Transcription
Hello everybody, and welcome back. Now, up until now, everything we've looked at has described the different steps needed to generate some form of image, and we classify these steps as Magnetic Resonance Imaging, or MRI.
Now, we're going to shift our attention to a different form of MR, known as MR Spectroscopy. Instead of taking the signal from an entire slice and trying to figure out where that signal is coming from and at what signal intensity, here we're going to look at a specific region on our image. We're only interested in the signal coming from this location within the slice. And if we take that signal and Fourier transform it, we can figure out which frequencies contribute to the signal at that specific location. And when we plot those frequencies in a graph like this, we generate a spectrum of frequencies, and that's why this is known as MR Spectroscopy, or MRS. We are generating spectrums, and we're interested in the different frequencies that contribute to signal in one location.
Now, as you'll know from the previous talks, predominantly signal comes from water and fat. That's where hydrogen atoms are most abundant within the human body. And we also know at the same magnetic field strength, hydrogen atoms in water will precess at a slightly different frequency to that of hydrogen atoms in fat. And that's because the local magnetic field, the local experienced magnetic field by the hydrogen atoms in these different molecules, will be slightly different because of electron shielding and because of relative de-shielding within the water atoms. And the difference between those precessional frequencies is what's known as chemical shift, and it's this chemical shift that leads to chemical shift artifact that we see in images.
Now, you can see that this precessional frequency difference has allowed us to see how much water and how much fat, relative to one another, in this specific location there are. It's this frequency shift that allows us to delineate specific molecules within a specific location, and this is the basis for MR Spectroscopy.
Now, it turns out that there are not only hydrogen atoms in water and fat, but other molecules have hydrogen atoms that can also exhibit nuclear magnetic resonance and generate some signal. The problem comes when we are generating a spectrum like this is that water and fat have over 10,000 times the signal of any other metabolite that we're interested in. And if we have a way of suppressing the signal coming from water and fat, we can actually then look at the metabolites, especially that have a frequency shift of that between water and fat. If we suppress signal coming from water and fat, we can actually then get some information from these metabolites here because we reduce some of that background noise, and we can get usable signal coming from these metabolites. And if we look at the spectrum that's generated, we zoom into this graph, we can see specific peaks corresponding to specific metabolites. And these peaks here, or this spectrum here, is corresponding to a specific location on our image. If we perform an MRS sequence for this location and then repeat that sequence for a separate location, we can compare the spectrums between those two locations and see how the concentrations or the signal differences between the metabolites change between these two locations. And that's how we go about analyzing MR Spectroscopy images, and we make clinical inferences from these spectrums here, from the changes in metabolites.
Now, in order to generate these spectrums, first, we've talked about, we need to be able to suppress the signal coming from water and fat. And secondly, we want to generate signal only from this location. We don't want signal from the entire slice, otherwise we're not going to be able to talk about the concentration of metabolites at a specific location. So, first, let's have a look at how we go about suppressing signal from water and fat. And we've touched on these concepts, so I'm going to go through these briefly here. If you want more detail, go back to the inversion recovery and saturation talks within this lecture series.
Now, we can suppress signal coming from water using what's known as a chemical shift selective sequence, or a CHESS sequence. Here, we apply a radio frequency pulse that has a very narrow bandwidth that's specific for the frequency of water. That narrow bandwidth means we don't provide an RF pulse that matches the precessional frequency of fat. And remember, it's only RF pulses that match the precessional frequency of specific hydrogen atoms that will cause those hydrogen atoms to flip into the transverse plane, or at least it will cause the net magnetization vector to flip into the transverse plane. So, with this RF pulse, only the net magnetization vector from water will flip into the transverse plane and get transverse magnetization. Everything else will remain in the longitudinal plane with longitudinal magnetization. Once that net magnetization vector for water has been flipped into the transverse plane, we apply spoiler gradients, which cause the net magnetization vector to rapidly dephase in the transverse plane. At this point of time, when that net magnetization vector has been flipped to 90 degrees, there is no longitudinal magnetization for water. And once we've applied the spoiler gradients, no transverse magnetization for water, the net magnetization vector for water is zero. We can then run a normal pulse sequence with our initial 90-degree RF pulse. There is no longitudinal magnetization vector in water to be flipped into the transverse plane. Only other tissues, fat, and metabolites will contribute to signal for this pulse sequence here. Here, I've presented a spin echo pulse sequence. We're going to look at a specific MRS pulse sequence that will occur after this spoiler gradient has occurred. This will allow us to select specifically water to suppress signal coming from it.
