to just mention this one. One of the methods that we are trying is to use a short optical fiber between the scintillator crystal and the MPPC array to make the distance from the coil to the RF shielding material as far as possible. My colleague, Prof. Hong, will tell you the details of this methodology at tomorrow morning’s session.

Another approach that we are now doing is to implement a depth of interaction (DOI) detector using the same detector module and electronics. This left one is the relative offset method using the same crystal but different crystal numbers. So I think

we can agree that our detector was quite successful in this method and also in the pulse shape discrimination analysis using the LGSO crystals with a different decay time.

This is my summary. I think these results indicate that it is possible to develop a PET camera using a promising G-APD or SiPM, which yielded reasonable PET performances in phantom and animal studies. And simultaneous PET/MR imaging was possible with multi-channel G-APD, especially multi-channel MPPC – of Hamamatsu – -based PET insert with no RF interferences on PET signals, but slight degradations in MR images. And finally, I think we need further work to improve our system, including various methods to improve the SNR in the MR images, incorporation of depth of interaction information, and finally, incorporation of temperature information that we measured to compensate for temperature-dependent gain fluctuation of the G-APD.

Let me appreciate [=express my appreciation for the help of] my collaborators, Prof. Seong Jong Hong and Prof. In Chan Song, and these are my post-docs and students. Thank you for your attention. Hidehiro Iida

Thank you very much, Jae Sung Lee. It’s a wonderful presentation, I think. I’d like to invite questions and comments from the audience, please. Any questions? Dr. Yamamoto, please. Seiichi Yamamoto

I’d like to ask you about signal detection by the PET scanner from the RF signal, but you don’t observe any RF signal. But if you look at the lower level of the PET detector signal, don’t you observe some fluctuation due to the RF or gradient signal? No signal… even if baseline… Jae Sung Lee

Actually, there was no signal although we lowered the trigger threshold in the oscilloscope, so… Seiichi Yamamoto

If you don’t use shielding, what do you think? Jae Sung Lee

Let me show you. Actually, this was an expected question. This is the signal that we measured without RF shielding… Actually, there were multiple channels, but if

you just look at this yellow one here. This is the PET signal and this signal with a very high frequency would be the RF interference. So even if we do not shield it, the trigger level is too high, so we cannot use that data. Seiichi Yamamoto

What kind of shielding do you use? Jae Sung Lee

We just use copper shielding. Seiichi Yamamoto

What is the thickness of the copper shielding? Jae Sung Lee

Actually, we are changing the thickness of the copper, but it is between 10-20µm. Seiichi Yamamoto

Thank you. Hidehiro Iida

Any other questions? Simon, please. Simon R. Cherry

Can I ask about your charge division circuit? Because that was very interesting. Am I right in understanding that all you did was you changed the resistor values through the SPICE simulation? Can you tell us a bit about what they changed to? Did all the resistor values basically go down and then you also adjusted the relative values, or…? Jae Sung Lee

Actually, this technology is not a new one, I think, but I think we can find a similar one in some literature. Anyway, you know that resistor values in the center of the resistor chain should be higher than the resistor values of the periphery of the network. But we need to, how do you say… I’m not sure that there is some systematic way, but we just adjust manually to obtain the linear result.

Simon R. Cherry

Right, thank you. Hidehiro Iida

I wasn’t sure that the amplifier was based on the charge-sensitive pre-amplifier. You may be just detecting the very, very low signal with a big magnification factor. Jae Sung Lee

Yes, yes. Okay, let me show you this. One second. Actually, if you look at this network, we do not use a pre-amplifier for each cell

of the MPPC. The MPPCs located here were directly connected to the position encoding board without use of a pre-amplifier. Actually, that was fine because the MPPC signal is quite high and the noise level of the signal from each MPPC was very clean. So we did not feel that we needed any pre-amplifier before the position encoding board because this network is the weighted summing circuit, so actually using the position encoding board, the signal has a better SNR. That is why we just used the amplifier after the position encoding board. So I think that is a good way to make our detector as compact as possible, without use of many pre-amplifier boards or AC board or something like that. Hidehiro Iida

Can I see the next one again? Jae Sung Lee

This is an equivalent [=the same] circuit for each MPPC cell, and it was directly connected to each node of the resistive chain like this. And we obtained four signals – A, B, C, D – and this was transferred to the amplifier. So this amplifier sent the signal to the outside. Hidehiro Iida

Maybe my knowledge was very old. Normally, the charge-sensitive pre-amplifier has feedback by the capacitor… Jae Sung Lee

Actually, this is a kind of a differential amplifier, so it’s a little bit different.

Hidehiro Iida

Okay. Yes, Watanabe-san, please. Mitsuo Watanabe

How can you get the energy signal – summing the ABCD signal? Jae Sung Lee

We obtained the summed signal outside the MRI room after obtaining… actually I should check whether we summed the signal inside the detector module or outside. Let me check… Let me see… Yeah, we calculated the summed signal inside the detector module by just adding the ABCD values. Mitsuo Watanabe

So did you estimate the timing resolution? Jae Sung Lee

We have not yet. But I don’t think that the timing resolution would not be… I’m not sure, but it would not be sufficient for time of flight measurement because we just used this charge divergence, okay? But the proposal was not to develop a time of flight PET detector, just a small animal one, so we did not need that kind of information.

Mitsuo Watanabe

Thank you. Hidehiro Iida

You said the image quality is worse with the PET insert. Is that because of the absorption of the RF transmission? Jae Sung Lee

Yes. As I showed you, we used this receiver coil, so the RF pulse came from the main body coil, so that is why some of the RF signal was blocked by our shielding material. That is why we need to work more on the optimization of the shielding material. Hidehiro Iida

You showed the energy spectrum; it was very broad. Is that…? Jae Sung Lee

Okay, so you mean this one? We did not align the peak of the energy spectrum from each scintillator crystal. Each scintillator crystal has a different position of the energy spectrum peak, so if we do not align such peaks, the energy spectrum would be quite broad like this, but after such a peak alignment we can get this kind of energy spectrum, yes. Hidehiro Iida

Any other questions from the audience? Okay, if not… Thank you very much indeed. Our next talk is by Prof. Seiichi Yamamoto. He will be talking about SiPM PET and iPET/MRI II. The “i” stands for “integrated”, he explained to me. Dr. Yamamoto, please.

SiPM PET & iPET/MRI II Seiichi Yamamoto

Thank you, Dr. Iida, and thank you for coming to this conference. [interval 0:33:00 – 0:37:45] Okay, I’m sorry to keep you waiting so long. Today I will talk about iPET/MRI

II and SiPM PET development. This a group collaboration work, with me and Neomax Engineering and Osaka University Hospital, Osaka University.

First I’d like to talk about iPET/MRI II development. This is a second generation ultra high resolution optical fiber-based PET/MRI system.

The objective: Development of integrated PET/MRI systems has been a new target for the researchers in this field. Most PET/MRI projects will use semiconductor-based photodetectors such as SiPM because they are less sensitive to static magnetic fields. However, they will have interferences from MRI and also have high temperature dependency. Optical fiber-based integrated PET/MRI systems have probably no interference from MRI and no temperature dependency. Optical fiber-based PET/MR systems were introduced by Dr. Simon Cherry more than ten years ago. We developed new high-resolution optical fiber-based PET detectors and used them for our new integrated PET/MRI system.

The optical fiber PET system detector concept is: selecting the highest numerical aperture and smaller diameter optical fiber for less light loss and smaller bending diameter. Most parts of the optical fiber are flexible for lowering cost and minimizing thickness. Dual-layer LGSOs with different decay times were used for depth of interaction (DOI) detection. This is a schematic drawing of the detector. Dual-layer LGSOs with different decay times were used for the delay detector block, the scintillator block. And the scintillator pattern goes up and changes direction and is fed to the position-sensitive PM tube. And most of the parts of the optical fiber are flexible so you can change the form of the optical fiber, and it can reduce the thickness of the optical fiber, and it is very good for reducing the diameter of the PET detector.

