false
Catalog
Practical Aspects of MRI (2023)
W8-CPH08-2023
W8-CPH08-2023
Back to course
[Please upgrade your browser to play this video content]
Video Transcription
Great, so thank you for the invitation to speak here. So MRI is a really amazing modality because it can produce images that have a wide range of contrasts. So in this talk, we'll go over the basic principles of T1, T2, and T2 star-weighted imaging, and also discuss some more specialized contrasts, including magnetization-prepared sequences and diffusion. So what do we mean by image contrast? Contrast is a relative term that describes the difference in signal intensity values between two tissues. And contrast is arguably one of the most important properties in MR, even more than SNR. So for example, if we're trying to identify a lesion, an image that has low SNR but high contrast can be more useful than an image that has high signal but poor contrast. From a physics point of view, contrast arises from differences in the transverse magnetization between different tissues. To create a signal in MR, we use an RF pulse that tips magnetization from the longitudinal, or Z-axis, into the transverse, or XY plane. And it's this XY component of the magnetization that we're able to measure and use to create an image. So to understand contrast, we need to understand what properties can lead to differences in this magnetization. So next, I'd like to move into MR relaxation properties, since these are what determine the contrast in most MRI scans. And we'll start with T1 relaxation. So T1 is a property of tissue that describes how quickly the longitudinal magnetization recovers after an RF pulse. And a longer T1 value means that it takes more time for the magnetization to recover. This T1 relaxation follows an exponential recovery curve, as shown by this equation, where T1 is the time constant of this function. It has units of seconds or milliseconds. And T1 is the time needed for MZ to reach 63% of its equilibrium value. Although we don't directly measure the longitudinal magnetization, T1 does affect the amount of magnetization that's available to be tipped into the XY plane to produce a signal. So here I'm showing a train of RF pulses, separated by some delay time called the TR. And if we use a long TR, this means that the magnetization has more time to recover in between pulses. This means that the next pulse will be able to tip a lot of magnetization into the XY plane and produce a large signal. In contrast, if we have a shorter waiting time, this means there's less time for MZ to recover in between pulses. And this will lead to a smaller signal. So T1 relaxation is mediated by a transfer of energy between spins and their surroundings. T1 values generally increase with magnetic field strength. And this table just shows some representative values. And you can see that fluids like CSF and blood have relatively long T1 values, whereas fat has a pretty short T1. The next relaxation property we'll talk about is T2. So T2 describes the loss of XY, or transverse magnetization, after an RF pulse. And T2 is caused by interactions between spins. Inside each voxel, there are many protons that are randomly moving. And as protons come closer together or farther apart, they will induce small local changes in the magnetic field strength. And this causes individual protons to precess or spin at slightly different frequencies. As a result, these spins will begin to fan out or dephase over time. And although we can't measure the individual spins, we are able to measure the vector sum of all spins inside of a voxel. And the spin dephasing causes this net magnetization to decrease over time, which is the origin of T2 relaxation. So T2 is described by an exponential decay function. T2 also has units of seconds or milliseconds. And it's equal to the time needed for the XY magnetization to decrease to 37% of its starting value. A tissue with a short T2 means that the signal loss happens more quickly. Most tissues have T2 values on the order of tens of milliseconds, up to a few hundred milliseconds. T2 is always shorter than T1. And fluids generally have very long T2 values. So T2 relaxation is caused by these random interactions between spins. And this is a type of signal loss that cannot be recovered. However, spin dephasing and signal loss can also be caused by macroscopic inhomogeneities in the magnetic field. And these are referred to as T2 prime effects. And this can be due to metal objects, iron levels in tissues, or differences in magnetic susceptibility between different types of tissues. Spin dephasing from these sources can be reversed with certain types of sequences like a spin echo. So T2 star refers to the combination of these T2 and T2 prime effects. T2 star is always shorter than T2. And in general, the signal loss with a spin echo sequence is described by T2, while for a gradient echo scan it's described by T2 star. And the last tissue property we'll cover is proton density. This refers to the number of protons inside of a voxel. The MR signal is proportional to proton density. But this is not often used as the main source of contrast weighting in MR. So to summarize, we've covered several different relaxation properties, including T1, T2, and T2 star. And proton density, which is not a relaxation property, but does affect signal levels. So next, I'm gonna show you how we can generate images that are weighted by these tissue properties. So most scans don't measure the actual T1 or T2 values. This is something called parametric mapping. Instead, we normally collect what are called weighted images and these enhance T1 or T2 differences between tissues. And we're going to start by focusing