We can also suppress signal coming from fat using what's known as a short TI, or short time to inversion, inversion recovery sequence, or short tau inversion recovery sequence. Again, we've looked at this before. We can flip all of the longitudinal magnetization vectors 180 degrees. They're lying anti-parallel to our main magnetic field. They are going to recover that longitudinal magnetization at a rate known as T1. There will be a specific point in time in which the tissues will each have their net magnetization vector in the longitudinal plane to be zero. And this occurs at different periods of time for different tissues. If we wait the perfect time to inversion time, where fat has no net longitudinal magnetization, it's got equal numbers of hydrogen protons in the spin-up and spin-down orientation, and then we run our pulse sequence again, we will have no longitudinal magnetization vector to flip into the transverse plane, and we'll get no signal coming from fat. I've gone through this in detail in the inversion recovery talk.
Another way we can suppress signal coming from fat is using what's known as saturation bands, where instead of saturating tissue based on their precessional frequency, we can saturate them based on their location within the slice. If we were looking at an MR of the brain here, we know there's a lot of subcutaneous fat in the scalp here that's going to be contributing to signal. We can place saturation bands over that fat to prevent that signal from contributing to the slice. In fact, we can place larger saturation bands that suppress signal coming from most of the region except the region that we're interested in. All of these techniques contribute to suppressing the signal coming from water and fat, allowing us to analyze the metabolites with much more accuracy. And again, when we look at those metabolites, we will see there are specific signal peaks that correspond to certain metabolites.
Now, we want to look at metabolites in a specific volume of tissue, or a voxel of tissue. Now, when we were looking at MRI imaging, the way we located where signal was coming from was by filling k-space and by using a frequency encoding gradient while we were acquiring the signal and using different phase encoding gradients every time we repeated a pulse sequence. That allowed us to figure out where signal was coming from. Here, we're not trying to figure out where the signal is coming from. We have specifically selected the region of interest. We know where we want to look at, and we're not interested, once we've selected this voxel of tissue, we're not interested where the signal is coming from within the specific voxel. We're only interested in which frequencies contribute to the signal from this voxel. So, how then do we go about selecting a specific voxel from a slice? Or this is the MRS sequence that is specific to MRS imaging and not MRI imaging. The predominant sequence that we use, I'm going to go through now.
If we were looking at a slice of tissue here and we wanted to look at this voxel specifically, we didn't want signal coming from the tissue here, how would we go about selecting this specific voxel? How would we localize the signal in this region only? Well, first, we only get signal from transverse magnetization. We need this voxel to have transverse magnetization in order to measure any form of signal. So, we can do what we've always done and select a specific slice. We apply a slice selection gradient in the longitudinal plane and we match the specific slice with an RF pulse, a radio frequency pulse that has the same frequency as the precessional frequencies along the slice that we're interested in. We will only get nuclear magnetic resonance in a specific slice of tissue, as we've done in all our MRI imaging so far.
Now, what happens if we apply that 90-degree RF pulse and just wait? We will get free induction decay, loss of signal due to dephasing of those transverse magnetization vectors, and we will get loss of signal at the rate of T2 star. And that loss of transverse magnetization will be happening in this specific slice. Nowhere else will have any transverse magnetization. Nowhere other than this slice will be providing signal.