This shows the minimum bending radius of the optical fiber bundle. We selected an optical fiber: Kuraray double-clad type, clear-PCM type, with a numerical aperture of 7.2. This is not a very new product. This is an old product, but still this optical fiber has the highest numerical aperture.

This formula shows the smallest bending radius. R larger, with the optical fiber radius and the difference of the clad and core, NC and ND. This is a graph showing this relationship. If we use a 0.25mm radius – it means a 0.5mm diameter optical fiber – the

minimum bending radius is around 5mm. So I used 0.5mm optical fiber and made a bundle using this optical fiber. This is

the optical fiber bundle used for the PET scanner. This has two inputs and one output. In between, this is flexible. The length is 72cm. This is the input part. The bending radius is around 10mm, so there is no light loss at this bending area. Input diameter is 11x22mm. This is the output part. The dimension is 22x22mm. If I place this figure in output, you can see the divided figure in input because this is an image guide. If you place it here, you can see the image here.

This shows the optical fiber-based detector block. We used an LGSO block with different decay times; two types of LGSOs. One had a decay time of 31ns with a cerium concentration of 0.025mol%. The size is 0.8x1.3mm and the thickness is 5mm. This is positioned on the lower side of the block detector. The other had a decay time of 46ns with 0.75mol% cerium concentration. The thickness is 6mm. This is positioned in the upper part of the block. And these scintillators are stacked to form a DOI detector and combined into an 11x13 matrix to form a block. And the pulse shape analysis was used to distinguished these two layers. And the block size is 10x18.3mm.

This block was optically coupled to the input part of the fiber bundle. This is the output part. It was coupled to the position-sensitive PM tube. The position-sensitive PM tube used was a 1-inch square crosswire and high-QE type.

The light loss due to the fiber was 72-80% depending on the position. However, we can obtain a very good separation on a two-dimensional map of the two LGSO blocks. This is the axial direction. This is the transaxial direction. In axial direction, most of the scintillators gave a clear result. In the trans-axial direction, not all the scintillators separated, but we can distinguish most of the scintillators.

This is the energy spectrum. We have two peaks, and this is the lower layer of the scintillator, and this is the upper side of the scintillator. This is the pulse shape spectrum; we have two peaks. This is the fast LGSO and this is the slow LGSO. We can distinguish the two because there is a valley between them.

Eight block detectors – it means 16 LGSO blocks arranged in a ring that has a diameter of 56mm. This is a detector ring. LGSOs are arranged here and position-sensitive PM tubes are arranged here, and in between, this is the flexible optical fiber. As you can see, this is really flexible here. This is the input part. These are the LGSO blocks – very difficult to arrange in the same separation, but I’m looking at it from this transparent part, and I arranged it as precisely as possible. And scintillation light is transferred using this fiber to the position-sensitive PM tube. This is the outside part of the detector ring.

This detector ring is contained in a black plastic case. The signal from the position-sensitive PM tubes are fed to the weight-summing board and the A/D [=analog-to-digital] converters and fed to the FPGA-based data acquisition system. This is the whole system. The outer detector diameter is 14cm, the length is 85cm, and the scintillators are arranged here, and the position-sensitive PM tubes are arranged here. Signals are fed to the weight-summing board and fed to the data acquisition system.

The performance of the PET system… The radial resolution was 1.2mm at the gradient at the center of the field of view (FOV), and you can see some improvement of the resolution using the DOI information. This is the result of the filter back projection. This is the sensitivity profile. The sensitivity was 1.2% at the center of the axial FOV. This is the count rate curve. The prompt-minus-delayed count rate is more than 70kcps (kilo counts per second). This shows the stability. This is the temperature change. Temperature change is around 2°C in this room, but the sensitivity measured by Ce-22 was less than 1%.

This shows the MRI used for the PET/MRI system. The magnetic field was 0.3T permanent magnet type. It has a hole at the yoke area; the diameter is 17cm. Using this hole, the optical fiber PET was inserted from the back side of the yoke. This is the optical fiber PET, and inside the optical fiber PET, an RF coil with a 4cm diameter was installed. This is an animal PET.

We measured the interference on the MRI from the PET using this dual-chamber phantom. It has two chambers and a gallium contrast material was contained in this phantom. These are the MRI images without PET. Upside-down – this is a bubble, so this image is upside-down. This is the MRI image with the optical fiber PET. There is no difference between these two images. This is a comparison of the signal-to-noise ratio – no difference. So no change in the MRI images with and without the PET scanner.

Next, we measured the interference on the PET from the MRI. This is the PET image outside the MRI. This is the PET image inside the MRI. There is no change with PET images. And this is the MRI image with the optical fiber PET. This image is with simultaneously measured optical fiber PET and MRI images.

We measured mouse images using the system. This is an image of a mouse chest; you can see the myocardium here. This is an optical fiber PET image of the mouse measured simultaneously; you can see the myocardium here. It was simultaneously measured with these two systems. And using the same mouse, we measured the mouse brain, and this is the optical fiber PET image of the mouse. This is an FDG-PET image, but for this mouse there is no accumulation in the brain. These two are also

simultaneously measured optical fiber PET and MRI images. Next, I will talk about SiPM PET development. This is a compact DOI SiPM

PET system with a temperature-dependent gain control system. The Geiger-mode APD – also called SiPM – is a promising photo-detector for

PET with its small size, high gain, and less sensitivity to static magnetic fields, so especially for use in MRI. Many groups are now developing SiPM-based PET systems. Seoul National Institute University has finished; Tuebingen is, I think, now developing; UC Davis is, I think, doing some kind of work; and Hamamatsu is now developing. Many people are developing this SiPM-based PET system. However, SiPM has temperature-dependent gain and that may be a problem for PET systems. So we developed a SiPM-based DOI PET system with a temperature-dependent gain control circuit and tested it in MRI.

This is a photograph of a SiPM array used for the system. The array was from Hamamatsu with a pixel size of 25µm. The size of the channel is 3x3mm, and arranged in a 4x4 array; the size is 18x16mm. The temperature dependency of the SiPM array… This is almost exactly the same data that the other speakers had. Jae Sung Lee told us the temperature characteristic of -4%/°C.

This shows a developed SiPM-based DOI block detector. Scintillator size was 1.1x1.2mm, and there were also types of LGSO stacked in depth direction and pulse shape analysis was used to distinguish the two layers. The scintillators are arranged in an 11x9 [matrix] and optically coupled to the SiPM array. The 2D histogram shows a very good separation, with a peak value ratio of almost 10. This shows the energy spectrum. The energy resolution is 14-27%. This shows the pulse shape spectrum; we can see two peaks, so we can distinguish two layers by using pulse shape analysis. The peak to value ratio is 1.1-1.9.

16 block detectors are arranged in a 67mm diameter rings and the cylinders are read from the side of the board using a very small-diameter coaxial cable, so there is no amplifier around the detector rings. And this was light-shielded by using black tape. Inner diameter of the detector ring is 6cm and the outer diameter is 11cm, and the width is 2cm.

This is a temperature-dependent gain compensation circuit. Signals from the SiPMs are fed to the weight-summing board and then fed to the variable-gain amplifier. We measure temperature by using a thermometer and then converted to the PC. The PC calculates the correction factor for the temperature-dependent gain and sends the data to the variable-gain amplifier to compensate for the temperature-dependent gain of the SiPM.

This is the SiPM PET system. This is the detector ring and bed; both are positioned in a flexible arm so you can position them very easily. Signal from the detector rings are fed to the weight-summing board and then fed to the data acquisition system. This is the inside of the data acquisition system. This is the data acquisition board and this is the bias provided for the SiPM. And this is a personal computer.

