Transcription
Hello everybody, and welcome back. Today, we're focusing on diffusion weighted Imaging. I want to show you how we can generate specific MRI images, manipulate the data of those images, and make assumptions about the underlying diffusion characteristics of the tissues that we're imaging. And in order to do this, we need to understand the physics behind generating a b0, DWI, and ADC images. After this talk, hopefully, you'll understand how we generate these images, what they represent, and how they relate to diffusion, as well as why we can't look at these images in isolation and need to compare them to one another in order to make sound clinical judgments.
Now, in order to understand diffusion weighted Imaging, we first need to understand what diffusion is. Now, diffusion is the random or spontaneous movement of particles within a medium due to thermal collisions. It's what's known as Brownian motion, and we can represent diffusion using a diffusion coefficient, which is a product of a constant, the Boltzmann constant, the temperature of the medium, the size of the particles we're looking at moving in this medium, as well as the viscosity of the medium. Now, when we're looking at MRI images, we're primarily interested in hydrogen atoms. It's the hydrogen atoms that are providing us with a signal that we use to generate our image.
Now, if we were to look at this example of a pure liquid where the particles can move at random in any direction, if we were to sum up the movement of these particles in all directions, it would be equal. The net movement would be zero. That's what's known as isotropic diffusion.
Now, when we look at the human body, it's organized in a structured way. Anatomy follows specific paths. If you look at nerve fibers or muscles or bones, there's limitations to how particles can move in those mediums, and it's very rare that we get pure isotropic diffusion. So, if we have a look at another example where the molecules or particles are still moving with Brownian motion, but they are limited by external structures, here the net movement of these particles is now not going to be zero. There's going to be more net movement from top to bottom than there is from side to side because of these limiting external structures. A good example in the human body is white matter axon tracks, where hydrogen atoms can move easily up and down the axon but find it very difficult to move between axons because of those cell membranes and the myelin sheath.
Now, this skewed movement is what's known as anisotropic diffusion, where we get diffusion preferentially in one plane as opposed to other planes. Now, we can represent this using what's known as a diffusion tensor, where we create a matrix of the diffusion coefficients in the Cartesian plane, where we look at the movement in the x-axis, y-axis, and z-axis, as well as a combination of the axes. Now, in isotropic diffusion, our diffusion along the x, y, and z axis is all going to be equal, and when we compare those axes to one another, these diffusion coefficients are going to be zero. In anisotropic diffusion, there's going to be a skewing of this matrix and preferential diffusion in one direction.
And we can represent this diffusion tensor in what's known as a diffusion ellipsoid. In isotropic diffusion, the diffusion is the same in all directions, and we get this 3D representation of the diffusion in that medium. In anisotropic diffusion, you can see our ellipsoid changes shape here. We get more diffusion, say this is the y-axis direction of our slice, than we do in both the x and z axis. There's a skewing of diffusion, and that's what's known as anisotropic diffusion, and this becomes really important when we look at diffusion tensor Imaging.
Now, there's a third type of diffusion, which is also common in the body, and that's what's known as restricted diffusion. If we were to look at how the hydrogen atoms are diffusing within a medium that has many, in this example, organelles, we're looking inside a cell here. We've got mitochondria, we've got endoplasmic reticulum and lysosomes and Golgi apparatus. It's been a long time since I've looked at cellular biology, but the hydrogen atoms are not free to move here. We could also have increased viscosity represented by the changing color here. We see that the hydrogen atoms are moving slower than they are in these other diffusion models.
Now, there are multiple things that can cause restricted diffusion in MRI imaging, and a lot of it relates to pathology. If we have infarction, say in the brain, we get swelling of cells, we get a reduction in that extracellular space. We can have increased viscosity, such as in an abscess. We can have increased cellularity, where the cells are packed tightly together, and hydrogen atoms don't diffuse freely in that medium, such as in a glioma, a highly cellular tumor.
The objective of diffusion weighted Imaging is simple. We want to try and calculate how freely the hydrogen atoms are diffusing within the tissue that we are sampling, and what we can do is represent that diffusion based on a grayscale here, where restricted diffusion is represented by a lower numerical value, and unrestricted free diffusion is represented by a higher numerical value. And these numerical values is what we call the apparent diffusion coefficient, the ADC. That's what we're trying to calculate here.