Now, what we can do is apply a 180-degree RF pulse, but instead of in the slice selection or the longitudinal axis, we can apply it in the Y-axis of our slice. We've seen this combination before when we look at spin echo imaging. If we apply a 180-degree RF pulse that is specific to this Y-axis location, we're applying a Y-axis gradient here, again, a gradient along the Y-axis, and selecting an RF pulse that matches only a specific band of precessional frequencies. What's going to happen to the transverse magnetization vector in our original slice here? Where there is no overlap between these two slices, we are going to continue to get free induction decay, continue to get loss of signal in these regions here. Where there's overlap, there's going to be residual transverse magnetization from this initial RF pulse, and we're going to get this refocusing or spin echo generated in 180-degree RF pulse. We're going to generate a spin echo where there's overlap in the region of this slice here, colored in blue, where we had no transverse magnetization. This slice never got this initial 90-degree RF pulse. What's going to happen there? Well, the longitudinal magnetization vector is going to be flipped 180 degrees and slowly regain its longitudinal magnetization. It will never have any transverse magnetization and won't be generating any signal.
Now, we have created a column of tissue that is going to generate a spin echo here. Now, you might be thinking, well, why repeat that? And that's exactly what we will do in the X-axis plane. We're going to provide another 180-degree RF pulse, but this time for a specific X-axis location. Notice how I haven't called this the phase encoding gradient or the frequency encoding gradients here. They're not encoding gradients. They're matched to an RF pulse. In fact, what they are, they're slice selection gradients, but just in an orthogonal plane to that original slice selection gradient. These Y slice selection gradients and X slice selection gradients are going to specifically select a slice of tissue in which there's going to be a 180-degree RF pulse. Again, this 180-degree RF pulse here now is not going to generate any transverse magnetization except in the region that's overlapping with its initial spin echo here. Remember, in our multi-echo slices, if we repeatedly give 180-degree RF pulses at the right timing, we can generate multiple echoes in our sequence, and that's exactly what's happening here. We're going to generate an echo coming only from this specific crossover between these three slice selection gradients. We have generated an echo in a specific voxel, or a specific point within our slice. We will then measure that signal at TE when that echo is being generated. We don't measure that signal with a frequency encoding gradient. We don't want any changes in frequency within this voxel here. We want to use a Fourier transform to figure out which frequencies are contributing to the signal, and we don't need that frequency encoding gradient to figure out where the signal is coming from. We've localized this signal with these slice selection gradients, and this sequence is the basis for most of the MRS sequences. It's what's known as PRESS, or Point Resolved Spectroscopy, and most of the time when you're looking at MRS imaging or spectroscopy imaging, you're using some form of PRESS sequence.
Now, you may have heard, just for completeness' sake, of a sequence known as STEAM, or Stimulated Echo Acquisition Mode. Here, instead of 180-degree RF pulses, we use multiple 90-degree RF pulses. And remember, when we looked at 90-degree RF pulses that were close to one another, we generate what's known as stimulated echoes, and we can also use those stimulated echoes to read out our signal, allowing us to get a shorter TE. One of the problems with PRESS images is that we can't get very, very short TEs, and there are occasions when we need shorter TEs. And using that STEAM stimulated echo acquisition mode, we can get a shorter TE, but that comes at the cost of signal to noise. Here, we've got great signal to noise ratio. We get a large voxel of tissue providing a signal, and no signal coming from anywhere else with a large echo here, coming up to the level of T2 instead of T2 star.
Now, we can take that voxel from a specific region in our image and generate the spectrum, and do that same sequence again for a different voxel in our image and generate a different spectrum, and we can compare these two spectrums to one another. Now, what happens? You may have seen images where we are comparing multiple voxels to one another, where instead of taking specific voxels, we can take a larger region and separate them into multiple voxels, and this is what's known as multivoxel sequences. Again, we often use that backbone of a PRESS sequence, but here we introduce some phase encoding gradients in both the Y and X axis, and that will allow us to delineate some of these spectrums from one another in a larger voxel that we have selected, and then we can generate spectrums for each one of these voxels.
Now, here again, we're going to get some reduced signal to noise ratio, and our acquisition time is going to be much longer. We need to repeat this pulse sequence many more times just to get the information, the spatial information for these different voxels here. And you may have seen images like this where there's color overlay, where we're looking at ratios between these metabolites and assigning specific colors along a spectrum to those ratios, and that can give us a visual representation of how the metabolites differ throughout this larger voxel here. And we combine the underlying anatomy with that color variation to represent these metabolite changes. Now, that is not the most accurate way, and that's just a visual representation. You need to look at the spectrums themselves to get any form of useful clinical data.