Performance radial resolution is 1.6mm detected in the center of the FOV. There is some improvement of the resolution of axis by using the DOI information. Sensitivity is 0.6% at the center of the axial FOV. Count rate is more than 20kcps. You can see some variation in this data. This is the temperature-dependent gain of the SiPM. We measured stability without the temperature compensation circuit. This is the temperature change. This is the sensitivity change using a Cerium-22 point source. With a 1.5°C temperature change, the sensitivity count rate was changed about 9%. With the temperature compensation circuit, the deviation is decreased less than 3.7%. So stability was much improved with the temperature-dependent gain compensation circuit.

Using this SiPM PET, animal imaging was performed. This is a picture of imaging a rat brain and imaging a rat foot.

This shows a rat brain imaging F-minus study. You can see the skull of the rat. This is an image of FDG. You can see the structure of the brain. And this is an image of a mouse heart. You can see the myocardium. And this is the image of a rat foot F-minus study. You can see the bones of the foot and you can see a finger [=toe] here.

Using this system, simultaneous measurements of PET and MRI were performed. This is the MRI used for the simultaneous measurement. The MRI uses permanent magnet MRI with a 0.15T – very low Tesla – magnetic field MRI. Our RF coil was a 3cm-diameter solenoid type. This is the inside of the MRI. This is the magnet and gradient coil installed here, and this is the SiPM PET, and this is the RF coil. The phantom or animal is positioned here.

Using the same phantom, we measured the interference on the MRI from the PET. This is the MR image without PET, and this is the MR image with PET without noise reduction. Noise is increased and the signal dropped, and you can see an artifact here. But using some noise reduction, we can see a better image, but not as good as without PET.

Here are the trials of noise reduction of the MRI. Copper shielding is effective, but it needs RF impedance tuning. A noise filter reduced some noise. Common grounding of the PET and MRI was very effective. Removing the PET from the RF coil was very effective, but we cannot measure simultaneously if we do that. Disconnecting

the PET output, the images improved a little bit. Other trials were not effective. This shows the interference on the PET from the MRI. This is without RF. With

RF, the big [=a large amount of] noise from the RF was observed in the PET weight-summing signal. But the image of the cylindrical phantom was not different because the noise is lower than the lower threshold of the PET’s detector.

We measured simultaneous measurements of the SiPM PET and the MRI using rat-FDG studies. This is an MR image without PET. This is an MR image with PET. There is what looks like no difference, but the noise level is higher than without PET. This is a PET image with RF and gradient. This is a PET image without RF and gradient. There is no observable difference between these data. And these data are simultaneously measured SiPM PET and MRI images.

Now this is the setup of the laboratory. In the center, a 0.3T new MRI system is installed. On the left side the SiPM PET was installed. On the right side, a new optical fiber PET was installed. And by changing the two modalities together or removing them, we compared performance.

Right now, we’re looking at comparison of optical fiber PET and SiPM PET for integrated PET systems. Spatial resolution: now, optical fiber PET is much better than SiPM PET, but SiPM PET has the possibility to have a much higher resolution because they don’t use optical fiber; no light loss, so basically, the SiPM PET can obtain a higher resolution image than optical fiber PET. Interference with MRI: with optical fiber PET, there is no interference. The SiPM PET has a lot of interference with the MRI, but it can be improved. Cost of the fiber is high; SiPM is also high because of the sensor. The optical fiber is very expensive. Stability of the optical fiber is much better than SiPM PET because there is no temperature-dependent gain. The detector ring diameter can be reduced smaller than optical fiber PET because it doesn’t use optical fiber space. So the total score: optical fiber PET is better than SiPM PET if these problems are not solved.

In conclusion, simultaneous PET imaging inside the MRI was possible for both SiPM PET and optical fiber PET systems. The signal-to-noise ratio of MRI was significantly reduced with the SiPM PET system inside the MRI, probably because the distance between the RF coil and the PET was small. There was no change of MRI images with the optical fiber PET system inside the MRI. So, so far, we conclude that the optical-fiber PET system is much easier to realize high-performance and reliable integrated PET/MRI systems.

I would like to thank Osaka University School of Medicine – Prof. Hatazawa and colleagues for many things, especially animal studies; Neomax Engineering for the

MRI system development; Hitachi Chemical for LGSO development; Hamamatsu Photonix for position-sensitive PM tube and SiPM development and supply; Furukawa Machines and Metals for reflectors for block detectors; Kuaray Corporation for development of optical fiber bundles; Espec-techno Company for development of electronics; Itoi-Jyushi Corporation for development of plastic parts; SilverAlloy for development of heavy metal material shields. These works are mainly supported by the National Institute of Biomedical Innovation and others. Thank you for your attention. Hidehiro Iida

Thank you very much—

Multimodal Image Fusion Hiroshi Watabe

—Multimodal Image Fusion. Actually I was asked to prepare a presentation about the fusion of iPET/MR, but actually I don’t have anything to say, as it is so easy to do image fusion. So today I will discuss more generally what multi-image fusion is.

There are so far so many image modalities in the world like PET, PET/CT and MRI, SPECT, angiography and ultrasound. Each image modality provides a different aspect for inside the living body. X-ray CT and MRI provide us anatomical information and PET and SPECT bring us functional information.

What is multimodal image fusion? It is to align, in sometimes 3D or 2D, multiple images obtained from different image modalities, and display all images together with different color tables.

What are the impacts of multimodal image fusion? Of course you get more information and more accurate diagnosis. For example, by using FDG-PET and X-Ray CT images you can detect cancer more precisely; or the assessment of efficacy of treatment by comparing two images acquired pre- and post-treatment, by fusing different treatment after post [=by fusing different images pre- and post-treatment] and you can see easily the difference after treatment. Or sometimes it’s applied for surgical planning by a fused image of both morphological and functional images.

To perform image fusion of course we need image registration. What is image registration? Images from one modality to other are different in terms of the position of the object and the time when the data was acquired. Therefore, multimodal image registration is required to avoid misalignment and false assessment of images.

What we do during image registration is: Basically we have two images, target and source, and you see somehow the differences or similarities between the two images. Then you find how the two images can be adjusted by some transformation metrics or transformation function and you finally get the fusion image.

The goal of image registration is to obtain the best transformation function between two images. The most simple way is to just take a look at two images side by side and manually do the image registration, e.g. by using a PC mouse, but it is very subjective and takes a long time and you cannot do it every time. So there are several approaches to perform image registration. The most common way for image registration is the software-based approach.

In the software-based approach you define an evaluation function which measures the similarity of the two images and estimates the optimized transformation

function which minimizes this evaluation function by means of computer software. There is a bunch of software which do software-based approach registration, for example SPM, AIR or FSL. There are so many lists and each piece of software has a different algorithm for registration.

For example, some software uses contour match such as the “Head and Hat” method; or sometimes they use the difference between two images; or the regression between two images; or the ratio between two images; or, most popular – mutual information, which helps doing registration between very different images.

So what are the good things and the bad things about the software approach? Good thing is that all you need is just a computer – if you have two images and you have a computer, you can do it very easily. However, one problem of image registration using software is that the results are very highly dependent on the image quality, for example, if you have a very poor noisy image, you will get completely wrong data sometimes, and if you change the software and the algorithm, you will sometimes get different results. So you will rarely have a solution obtained.

On the other hand, we have another approach – the hardware-based approach. This is basically using an external device, like PET/CT or PET/MR, determining the transformation function and image registration is performed.

Nowadays most PETs have CT combined and clinically they use every ?????tingly.

Here we have the iPET/MRI developed by Dr. Yamamoto. The good thing about iPET/MRI in terms of registration is the simultaneous acquisition of PET and MRI images. In CT, principally you cannot acquire data at the same time, but in PET and MRI you can really acquire data at the actual same time. So here, this is the MR field of view, the PET insert was already shown by Dr. Yamamoto. The unique feature of this iPET/MRI is that if you dial here, you can move it like this. Not only you can fuse PET and MR images, but also you can fuse a different position of the PET image.

What are the good things and what are the bad things about the hardware-based approach for image registration?

Of course we know the position of two modalities very fast. You can immediately get the results of image registration and the accuracy of the registration could be fine, due to mechanical accuracy.