on a spin echo scan and I'll show you how we can generate images that have different T1 or T2 contrast weightings. So this is a pulse sequence diagram of a spin echo sequence. To keep things simple, let's assume that we're using a 90 degree excitation pulse and a 180 degree refocusing pulse. The two parameters that we can easily adjust at the scanner are the echo time and the repetition time. So if we wanna create a T2 weighted image, we wanna choose a long TE to maximize T2 contrast. And we also wanna minimize the effects of T1. And we can do this by using a long TR. With a long TR, MZ will have plenty of time to recover regardless of the tissue's T1 value. So here I'm showing a plot of the T2 relaxation as a function of the echo time for two different tissues. And you can see that we wanna use a relatively long TE to enhance differences in the signal due to T2 relaxation. A tissue that has a long T2 value will experience less signal decay, so it will appear bright on a T2 weighted image. And you can see that in these examples where CSF in the brain and synovial fluid in the knee both appear bright. We can also use a spin echo sequence to create a T1 weighted image. Here we want a short TE to minimize T2 contrast and also a short TR to maximize T1 contrast. So you can see that here on the plot of the T1 relaxation curve, we wanna choose a TR that's in this region, so relatively short TR to enhance differences in signal due to the T1 value. And tissues like fluid that have a long T1 will show up darker on a T1 weighted image, which you can see in the CSF in the brain. In general, for spin echo sequences, the acquisition time is shorter for a T1 weighted scan than a T2 weighted scan just because of the shorter TR. And finally, to create a proton density weighted image, we want to minimize T1 contrast using a long TR and also minimize T2 contrast with a short TE. So this type of contrast isn't very common for brain imaging, but it is used in other applications like MSK. So, so far we've focused on spin echo sequences, and I've shown you that just by adjusting the TE and the TR we can generate T1, T2, or proton density weighted images. In general, spin echo sequences have high SNR but relatively long scan times. Another class of sequences that we can use is called a gradient echo scan. And these tend to use shorter TRs, so they're generally faster than spin echo sequences, and they often use small flip angles less than 90 degrees. One important difference is that gradients are not able to recover those T2 prime related effects. And this means that the signal loss with a gradient echo scan is governed by T2 star instead of T2. And we're going to cover two flavors of gradient echo scans called unbalanced or spoiled and balanced. So the term spoiling means that the XY magnetization is eliminated before the next RF pulse, and this can be achieved by varying the RF phase from one pulse to the next. And I'll show you that by adjusting the flip angle, echo time, and TR, we can achieve images that have very different contrast weightings with T1, T2 star, or proton density weighting. So just like before, by using a shorter TR, we can increase the amount of T1 contrast. And by using a longer TE, we can increase the T2 star weighting. Another way to increase T1 weighting with this type of sequence is to use a larger flip angle, which I'll show you on the next slide, which is a plot of the transverse magnetization as a function of the flip angle for this type of spoiled GRE sequence. And you can see that the flip angle that maximizes signal levels is quite small, so it's around 10 to 15 degrees for these tissues. And this is called the Ernst angle. But if we're interested in maximizing the contrast or the signal difference between these tissues, we want to use a slightly larger flip angle. So here I'm showing an example of a T1-weighted, spoiled gradient echo image, which is using a short TR, a short TE, and a relatively large flip angle for this type of sequence. And using the same type of sequence, but by increasing the echo time and increasing the TR, we can generate a T2-star-weighted image. The second class of gradient echo scans that we'll discuss is called Balanced Steady-State Free Procession, or BSSFP. And this term balanced means that the gradient areas, positive and negative areas, are equal within one TR. And this sequence is very special because even though it's a gradient echo scan, the signal is not weighted by T2-star. And instead, it has this very interesting contrast proportional to T2 divided by T1. So BSSFP scans are widely used in cardiac imaging because blood appears bright and myocardium appears very dark because of this T2 over T1 contrast. So in summary, I've shown you that just by adjusting the TR and TE with a spin echo sequence, we can achieve images that have T1, T2, or proton density contrast. And we've also talked a little bit about gradient echo sequences where we can use a spoiled sequence for T1, T2-star, or proton density-weighted imaging, or a balanced SSFP sequence for T2 over T1 contrast. Very quickly, I'd like to talk about gadolinium contrast since this is another way that we can enhance the contrast of our images. So gadolinium shortens the T1 of nearby protons, and the contrast agent leaves the blood vessels and accumulates in the extracellular space. However, it can't enter inside of the cells or cross the blood-brain barrier in normal healthy conditions. So because gadolinium shortens T1, we normally collect images post-contrast using a T1-weighted sequence where areas that accumulate contrast will have increased signal. And this can be used to detect focal lesions, to image the heart or blood