Now, why is it apparent? It's not the actual diffusion. It's a result of an experiment that we run. We create specific MRI images that changed based on the diffusion of tissues, and we use that data to try and figure out what is the diffusion in those tissues, what's the apparent diffusion in those tissues. Now, remember, this is what we're trying to calculate at the end of the day, the ADC. So, how do we go about doing that?
Well, the first thing that we need to create is what's known as a b0 image and a DWI image. It's the combination of these two images that we actually take using a pulse sequence. We use a combination of these images to create what's known as an ADC map. This is not an actual image that we use a specific pulse sequence to create. This is a mathematically calculated image. That's why often your ADC map is very pixelated and looks very mathematical. It's a combination of these two actual images that we create to create a mathematical formula that gives us the apparent diffusion coefficient within the tissues that we're looking at. We don't have a specific pulse sequence that can just spit out an ADC map. There's no such thing. The only way we can create an ADC map is by creating a b0 image and a DWI image.
So, let's look at how we go about creating those images and how they help us to create an ADC map. The first image we're going to look at is what's known as the b0 image. Now, you'll notice this pulse sequence we've seen before. This is an EPI, an echo planar image. What we've done is we created a spin echo using a 90° RF pulse and a 180° refocusing pulse. So, in this example, we're creating a spin echo. Here, we rapidly fill k-space using oscillating frequency encoding gradients coupled with additional phase encoding gradients, and we can fill k-space rapidly.
When we look at diffusion weighted Imaging, we want to get a snapshot of the specific anatomy that we're looking at and prevent movement within that image, especially bulk motion. If we've got pulsating vessels, that movement is going to obscure or prevent our ability to assess different diffusion because that movement is going to introduce artifact into our image, and that's why we generally use rapid sequences, especially EPI, in diffusion weighted Imaging.
Now, this b0 image, if we look at this diagrammatic representation that I've made here, the gray matter is lighter than the white matter in this image. You can see this is a T2 weighted image, and it's a T2 weighted image because we've got a slightly long TE. We've created space here and here in the pulse sequence that we're going to use to create our DWI image. Now, what is contributing to signal here? It's the differences in T2 decay within tissues. Signal that is brighter here represents a tissue that keeps the transverse magnetization for longer than tissues that have darker signal here, which lose their transverse magnetization more quickly. So, our b0 image here has created a T2 star weighted image.
Now that we've created a baseline image, a T2 star weighted image, we need to figure out how can we change this sequence in order to null signal or reduce signal coming from tissues that have high diffusion in them. Now, in order to do that, we're going to add another line into our pulse sequence, another gradient. We've seen we have a slice selection, a frequency encoding, and a phase encoding gradient. These all help us to localize signal from our image and allow us to plot those signal values to create the MRI image that we're looking at.
Now, you can see this image has changed quite drastically, and it's changed because of these two gradients that we've added in. These are what's known as diffusion gradients. Now, these diffusion gradients are much stronger than the gradients that we apply when we're doing our spatial localization gradients, and these diffusion gradients are going to help us to figure out how much diffusion is happening within the tissues.
So, let's take a look at two separate parts of this MRI image. First, we'll look at a part where we know there's free diffusion. CSF has rapid diffusion of water within the CSF. It's relatively unobstructed. Then we can look at the head of the caudate, where we know that's highly cellular. We've got cell bodies, we've got slight restriction of diffusion here, and we can see that the caudate is brighter than our CSF here. Now, why is that the case?
Well, if we were to look at the hydrogen atoms that are moving within the CSF, they're moving rapidly. We flip all of our spins into the 90° plane, into the transverse plane, using this RF pulse. Those spins are going to lose transverse magnetization at a rate of T2 star. Now, we know that they lose transverse magnetization, and eventually we're going to apply a 180° RF pulse that's going to generate a spin echo, allow the leaders to become the laggers and catch up, bringing our signal back up to levels of T2. If this is all looking forward to you, I'd encourage you to go back to the spin echo talks as well as the EPI talk.