However there are several limitations of the hardware-based approach. Of course you need special hardware which is only like PET/CT or PET/MRI. And sometimes there’s a restriction of usage and you cannot insert everywhere, for example PET/CT has its own field of view and with PET/MRI, sometimes you cannot insert some

magnetically sensitive material. Usually [=often?] internal structure may not be correctly registered due to motion or non-rigid deformation.

So I think that the future direction should be the fusion of the hardware and software approaches for image fusion. So combine the software and hardware approach together, and finally you get better registration results. I will show some hints toward this fusion of the software and hardware approaches with a couple of slides.

I have done some multimodal image registration using an optical tracking system. This is POLARIS, which detects the geometry of the target by using two CCD cameras looking at the target, and POLARIS gives us the exact, very precise position of the target.

The concept of multimodal image registration by optical tracking system is you have different scanners in different places, but you share the reference marker, so share the same coordinates between two scanners by using a reference marker. It doesn’t matter, where you are and where the scanner is. First of all, we pre-determine the relationship between the reference marker against imaging coordinates of each scanner. If you take a look at the images together, we virtually combine the two positions of the images according to the reference markers so sharing the same coordinates between two scanners by using the reference marker. The relationship between the coordinates of the reference marker and the imaging coordinates of each scanner is predetermined; therefore we can virtually combine the two image positions together.

What is good about this optical tracking based image registration is that the method can be applied to any scanner, it doesn’t matter if its MR, PET, SPECT or CT, and it’s cheaper than real PET/CT or PET/MRI. In principle you can acquire data at different times, like ten years later, so 4D image registration is possible.

One example of this image registration, we took a look at the MR image of the phantom, we moved this phantom to PET and took another shot with PET, so the PET scanner and the MR scanner have different locations, but we finally combined them together.

Another example of image registration is using ultrasound and PET scanners. We prepared a gel with a ball inside, which is sensitive to ultrasound, and we put the target of the optical tracking system at the probe of the US [ultrasound]. Then we determined the coordinates of the probe against this box. Inside the box is the gel and ball, and inside the ball, there is an F-18 source; therefore we combined the ultrasound image and the PET image together.

In summary: There are many, many ways to do multimodal image fusion and you must understand your problem and choose the proper method for image

registration. Today I just showed the combination of hardware and software registration.

I believe that more applications can be anticipated using multimodal image fusion in future.

If you combine human and Navi [this is a reference to the movie Avatar], you will fly. Fusion brings you more power. Thank you!

Hidehiro Iida

Thank you very much, Prof. Watabe, for a very interesting presentation. Any comments or questions? Dr. Kanno please!

Iwao Kanno

Thank you very much. I disagree that POLARIS system is one of the best methods. I have no experience of POLARIS in my group, but because of my previous experiments using a marker on the head, I know that the head skin is always moving a lot, so on that point the POLARIS system might have the same problem. So I’m just wondering, why you recommend it.

Hiroshi Watabe

I don’t recommend it, actually. This was just another example just to show the concept of image registration. I agree that POLARIS is not accurate enough. There is another way, for example you can detect the position using a camera, so it doesn’t matter where the marker is. There is other hardware that should determine the position and location more precisely than POLARIS, I agree.

Hidehiro Iida

Any other questions? Last week we did mice measurement using micro-PET. The mice were nude mice, the weight was about 25g, with lung cancer. We injected FDG and we immediately did the FDG scan after a short period of CT. The FDG scan was ten minutes. We expected good matching between FDG and CT, but among four experiments [=in four experimental trials], we were never satisfied by the agreement so we wondered about the possible modification change of the total figure in the lung area during anesthesia. If you use CT, you cannot look at the time course of the modification of the shape, but I think it is very much the strength of using a hybrid MRI system. You have the opportunity to look at the time course of the modification of the structure, which you cannot do with CT. I think this point should be promoted in this symposium,

if you agree.

Hiroshi Watabe I agree.

Hidehiro Iida It is also true that there are lots of different options and also it is very

important to pick the right way – the software and the hardware. Thank you very much for your comment.

Any other questions? If not, thank you very much, Prof. Watabe. Okay, we will move onto the next talk by Dr. Tadashi Watabe, from Osaka

University. He will be talking about small animal experiments. Please.

Small Animal Experiment

Tadashi Watabe Thank you for my introduction. My title is Small Animal Imaging with

Integrated PET/MRI. The major advantages of iPET/MRI in small animal experiments are the

following three points. The first advantage is tissue contrast. The tissue contrast of MRI is much better than micro-CT, especially in the brain and abdominal organs. It’s easier to identify uptake sites in accurately contrast registered PET/MRI images than PET/CT. The second advantage is simultaneous PET/MRI acquisition; for example in the acute phase of ischemia model and early phase after drug administration. Simultaneous evaluation between changing MRI and PET is important. The third advantage is the shortening of anesthesia time. With simultaneous PET/MRI acquisition we can reduce the load of anesthesia for small animals.

Simultaneous PET and MR imaging was performed in rats and mice with various tracers using iPET/MRI, using optical fiber-based PET and permanent magnet MRI developed by Dr. Yamamoto and NEOMAX Company. We can easily access the animal in this open-type MRI system.

The ????? was positioned under anesthesia and simultaneous PET/MR imaging was performed. PET images were acquired in list mode, MRI was done with the first ????? FLASH TR/TE sequence. From now, I’d like to show many images of small animal experiments.

Firstly investigating immediate metabolic and morphological changes associated with acute traumatic brain injury (TBI) in rats. Two rats survived from TBI and the resultant MRI regions. Cerebellar FDG uptake was predominantly decreased in the brainstem and cerebellum compared with control rats. Cerebral glucose hyper metabolism associated with normal MRI was documented in rats immediately after TBI. Other two rats did not survive the TBI and showed brain contusion. Ventrical hematoma and sagittal lateral hemorrhage were evidenced by MRI and autopsy.

This slide shows physiological F-18 sodium fluoride uptake in relation to new growth in normal rats. In old age, these are groups of three-week, five-week and nine-week-old rats. On average groups’ shoulder joints show the highest uptake of all bone structures in most of the half body. In three-week-old rats, accumulation on skull base was significantly decreased [=lower] as compared to nine-week-old rats. In five- and nine-week-old rats, accumulation on temporomandibular joints and mandibles had significantly increased as compared to three-week-old rats. Furthermore, accumulation

on shoulder joints shows increased uptake as compared to three-week-old rats. Accumulation on vertebral bones was relatively stable among the three groups.

We tried to image normal rats using I-124. I-124 is one of the alternative MRI/PET nuclides. Its physical half-life time is approximately 4.2 days. One day after the injection, high uptake was clearly observed in the right lobe and left lobe of the thyroid gland in the PET image. In the fusion image, high uptake corresponded to the thyroid gland and trachea.

This slide shows physiological uptake of C-methionine in the normal rat abdomen. Methionine was used for evaluating amino acid metabolism on pancreas tumors; you can see methionine uptake in the liver and pancreas accurately corresponding to MRI.

This slide shows the image of a tumor model mouse. This mouse had a prostate cancer tumor in the dorsum, you can see a big tumor with low intestine necrotic region in the left MRI image. On the PET image, heterogeneous FDG uptake was observed in this tumor. The fusion image demonstrated that the FDG uptake corresponded to the viable region of the tumor.

From now on, I’d like to show an O-15 gas PET/MRI study. ????? was installed under anesthesia, tracheotomy was performed in normal rats. Artificial ventilation of O2 gas with O-15 labeled gas, i.e. O-15 CO, O2 and CO2 were performed in order. The study procedure was done within 60 minutes. You can easily access these animals not only to inject drugs from convoluted tail/belly but also to collect arterial blood sampling for the measurement of radioactivity.

In the O15-O2 study, the three-coincidence count reached a steady state after approximately seven minutes of continuous inhalation, and PET images during steady state acquisition clearly showed high uptake in the brain accurately corresponding to the MRI images. The highest uptake was found in the basal ganglia and thalamus.