Now, you can see how getting spectrums from a multivoxel sequence is good for heterogeneous lesions or small regions within the body where we're comparing signals very close to one another. Now, there is a downside to using such a large voxel area here in comparison to those single voxel sequences. If we have any magnetic field inhomogeneity here, we're going to drastically reduce the accuracy of these spectrums.
Regardless of which sequence we use, whether we use multivoxel sequences or single voxel sequences, a PRESS sequence or a STEAM sequence, ultimately we're going to generate a spectrum of frequencies from a specific location in our image, and those spectrum of frequencies correspond to specific metabolites.
Now, this talk is not a clinical talk, and there are multiple other metabolites to talk about, but I want to show you just with some examples how we can use these metabolites to make some form of clinical judgment from these spectrums. Now, here, these are common metabolites that we look at in the brain. We have a metabolite called NAA, or N-acetyl aspartate, that corresponds to healthy neuronal tissue. If we've got any form of disease process that decreases the number of healthy neurons, say we've had a stroke, say we've got MS plaque, or say we've got a metastasis that's not made up of neurons, we will get a decrease in signal in that location coming from the NAA or N-acetyl aspartate metabolite. And signal roughly corresponds to the concentration of the metabolite within that region. Creatine, as you may know, is a form of, as a marker of metabolism. It remains fairly constant no matter where we're looking at within our image, and it's a good internal standard when we're comparing multiple voxels. We can't look at a spectrum in isolation. We need to compare it to something else. These are not absolute values. These are values that we compare and see changes in different regions in our image.
Now, choline is associated with cell membranes. If you've got a highly cellular primary CNS tumor, we could get an increase in choline relative to our other metabolites. If we have a demyelinating disorder, we've got less myelin, we've got a higher concentration of axons within a specific region, we could also get an increase in choline here. And you can see how comparing these metabolites could give us some idea as to what's going on in the underlying pathological process.
Now, as I've said, we can use signal intensity as a proxy for concentration of the metabolite. Now, it's not only concentration that's going to affect the signal intensity of these different metabolites. We may get a change in our graph that looks like this, a reduction in the signal intensity from NAA. Now, that could be a reduction in the absolute concentration of N-acetyl aspartate, or it could be the type of pulse sequence that we use. These metabolites are going to relax at different rates. Any different tissue has a different relaxation rate, a different T2 star rate. It could be the type of pulse sequence that we use. If we use shorter or longer TEs or shorter or longer TRs, we're going to get changes in signal intensity, just as we got changes in signal intensity between water and fat. We may get overlapping of metabolites, and maybe another metabolite close to this that we've got an increase in concentration, and their frequency shifts are close enough that we get overlapping and widening of this curve here. There may also be local magnetic field inhomogeneities. Remember, this technique relies on having a very homogeneous magnetic field, a magnetic field that's constant, at least throughout the voxel that we're interested in. If there is a non-constant magnetic field within that voxel, we could get changes in signal intensity between our different metabolites.
Now, don't get caught up in the specific metabolites and how the signal changes. What I want to get across in this talk is how we localize a specific region to generate signal from that specific region, and how we can then take that signal, Fourier transform it, and generate a spectrum, and then make some form of informed decision on that spectrum when we compare that spectrum in one region to the spectrum in another region. And importantly, we need to suppress that high signal coming from water and fat, and there are multiple different techniques that we can do. We can use inversion recovery, saturation bands, or a chemical shift selective sequence to prevent that signal from interfering with these much smaller signals coming from metabolites.
Now, important to remember, MR Spectroscopy is not only for neuroimaging. We see it in breast imaging, prostate imaging, multiple different subspecialties within radiology can utilize this, and often we are looking at different types of metabolites depending on the region of the body that we're looking at.
Now, I know this is just a basic introduction to MR Spectroscopy, and we could spend days on this topic, but hopefully, I've given you a good basis, a good foundation to figure out how we go about actually generating these spectrums. Now, if you've made it this far in the video, you're interested in this physics, or you are preparing for a specific exam, you can test yourself in a question bank that I've linked down below in the description. Otherwise, I'll see you all in the next talk. Until then, goodbye everybody.