Now, what happens if we were to apply a strong gradient across each one of these voxels? We are applying a gradient across the whole image here, say in the x-axis direction. Hydrogen atoms on this side of our voxel are going to now precess at a slower frequency than hydrogen atoms at this side of our voxel. We are going to get a frequency difference based on this large gradient that we've applied across our image. Now, if these hydrogen atoms are moving freely while this diffusion gradient is being applied for a specific period of time, the movement of the hydrogen atom is going to cause a change in frequency, and that change in frequency is inducing a change in phase.
Now, what causes signal loss? It's dephasing of spins in the transverse plane. If these hydrogen atoms were to move during this diffusion gradient, we are getting a change in phase, and we are getting a reduction in transverse magnetization. The net magnetization vectors are still in the transverse plane, but they are more out of phase with one another as opposed to if they stayed stationary. If we were to think about spins that were staying stationary because there was some restriction to their diffusion, there would also be a phase change across this voxel. Spins on this side will be precessing faster than spins on this side. There would be a net phase change and also signal loss, but it would be organized based on the location in the x-axis here, assuming that these spins aren't moving freely across this voxel.
We then apply the 180° RF pulse, and we apply an equal diffusion gradient. Now, because of this 180° RF pulse, that diffusion gradient is effectively equal and opposite to the initial diffusion gradient. Now, remember, the spins in the restricted sample have stayed in the same location. They are experiencing an equal and opposite diffusion gradient, so that dephasing that was initially occurred because of the diffusion gradient is now equal and opposite and allows a rephasing. We will get larger signal from the spins that have restricted diffusion because they are rephasing with this second diffusion gradient, and we get a reaccumulation of phase back to baseline, and when we sample that signal with our EPI, we get a signal that is similar to if we didn't have the diffusion gradient on.
Now, the hydrogen spins that are freely diffusing, the gradient that they experience now is not equal and opposite because they have changed location within the voxel. They may have completely left this voxel, they may have completely left the slice and not contributing to signal. So, we get a rapid loss of signal here. Now, the loss of signal represents the degree of diffusion that's happening with those hydrogen spins. We know that when hydrogen spins move along a gradient, we get a change in phase. We actually get an exponential change in phase, and that change in phase is what's causing the signal loss in our DWI image.
Now, the strength of this diffusion gradient, as well as how long we have the diffusion gradient on, as well as the time between the diffusion gradient is what's known as our B value. The stronger the gradient, the longer we apply that gradient, and the greater the time between these gradients, the higher our B value. Our B value is something that we input into the MRI machine. The higher the B value, the more signal loss we get based on the diffusion within the tissues. And as I've said, that B value is related to the period of time that we have the gradient on. The longer we have that gradient on, the more time there is for diffusion to occur and signal loss to occur in a freely diffusing tissue. The steeper the gradient, the greater the amplitude of this gradient, the greater the differences in precessional frequencies based on their location, the more phase change. And lastly, the period of time between these two separate diffusion gradients. As we increase that period of time, there's more opportunity for these spins to move and not experience an equal and opposite gradient. Remember, the spins that have restriction are staying in the same location, so they experience the first gradient, they experience an equal and opposite second gradient, and allow for no net phase change to occur.
This B value is something we set. It's generally between 0 and 1,000, generally 500, 1,000. Protocols change between different hospitals, and there's no set B value that's going to give us the best image. So, that's chosen by us, and that B value, you can see, will manipulate the signal that's generated on this image.
Now, we've looked at this diffusion gradient in one plane. Now, we know that in anatomical structures, we don't have isotropic diffusion, we have anisotropic diffusion. So, we need to repeat this diffusion gradient in the x-plane, the y-plane, and the z-plane, at a minimum, the three planes of our Cartesian plane, to see if there's restricted diffusion along all of those planes.
Now, when we generate signal from this DWI image in all of our planes, what is actually giving us signal? Well, firstly, we are running a T2 star weighted sequence here, an EPI sequence. Secondly, our B value is going to determine how much signal loss we get. The higher our B value, because this is a negative exponent, as our B value increases, the more signal loss we get based on diffusion.
Now, what we want to calculate, as I've said before, is what is the actual or apparent diffusion coefficient within our image? How much diffusion is actually occurring in this tissue? As there is more diffusion, such as in our CSF here, we get more signal loss. Again, this is a negative exponent here. We repeat this in all three planes, and we're going to get three separate images. These are what's known as our source images. Now, in practice, we may repeat this six to 20 different times, but here we've got an example of three. We combine the signal of all three of these images that we've created to create our DWI image.