In O15-CO2 study, the three-coincidence of the head also registered after approximately seven minutes. You can see clearly a high uptake in the brain corresponding to the MRI images. These images are similar to the previous O2 images.

In O15-CO study high uptake was observed in the venous sinus and bilateral internal and external carotid vein. Accumulation in the brain was relatively low, iPET/MRI demonstrated visualization of cerebral oxygen metabolism and blood flow in normal rats. For the next step we are developing a method for quantitative measurement of cerebral blood flow and cerebral metabolic rate of oxygen.

In conclusion, simultaneous imaging of small animal experiments was successfully performed with iPET/MRI system using optical fiber-based PET and

permanent magnetic MRI. At last I made a very useful iPhone application, iCurie meter, this app enables

you to know radioactivity any time, once you measure the baseline radioactivity. It will be useful in using short half-life isotopes like O-15 and C-11. You can install it from the App Store in your iPhone and iPad. Thank you.

Hidehiro Iida

Very nice. Thank you very much.

Lecture I

PET/MRI: Historical Perspectives and Future Opportunities

Hideo Murayama Our next speaker has received the degree of Doctor of Philosophy in medical

physics from the University of London in 1989. His major research interest is molecular imaging technology, particularly positron emission tomography, multimodality imaging systems, gamma and X-ray detector technology, 3D image reconstruction and use of imaging techniques in drug development, and so on. He also contributes to education and is co-author of the famous textbook Physics in Nuclear Medicine. He has also built up and fostered the Center for Molecular and Genomic Imaging Biomedical Engineering Graduate Group at UC Davis. As you know, the laboratory team is one of the most active research groups in the world. They are exploring the integration of PET imaging technology with high resolution anatomical imaging provided by magnetic resonance imaging or X-ray computed tomography. Today we are expecting a special lecture entitled PET/MRI: Historical Perspectives and Future Opportunities. Prof. Simon Cherry, would you please open your exciting talk.

Simon R. Cherry

Thank you very much for the kind invitation from the organizers to speak here, it’s a great privilege to be here, we’re having a few technical problems which hopefully we will resolve soon. It’s very hard to give a talk on imaging without images! Thank you so much, thank you.

What I want to do today is to give you a little bit of an overview, both going back in time and looking a little bit at the history of this field and it has developed and giving you an overview of some of the more recent developments.

As you heard earlier today it really started here in Japan 25 years ago, with the work of Prof. Iida looking at the use of magnetic fields to constrain positron range, and you’ve already seen this data today, and Prof. Iida was very smart in what he said in his paper, if you read what he said. He talked about the fact that it would be very difficult to reduce positron range using current instrumentation based on photomultiplier tubes (PMTs), but he already could see into the future, he said in the future a new detector might be invented having good spatial resolution being unperturbed even by strong external magnetic fields, and of course, with the advent of APDs and SiPMs, that is exactly what has happened.

Going back to the early days, following that paper, there was… the first real suggestion of actually combining these into some kind of combined imaging system came in this patent application from Bruce Hammer and colleagues where they showed this concept of placing detectors inside of a MRI system. These would be detectors based on scintillators with long light guides to bring the light out from the magnet to PMTs that would sit outside of the magnet. And again the main motivation at this point was again to be able to reduce positron range.

Some of the first experimental studies were done by Ray Raylman, who was at the University of Michigan at the time doing his PhD. The title of this PhD was Reduction of positron range effects by use of magnetic fields, this was also really the first use of an avalanche photo diode (APD) in this context as well. It’s amazing – I contacted Dr. Raylman a few months ago, to ask him for some information from his PhD thesis and he said, “Well, I still actually have the experimental setup, it’s in my garage at home,” so he went back and took some photographs for me of this setup that he used back in 1990 to take some data to look at positron range effects in different magnetic fields using a single APD with some water cooling and attached to a plastic scintillator and then he looked at the absorption and took the path length of positrons in different materials as a function of different magnetic field strengths. Unfortunately that work never got published.

The work that did get published was a few years later, again from Bruce Hammer’s group, where now you see an experimental setup with a magnet, you see the long light pipes coming out – so this was very much like what they showed in their initial patent application. So they have two coincidence detectors here, long light guides, and they measured the reduction in positron range at 5T in this case using a Ga-68 source. There were three different papers that described this work and some of the modeling and the experimental data.

The first real biomedical experiment involving both PET and, not MRI in this case, but NMR spectroscopy, was this study that came from Paul Marsden’s laboratory at the University of London where they took a standard NMR spectrometer and inside that spectrometer they placed a scintillation crystal with a long optical light guide going to a PMT, this was at 9.4T. And the sample they were looking at was an isolated perfused rat heart that was infused with F-18 fluorodeoxyglucose, constant infusion, and they tracked the radio tracer uptake under different conditions, normoxic and hypoxic conditions, and at the same time they acquired phosphorous NMR spectra. So this was really the first acquisition of simultaneous NMR signals and PET signals or at least radiotracer signals in a combined experiment.

Dr. Marsden presented this at a conference that I was at, and this is what got my laboratory interested in this concept, because around at the same time we were developing a system for small animal imaging called the microPET scanner and the detector that we had based this scanner on utilized an array of LSO crystals, as you see here. We were using a multichannel PMT to readout those crystals, but the problem was that we had to pack these crystals in tightly in a small bore ring for an animal system and we were using these bulky PMTs. We had to deal with a packing issue and to get around that we were using an optical fiber connection between the scintillation crystals and the PMT.

When I saw Dr. Marsden’s talk, I immediately realized we could take this same concept and put this inside of an MRI system to actually do imaging where rather than using a single light guide and single crystals, we would use bundles of optical fibers to take the signals out of the magnet. So we immediately started a collaboration with Dr. Marsden at that time to develop a system.

So here is the very first system we developed, it’s very crude as you can see, it’s a ring of LSO crystals wrapped in Teflon reflector, and rather than reading out in the normal radio direction, the fibers are connected at the ends of the crystals and readout along the axial direction. So we have crystals here, fibers coming off the end, and we use very long fibers, 3-4 meters long, to bring the signal out of the magnet which you see here. This was an interventional MRI system we were using, relatively low field and we bring the signals out through the long optical fibers to PMTs that were sitting a long way outside of the magnet. So we imaged a very simple little phantom inside this very small scanner but we were able to resolved the little spots, this is the MR image, this is the PET image taken the same time, and there’s [a photo of us sharing] the obligatory glass of champagne to celebrate getting our very first images. Some of you may know some of the people in our field that are in this picture, Dr. Stefan Siegel, who now works for Siemens, Dr. Yiping Shao, who is at MD Anderson Cancer Center.

So this got quite a bit of attention at the time, this was a write-up that was done in Science and they got a couple of people to comment on what they thought about this work, and so Dr. Z.H. Cho made the comment that he thought performing feats like this, simultaneous PET-MRI, would be very difficult in people. It’s rather ironic but he was actually one of the first people to do hybrid PET-MRI imaging in humans as you’ll see in a little while. And Paul Lauterbur, the Nobel Prize winner for MR, just reminded us that, of course, this is not really new, people have been discussing this kind of thing for a long time.

So we moved on to try in vivo imaging, I think these are some of the first in vivo

images back in 1997. So this cross-section through the head of a rat is the MR image here, here is the FDG-PET image, not particularly great image quality, I realize, but you can see the FDG uptake in the brain quite easily. This was a somewhat more advanced system, we had gone up to 72 detectors. Remember earlier today from Hamamatsu, we saw a proposal for a scanner with over 600,000 detectors, this one has just 72 in it, but you can still create a single slice image like these, and this time we were using a clinical 1.5T MR system.

Now, at the same time Dr. Ben Pichler was already thinking about the concept of getting rid of the PMTs, getting rid of the optical fibers, which were clearly the weak link in the chain because you were piping the light down very long optical fibers losing a lot of the signal. It hurt the energy resolution, it hurt the timing resolution. So he was already thinking ahead to what kind of detectors could we use without optical fibers. He was looking at the possibility of using APDs and so he was testing these at very high magnetic fields up to 9.4T and showing that they worked extremely well. That led to a subsequent development of detectors based on APDs for PET-MR. Again, this was back in 1997, so a long time ago.

So there was this initial burst of activity in the field that is shown here; if we look at the number of publications on PET-MR versus year, there was this little burst of activity in the mid 1990s based on the work that I’ve just shown you, but then everything came to a complete stop in the late 1990s. And the reason, if you are wondering, is because of the advent of PET-CT, it completely took over the field, it completely took over the research funding, it was very difficult for us in the US to get funding for PET-MR at that time, everybody was so keen on PET-CT. But in the long run I think it has helped us because now we are building on the established foundations that PET-CT has laid for us and its huge success, but certainly at that time it was very challenging and trying to get industry interested in PET-MR was just impossible.

So, of course PET-CT, both in the clinical world and the pre-clinical world, has been a hugely successful tool, but of course at this conference we know what its limitations are, and we know why we want to still pursue PET-MR.

So a number of groups carried on working on a fairly low level at this with whatever funding we could put together, so P. Marsden’s group continued with the fiber-optically-based systems building more and more sophisticated systems with multiple layers of scintillator material. And also Ray Raylman, who had originally done that very early work in his doctoral thesis, also built a fiber-optic coupled-based systems, as you can see here. And of course, you have heard about the efforts here in Osaka and Kobe to develop optical fiber-based systems as well. So a number of these systems came

into being, a number of animal studies started to be performed, and slowly but surely interest started to build again.

There were also some novel takes [=ways of taking things] on how to integrate these two technologies, everything that had been done up to this point used existing MR technology, and tried to make the PET fit around that, essentially. The group at the University of Cambridge took a different approach where they went to a very novel magnet configuration, a split magnet. As you can see it has a gap here in the axial direction and they decided to fit a ring of PET detectors inside that gap. They’re still using optical fibers, but now those optical fibers are coming out in the radial direction rather than in the axial direction. That is much better because it gives you a lot more space to play with inside of the magnet. So here you see the scintillation array on the front very long but straight optical fibers going to PMTs. This system has now been completed and images have been published.

The group at the University of Western Ontario looked at this concept of a field-cycled MR, the idea of switching the field on and off and acquiring PET data when the field was off. So this is a cycled system where you have a polarizing magnetic field and a readout magnetic field, they are then switched off for a period of time during which you can take PET data, so now you can just use conventional PET detectors, you don’t have to use optical fibers, you don’t have to use or any clever tricks. The downside of course is that you are only acquiring the PET data for a fraction of the imaging time, so in terms of your effective sensitivity, that’s quite a problem. The other problem is that these field-cycled systems are relatively low in field strength.

Our next attempt at working on PET-MR was to move to APDs, so replacing the PMT with a photosensor that was relatively immune to magnetic fields. We still weren’t confident enough to completely ditch the optical fiber coupling, so the detector module we developed is based on a scintillation array, a short length of optical fibers going to a position sensitive APD. So this is a single photodiode that has a resistive readout that allows it to be continuously position sensitive, so a little different than the photodiode arrays we heard about earlier.

And then we have our electronics, as much as possible nonmagnetic, but we moved this far enough away that it is outside the MR FOV, so we felt that would give us the best chance of avoiding any interference between the two systems.

So we built an insert based on this idea, you can see there are 16 detector modules in the ring here, with 8 of them with the optical fibers coming out to the left and the other 8 to the right. We have some copper shielding to reduce interference that you can see over the electronics on this side, it has been removed on this [other] side, so

you can actually see the electronics. And then here is this insert sitting inside of a 7T pre-clinical magnet, it was

designed to fit very snuggly inside the available bore there, which is about 12cm, and here is the acquisition system that sits outside of the magnet.

So we had done quite a lot of animal studies with this system. I am just going to show you one example here, which combines both structural MR, but also looking at apparent diffusion coefficients, which is often being used as a way to monitor cancer treatments, and also then combining that with simultaneous FDG-PET. So this is a mouse model of colon cancer, there’s a xenograft tumor in the flank of the animal that you can see here. What you are looking at here is a control animal. Day 1, day 3, day 6, FDG-PET, the tumor is gradually growing, metabolic activity is going up, diffusion coefficient is staying reasonably constant. In this treated paradigm they were looking at a combination drug treatment here. What you see is that early on after initiation of treatment you see the FDG uptake increases, this is probably an inflammatory response, you often see this in response to chemotherapy, and then by day 6 you see a dramatic reduction in FDG uptake.

If you look at the apparent diffusion coefficient images here, you see that very early on you see an increase in diffusion suggesting that you have got cell death going on, that again increases at day 6. So a nice combination here of two of the techniques that are used to monitor tumor treatments, overlaid on the structural MR.

Earlier on we had a discussion about positron range measurements, so I put this slide in at the last moment to show our attempts at measuring positron range reduction, using that scanner that you just saw. So this is at 7T, we have three different PET radio nuclides here, F-18, Y-86, Br-76, these are the average positron energies ranging from about 250keV up to 1.18meV. The blue lines here are acquired inside the magnet, the red lines are acquired outside the magnet and these are point spread functions through the reconstructed images. And so you can see that with F-18 we essentially get no difference with magnetic field but once we got to B-76, we get a very clear improvement in the point spread function due to the restriction of the positron range.

So at the same time Ben Pichler’s group at the University of Tuebingen was also developing an APD-based PET-MR system with no optical fiber coupling at all, so they were directing coupling the scintillator to an APD array. They build this insert pretty much exactly the same time as we were building ours. They have also done a huge amount of work with this, both characterizing the interference between PET and MR showing, that there is essentially no interference between the two modalities.

Mostly doing a wide range of animal studies. For example here you are seeing a glioma model in a mouse with various radiotracers and a whole slew of MR sequences as well. And so I think both these prototype systems that we developed really convinced the community that you really could do PET and MR together without major artifacts and it could be a very useful tool for biomedical research.

A number of other groups have subsequently developed APD-based PET-MRI systems as well, this is a system developed at Brookhaven National Laboratory. This is a very compact system as you can see here. It was a derivative of a system that they developed for imaging awake animals, animals that had not been anesthetized, originally the animal wore one of these small PET scanners around its head and it was freely moving, they then adapted this to be a magnetic field insensitive system that could go inside of an MR scanner.

So a lot of development pre-clinically, the first kind of hybrid combination for human imaging was in Korea, where they took a HRT, a high resolution brain imaging system. You see on the left here, in an adjacent room they put a 7T MRI and they had this shuttle system that moved the patient back and forwards between the two. This is perhaps taking avoiding interference to the extreme, by having these things pretty well separated and moving the patient some distance, but it certainly allowed for the ability to very quickly move from the PET to the MR with minimal patient movement and therefore good registration of the data.

So all of this activity as you can see has led to a complete resurgence in our field, very, very strong growth. The numbers for 2010 are probably underestimated because they are based on the publications that are currently in the databases which are not complete yet for 2010. So we’ve seen this just incredible growth over the last 3 or 4 years in the field, a lot of exciting things, much of which you have heard about today already and some other things you will hear about tomorrow.

And of course the big thing that has changed is that industry has got involved. And I think that that has made a really huge difference to the field. So Siemens were really the first to get heavily involved in PET-MR, and they developed this brain PET insert as you have already heard about today. So this is the system installed at Massachusetts General Hospital and some of the first simultaneous PET-MR images of the human brain taken from that system.