This DWI image represents diffusion within the sample, as well as the underlying T2 characteristics of the tissues that we are sampling. Here, we're running a T2 weighted sequence, a slightly longer TE. That longer TE is because we need space for these diffusion gradients to be applied. You can see it's still T2 weighted. Our gray matter is lighter than our white matter here. Now, if something is bright, what does that represent? Well, it could represent the inherent T2 characteristics of that tissue, or it could represent the restricted diffusion within that tissue. There's not been much movement. We haven't got signal loss based on the diffusion gradients that we've applied, and that diffusion or restricted diffusion is also a function of how strong our diffusion gradient has been that we've applied here. The degree of signal loss is also a function of that B value there.
So, now we've created two separate images: a b0 image and a DWI trace image, where we've combined our source images. This signal-averaged image is what's known as our DWI trace image, and we've seen that the signal in each voxel here is represented by T2 star. Signal loss is a function of how rapidly tissues have lost their transverse magnetization. Bright tissues here lose their transverse magnetization more slowly than the darker tissues. Our DWI sequence, each voxel here represents both the T2 waiting within the sample, as well as the inherent diffusion within the tissues that we're looking at here.
Now, what we can do is mathematically combine these two images to try and isolate what is the apparent diffusion coefficient within our image. You see this B image, each voxel here represents the T2 star of that tissue. In our DWI sequence here, you can see we use T2 star. The anatomy hasn't changed here. Our pulse sequence, the only thing that's changed is this part of our pulse sequence, adding in a diffusion gradient. Otherwise, nothing else has changed. Our B0 image actually is the same sequence, but we haven't applied a diffusion gradient. Our diffusion gradient is zero. Anything to the power of zero is one, so we can effectively cancel out the diffusion component here. We've got no signal loss based on our diffusion gradients in this image.
Now, what we can do between these two images, you see our b0 image is in fact just representing the T2 characteristics of the tissues again. Nothing has changed between these. The only thing that's changed is the strength of the diffusion gradient that we've applied. Now, when we look at our DWI image, the signal that we're generating here, it involves the T2 star component. We can take the signal of this DWI image and divide it by the signal that we got in our b0 image. As we take the signal from our b0 image and divide the numerical values of these signals, we can see that DWI image divided by our B0 image, B0 is the same as T2 star. We can cancel out the T2 contribution in our DWI image here. We can get rid of signal that's coming from T2, and what we're left with here is a calculated image, a mathematical calculation that is going to allow us to figure out what signal is coming from the diffusion within our sample.
Here you can see now we can reorganize this formula here. We know our DWI values from our trace image, we know our b0 values from our b0 image, and we know our B value. We selected the B value. The only variable that we don't know is our ADC, our apparent diffusion coefficient of each voxel within these images. If we isolate ADC in this formula, we can plug in the known values for each voxel within this image, and we can create what's known as an ADC map. You see, this is a mathematical calculation, and it fully relies on the DWI trace images as well as the b0 image. It's the mathematical calculation between these two that's giving us our ADC image.
Now, remember, in DWI, when there was free diffusion, we get a loss of signal. When there was restricted diffusion, we get higher signal. The opposite is true in our ADC maps. Free diffusion now is represented by bright signal values. Remember, as ADC increases, DWI decreases because of this negative exponent. The ADC values here give us some value to figure out how much diffusion is happening. The more diffusion there is, the higher the ADC value. Notice how we've lost though that contrast between our gray and our white matter here. The contrast here wasn't really due to the differences in diffusion. It was more due to the differences in the T2 weighting of the image. We have now negated that T2 weighting, and we really struggle to see that contrast in our ADC image because the contrast here that we get is based on the apparent diffusion that we have calculated within the sample here. The apparent diffusion is much more in the CSF in this example than it is in these cellular components of our image. That's how we go about creating our ADC map.