As we have heard we are even up to 9.4T now. It is nice to see that Prof. Iida, who is the one who started this field, is still involved 25 years later and through the positron range measurements that he talked about earlier, and I think many of us many years ago would have had a hard time imagining that we were going to be doing

simultaneous PET-MR in humans at 9.4T, but the reality is obviously here. Now in terms of clinical applications, there was somewhat of a debate about

whether you need to do PET and MR simultaneously or whether you could do them sequentially and just have the machines close to each other. So the Philips system that has been developed so far, involves sequential imaging. They have an MR and a PET system in the same room in fairly close proximity and again a shuttle system that moves the patient between the two, and this is installed in a couple of medical centers now and is being investigated as a clinical tool.

Obviously the advantage of this is that you don’t have to do much modification on existing PET and MR systems, the downside is that you cannot get data at the same time and of course there’s an increased risk of patient motion between the two studies. It will be interesting to see as we go forward which of those wins out in terms of being the clinical tool that people will prefer.

So moving on, I want to tell you a little bit about some of the work that we are looking at for next-generation PET-MR systems in our lab. We have been looking at fairly high resolution scintillator arrays going well beneath a millimeter in terms of pixel size, these are actually 0.5mm pixels in this array here. We have continued to look at these position sensitive APDs. We are putting them on both ends of the scintillator array so that we can get depth of interaction information; the idea being that when the source interacts close to this APD, you get a bigger signal here and a much smaller signal here. When the source interacts towards the bottom, that ratio flips, therefore by looking at the ratio you can determine not just which crystal the interaction took place in but the depth at which it took place as well. This is important if you want to get uniform spatial resolution across the field of view (FOV).

So we have been able to show that in the depth direction, so the direction between the two APDs we can get a resolution of about 2mm or so, and that is fairly good.

We have also been looking at this idea of moving towards tapered scintillation arrays, because one of the problems is that as you go from one of our previous scanner geometries that look like this, if we want to bring the detectors in to get good sensitivity and we make them thick, which we need also for good efficiency, you see that we have these big gaps where gamma rays can come through and not be detected and that overall is throwing away potential sensitivity. So if you could angle the scintillation arrays and close down those gaps a little bit, you could potentially significantly improve the sensitivity. So we have now fabricated these tapered scintillator arrays and they work very nicely. Simulations tell us that we should get quite a big sensitivity

improvement by using such arrays. In the red is a plot of the sensitivity with standard rectangular arrays, in the blue, as we move across the transaxial FOV, it shows you what the sensitivity profile looks like with these tapered arrays, we get a significant increase. So we are looking at building a small animal PET scanner for mouse brain imaging based on this approach; and again using very small crystals, 0.5mm sized crystals.

Also we’re interested in how far we can push this crystal size and so we have managed to have manufactured arrays that have crystals that are down to now 220 microns in size, you can see there are problems with this, you can see first of all the crystals are not regular in size, it is very hard to control the cutting here. You will also see that the reflector that we have around the crystal now starts to be a substantial fraction of the area here, so that is going to hurt our efficiency. Nonetheless, we were very happy to actually manufacture this in the first place, and as you can see here, at least in the middle of this array, we did a fairly good job in resolving these very tiny elements.

In theory we can go to even smaller arrays. Whether it is really practical to do this yet in a system is questionable, but I think for very high resolution imaging in, for example, the mouse brain, we have convinced ourselves that we need to get down to elements that are around this size if we are not going to be limited by the detector but we are going to be limited by other things such as the positron range and the non-collinearity. So we are quite serious about trying to pursue this eventually to a system level.

Now in terms of a PET-MR we have been developing a second generation system based on APDs. We started this work before the revolution with SiPMs and we were too far down the path to change when all the recent activity happened. So this is going to be a more sophisticated system than our previous ones, four rings of detector modules, really good cooling and air distribution system to keep everything very stable thermally, which is really important.

We have actually built this system now, well, we have built two of the four rings, so it is almost finished, those two rings are complete. You can see the detector modules here, the scintillation crystal sitting on top of the APDs, and here is some of the electronics we have developed.

Fresh data just last week, we just got our first results from this system: Not a very exciting image, it is just two point sources, as you can see the sinogram data here, the reconstructed image, but this is essentially first light from this new system. You are the very first [people] to see it.

So of course the excitement is SiPMs and the possibility of them being the future for PET-MR. We all know their advantages, they have the gain of PMTs and they also have the field insensitivity and the compactness of APDs. So just like everybody else, we have been looking at these devices, we have been collaborating with a company in the Boston area to look at some fairly novel SiPMs. We like others have convinced ourselves that these things work just fine inside of a magnet.

The devices we are working on are so-called position-sensitive SiPMs. They have a very small resistor between each microcell, between each Geiger-APD, so they are continuously position sensitive across their surface. You don’t have to worry about matching your crystal size to the pitch of the SiPMs, and so with this technology we can resolve small crystals, these are again 0.5mm crystals that are easily resolved.

But we’re a long way behind other people as you have already heard, Jae Sung Lee’s group presented just earlier today, showing beautiful results, already constructed a complete system based on SiPMs and shown first simultaneous PET-MR images. Of course here in Osaka as well Dr. Yamamoto’s group, as you already saw in the earlier presentation as well, developed a very flexible, very compact system, and also have first images with that system as well.

So about five years ago I gave a talk at the Society of Nuclear Medicine [Annual] Meeting where I showed this slide. I was being a bit provocative talking about where the field was going and this idea that perhaps eventually we would have very long axial FOV PET scanners inside whole body MRs, we would be able to quickly produce whole body MR and PET images of the body. And in terms of PET we’d incorporate time of flight (TOF) we would have a huge improvement in signal-to-noise ratio and I have to say people laughed at this and they said “Well, this is a nice suggestion, but it seems a long way off.” I must admit I thought it was a long way off as well, but if you look out there now at what people are planning I think we are not far off that.

This is the HYPERimage consortium that is funded in Europe to develop a whole body TOF PET-MR system, and if you look at this schematic here it is really not very different than what I showed on the previous slide. So they are using SiPMs, they are looking at fully digital SiPMs as well, as a possibility for this project and so I think we are very close, just a few years probably from realizing this concept where we can rapidly scan the whole body with both a PET and MR in a hybrid system.

So really I think the challenge is changing from the challenge that we have addressed as a community, which is more on the technological side, to now, how we are going to use this technology. And really I think the sky is the limit here. A lot of what we

have been doing so far, most of this has been combining structural imaging with metabolic imaging and that is all fine and good, there is lots of interesting things that you can do with that.

But MR is just such a powerful technology, there are so many different aspects of biology and medical applications you can look at with this. Of course with PET we have got the flexibility, the chemistry that allows us to look at so many different biological processes, so I think the key now is for us to find how to use this technology.

And so my new pie-in-the-sky slide, since the one I had five years ago is almost coming to fruition already, is to think about this combination. So we are all hearing about hyperpolarized MR and it is very exciting, the idea of being able to trace metabolic pathways with very high sensitivity, albeit over a fairly short time scale using C-13 labeled compounds. So this idea of perhaps being able to do dynamic combined C-13 hyperpolarized MR with C-11 radiotracers with PET to really interrogate metabolic pathways and different aspects of them simultaneously and during interventions, I think is perhaps an exciting area that we can look forward to and certainly one that I think would be very, very challenging right now to do, but no doubt somebody will do it pretty soon and show me that it can be done.

With that I would like to summarize and say hopefully I persuaded you that PET-MR has a much longer history than perhaps some people think, it has been going for 25 years now, and started here in Japan. PET-MR systems of course have been successfully developed for small animal and human imaging. It is a tremendously powerful multimodality platform for research and we are going to see where the clinical applications lie, I think very soon. And although most of the systems currently out there are based on APDs, there is not much doubt in my mind that in the future most of the systems will probably be based on SiPMs.

So with that I would like to acknowledge all the people who have worked in my laboratory over the years on PET-MR, and also many people from the field who have kindly provided materials for this talk and thanks very much for your attention.

Hideo Murayama

Thank you for your excellent lecture. Then, on to the discussion or some comments. Alright.

Participant 1 (male, Japanese)

Thank you for the nice lecture, I would like to ask you about the future of optical fiber-based PET.

Simon R. Cherry

Well, I thought you would want to tell me about that. So I think you made some good points today, you actually gave me hope that optical fiber PET is not dead yet. I think certainly as you point out, using optical fibers you can really eliminate any interference of any significance and that is good. On the other hand I think if I look to see what people have done both with APDs and SiPMs, I think it is clear that we are slowly understanding how to solve the interference problems and many people have been very successful. And I think with the latest results I am hearing from the Siemens whole body insert that they have developed, which is a lot of detector material and a lot of APDs, and it seems to be working pretty well. So I think that the interference issues are slowly going away, we are understanding how to solve them and I have to say, although you may not want to hear it, that the days of optical fibers I think are over or will be soon.

Hideo Murayama

Any other comments or questions?

Participant 1 (male, Japanese) It was a wonderful lecture, thank you very much. I think we should be aware

that a lot of technologies in the MRI field are also going on. One of the most important ones is the multiple transmission system to guarantee [against] the possible distortion of the static magnetic field, and as a PET person, when looking at the real exam of hybrid MR/PET scanner, I feel a little bit ashamed to be able to provide only one image from the PET side, at the same time as MR are providing lots of different functional images, so I think we should think of a methodology to extract the transient change or multiple functions from single session of the scan, maybe multiple injection studies or something like that. What do you think?

Simon R. Cherry

I agree with you completely, I feel equally badly when I look at how long it takes to take data from just one radio tracer and in that same time you can run through a whole bunch of MR sequences and get all kinds of different information, and so I absolutely agree. This idea of injecting multiple things is interesting if their kinetics are different enough then possibly you can extract it. Technically now we have higher and higher sensitivity systems, so the statistical quality of the data is getting better and

reconstruction algorithms are getting better. So maybe it is time to re-examine this but I think it is hard because our data is always still going to be pretty noisy so how much can we really extract? But your point is a very good one and we should brainstorm about it, think how we can do better.

Hideo Murayama

Alright. Other questions? How about you Prof. Hatazawa?

Jun Hatazawa Thank you very much. I now understand the long history of PET-MRI. We have

a very short history in Osaka, I think we just started the PET-MRI project for the clinical demand, I mean [with the idea that] the CT of PET-CT should be replaced by MRI. In that stage only maybe T1-weighted image or T2-weighted image or some morphological imaging of MRI is good enough, I think. So we chose a very cheap system for MRI to develop broadly, I mean 9T or 7T MRI combined with PET, it is maybe excellent, but who or which center can install such a system? That is just from the clinical point of view. Maybe it’s just like a F1 car, but in the clinical setting we need Corolla, Opel, Toyota, maybe, so our way is just like the convenient car for driving, that is maybe 80% good enough for clinical use, I think. So we are still working on our way and of course we are trying to do the F1 car just like Toyota. They started the Corolla and now they attempt the F1 race. Maybe we need both of them, but on the clinical side we always need very convenient and stable, and hopefully cheap imaging modalities. That is my comment.

Simon R. Cherry

I think those are very good comments. If you look historically, I have to say that people seem to go for the best they can get, so if you look at the PET field, a number of companies have tried to come out with products that are lowering cost and where the performance is a little bit worse, and none of those companies survived. And it is interesting that as new generations PET scanners came out everybody was scrambling to get the latest and the greatest, so cost did not seem to deter people. And although I don’t know the MR field as well [as thoroughly as I know the PET field], it seems that similar things are happening there, that as field strength has gone up, the clinical field strength has moved up as well. On the other hand, I think we are entering a time when there is a lot more focus on health care cost and the sustainability of the cost of technology increasing so we may see a different emphasis on cost coming in the future

that may change the balance. So it is a very interesting question, and I don’t know which way it is going to go, and so your arguments are very good, that maybe for the majority of examinations you might do clinically, a modest field strength MR with a decent PET, is good enough, and would serve us better. On the other hand, I think those of us working in technology always want to push the frontiers and do the best we possibly can, and let other people worry about the economics. Because sometimes you develop something that at first is very expensive, but then the costs rapidly come down, either because the technology becomes cheaper to make or because the volume goes up. So it is not always necessarily the case that something that appears expensive first has to end up expensive, so that is the other thing to bear in mind. So I think it is quite a complex thing. I have to say that during my career I have tried not to worry too much about the cost, sometimes we have been criticized for that, but we try and focus on trying to push the science forward.

Jun Hatazawa

Just like you said, this time, from the clinical point of view we need the MRI, maybe it should be… a low Tesla MRI is good enough, but now we are testing a new medical therapeutic methods or new medical applications using IPSN or many regeneration medicine testing the effect of the therapy, so in such new therapy we need the basic imaging animal experiments and with patients we need more information not provided by the present PET-CT, or MRI alone, or PET alone. So for that future medicine, I think that PET-MRI should be used, and I remember [=recall] the history of FDG-PET oncology application. The FDG accumulate at the tumor, this evidence was proved in the early 1980s, but it cannot be used in clinical practice because only one slice or several slices were imaged using PET at that time, so whole body PET imaging was possible in 1990s after many technological developments. Then the FDG-PET oncology is used in the clinical setting dramatically [=Then there was a dramatic increase in the oncological use of FDG-PET in the clinical setting], so we need technological development just as discussed this afternoon and we need the future clinical needs. Maybe tomorrow we’ll discuss more about this problem after Prof. Pichler’s lecture and Prof. Sorensen’s lecture.

Simon R. Cherry

They have been very quiet about it but I am sure they have opinions!

Hideo Murayama

Please.

Participant 1 (male, Japanese) Let me ask another question. So you know although we have worked hard on

the development of SiPM and PET-MRI. Some people ask me whether such a SiPM-based PET-MRI will be really necessary, because the TOF information would not be necessary in the PET-MRI, because the scanning time of the PET-MRI is mostly dependent on the MRI, because as you know the MR scan time is much longer than the PET. Do you have any idea how we should answer that question?

Simon R. Cherry

So they are saying you don’t need TOF because the patient is going to be in the magnet for a long time anyway. Not sure that is true, I think it depends on the MR examination you are doing, but there have been considerable speed-ups. And I think the reason people are often in the magnet for a long time is that they are doing many different things on the MR side, and maybe Dr. Sorensen can comment on that, but that doesn’t seem like a good reason not to use SiPMs, because I would use SiPMs if I was doing TOF or not. What is your alternative, your alternative is PMTs with optical fibers which in a human scanner is virtually impossible because of the volume of fibers you would need. Pre-clinically it is ok, but for humans it would be very tough, you can use APDs which is what Siemens are currently using, and while they have done a good job, I am sure they would rather have the gain and the timing performance that SiPMs provide. So even if you are not going to do TOF, I think you would still make that technological choice. And if you can have TOF, then why not have it, if it improves the signal to noise a little bit, you might as well have that. If the patients have to sit in the magnet for longer you might as well keep taking data from the PET I guess, and improve the signal to noise.

Participant 2 (male, Japanese)

Thank you very much, that was a very excellent lecture. I have a simple question. Do you think that the PET-CT is going to be replaced by the PET-MR completely in the near future?

Simon R. Cherry

If PET-MR will replace PET-CT? That is a very simple question, but not one that has a simple answer! I don’t know, I think CT and MR have co-existed quite

happily because they have their different strengths and I think the combinations with PET will also co-exist happily for at least quite a long time. I think there will be areas where PET-MR will be the technology of choice and there will be other areas where you are going to use PET-CT. I think the one force that was pushing against PET-MR was perhaps the expense, but the force that is pushing in favor of PET-MR is now a much bigger awareness about radiation dose. At least in the United States it has become a big issue, it’s a big issue on the public side, because of some unfortunate incidents with CT dosing, so I think that is one area where PET-MR has a big advantage and will help it gain some acceptance perhaps.

Hideo Murayama

We don’t really have time but one more question? No, alright. Thank you again for your lecture.

END