So, let's look at a few examples to understand the underlying processes as well as drive some of these points home. Now, in our first example, we can see that there is a lesion here that changes signal intensity based on the image that we're looking at. In our b0 image, that lesion is bright. It's got a high T2. In our DWI image, the lesion is also bright. Now, we know that in our DWI image, signal intensity either comes from the T2, the prolonged T2 of that tissue, or it comes from restricted diffusion. Now, we couldn't look at this DWI image in isolation and say that this is confidently restricted diffusion because we haven't yet canceled out the T2 contribution to this image. Once we cancel out the T2 contribution by combining these images here, we can see now that the lesion is dark on our ADC map. Remember, ADC represents the apparent diffusion coefficient. If the value is low, there is a low amount of diffusion. There is restricted diffusion within this lesion. So, our DWI image is bright because of the restricted diffusion and probably some contribution from T2 star here. So, this represents restricted diffusion.
Now, take a second example here, where we've got a bright b0 image and we've got brightness in our DWI, the same as the previous example, but now we've got brightness on our ADC map. Now, brightness on the ADC map represents free diffusion, unrestricted diffusion. If we look at our DWI image, we see that there's brightness on the DWI image that we know can either be restricted diffusion or T2 star contribution, prolonged T2 in that tissue. Here, we look at our B0 image, and it's very bright here. We know that this lesion has a prolonged T2. It has T2 brightness. So, this DWI image, we have to look at in combination with our ADC map because alone we could say that this is restricted diffusion. In fact, this is what's known as T2 shine-through. The lesion itself has a prolonged T2. It's got a prolonged free induction decay, and that's what's causing the brightness in our DWI image. It's not a restricted diffusion. When we look at our ADC map, we see that there is free diffusion within that image here. So, you can see how we need to look at our DWI in combination with our ADC and b0 image. Both of these examples, we couldn't look at the DWI image in isolation and confidently say this is either restricted or unrestricted diffusion.
Let's look at another example here. We've got a DWI image that is completely clear. There's no lesion on our DWI image. Now, this is a reminder. You can't just look at the DWI image and say if it's clear, I won't look at my ADC map. Many times, we look at the DWI image because our resolution is much better. The ADC map has really poor resolution. It's very pixelated. It's quite hard to spot the lesion. When you don't spot a lesion in DWI Imaging, you need to look at the ADC Imaging, and we look, and suddenly there's a lesion there. Now, what does this represent? Does it represent restricted diffusion? That would be strange for it to not show up on our DWI image. This is when we need to go back to the b0 image, and we see there is a low T2 value here.
Now, we are creating these images using echo planar Imaging. They're susceptible to many different artifacts, especially susceptibility artifact. We could get an artifact either in our b0 image or in our DWI image that doesn't represent the true underlying anatomy, and it doesn't represent the diffusion characteristics of that tissue. Because we use these images to create our ADC map, that artifact is going to come through in the mathematics. It's going to show up in our ADC, yet it doesn't truly represent anything to do with diffusion in this region, and we can figure that out by looking at our source images. So, here we've got an artifact that's being propagated through to our ADC image.
One last example to really drive this point home. Here we've got a dark lesion on our b0 image. Again, it could be an artifact, but we see it's also dark in the DWI image. All of our source images in our DWI average out have still created this dark region here. What we've got is a lesion that has a very short T2. It loses transverse signal very quickly. It's got a low T2 value. The low value in our DWI image here is not representing diffusion, free diffusion. The signal has been lost quickly because we're still running a T2, an echo planar image during this diffusion weighted image. Now, we don't know the diffusion characteristics within this lesion. Is it dark here because of the short T2 times, or is it dark because there's freely diffusing tissue in this region? And when we look at our ADC map, we get a confusing picture here. I've represented it by this kind of pixelated region here. The math here to create our ADC image doesn't give us information about the diffusion characteristics within this lesion, and this is what's known as T2 blackout.
So, hopefully, you have a better understanding of how we create these specific images and what the signal intensities on these images represent. I hope that you are convinced now that you can't look at a specific image in isolation without comparing it to the other images within the sequence. I hope that hasn't been confusing. I hope you have got more clarity when it comes to this type of Imaging. Many people get confused. And remember, if you are studying for a specific exam, I have a question bank linked below here. You can practice your skills that you've learned throughout these talks. You can figure out where you're strong and where you're weak and where you need to focus on prior to heading into the exams. I don't want you to be thinking, "Oh, I wonder if this is going to come up." I want you to be thinking, "When this comes up, I'm going to ace this." Go into your exam instead of with trepidation, with confidence. So, that's all for today's talk. I'll see you all in the next one. Goodbye, everybody.