vessels, and to quantify tissue perfusion. Now, during the last few minutes, I'd like to talk about some more specialized image contrasts starting with magnetization-prepared sequences. So this refers to special RF pulses that can enhance or suppress a particular tissue property. And this special RF pulse comes before the image acquisition. And this includes things like inversions and saturations to enhance T1 contrast, T2 preparation modules, or fat suppression pulses. We're mainly gonna focus on inversion pulses. So an inversion is a 180-degree rotation that flips the magnetization from the positive to the negative z-axis. And this can be followed by a waiting period called the inversion time, or TI. And then this pulse is followed by our image acquisition using the T1-prepared magnetization. So inversion pulses are used in the MP-RAGE technique, which is widely used for brain imaging. And this consists of an inversion pulse followed by a fast T1-weighted gradient echo readout. And this type of sequence has excellent contrast between gray and white matter in the brain. We can also use inversion pulses to eliminate or null signal from certain tissues based on their T1 values. So by using a short inversion time, we can eliminate the signal from fat. And this is used in the STIR sequence. We can also select a longer inversion time to eliminate signal from fluids, which have longer T1 values. And this is done in the fluid attenuated technique, or FLARE. And this type of inversion preparation can be combined with either a T1 or a T2-weighted sequence. The next type of specialized contrast I'd like to talk about is fat and water. So one type of preparation pulse is called a fat suppression pulse. And this makes use of differences in resonance frequencies between water and fat. So this pulse will only tip and then spoil or eliminate the signal from fat while leaving the water spins untouched. And this can be useful because fat has a short T1 and a longer T2, so it often appears bright on T1-weighted or T2-weighted images. Another strategy, rather than suppressing fat, is to perform fat-water separation by collecting images at different echo times, where water and fat will accumulate relative phase differences. So we can time our TEs to collect images where water and fat are exactly in phase or out of phase. And then we can combine these images by either adding them together to get a water-only image or subtracting them to get a fat-only image where water has been suppressed. And this is called the Dixon technique. But here you can see examples of these in and out of phase images and the water and fat-only images that are derived from them. And then finally, in the last minute or two, I'd like to talk about diffusion-weighted imaging, which is an important part of many clinical protocols. So this is a type of scan that provides information about tissue microstructure and cellularity, where diffusion is this thermally-driven random motion of protons. So diffusion-weighted sequences are generally based on a spin-echo scan where we insert these two gradients before and after the 180 refocusing pulse. This first gradient causes spins to dephase. And if the spins are stationary, they'll be refocused by the second gradient. However, if spins are diffusing, they won't be completely refocused. And this leads to a loss of signal whenever we have unrestricted diffusion. The amount of diffusion weighting in an image is described by something called the B-value. And this is controlled by the strength and timing of those gradients. And during a typical protocol, you might collect images at different B-values, including a B0 image that has no diffusion weighting and is purely T2 contrast, and images at higher diffusion values. So here I'm showing an example of a B0 image. So this is T2 contrast. And then a B1000 image, which has higher diffusion weighting. And on the right is a parametric map of the diffusion coefficient that's calculated from multiple B-values. So in summary, in this section, we've gone over several types of inversion recovery sequences, including MP-RAGE, FLIR, and STR. We've also talked about fat suppression and fat water-separated imaging, and a little bit about diffusion contrast. So in conclusion, I hope that I've shown you that we can generate images that have very diverse and interesting contrasts, in many cases by adjusting just a few parameters at the scanner. I know that if you're kind of new to MR physics, just the sheer number of sequences can be overwhelming at first, but I hope I've given you some of the tools to start to make sense of these different contrasts. All right, thank you. My name's Walter. I'm an associate professor of radiology at Penn Medicine. So we have some learning objectives today. At the end of the presentation, you should be able to understand the concepts of spatial and temporal resolution in MRI. We'll define k-space and explain the difference between image space and k-space, and understand its relationship to spatial and temporal resolution. I also, I promise I'll tell you why you should care about that. And finally, we'll understand how spatial and temporal resolution affect image quality in magnetic resonance imaging. So the fundamental principle here is that when you take a picture with any kind of camera, and MRI, CT, nuclear imaging, these are all types of cameras, we ultimately are corrupting and distorting our true image of the object. So you can see that when we compare the image on the right to the image on the left, there's some blurring, and also a loss of color in the image. And these are the types of things that happen when we encode images of the body using magnetic resonance imaging systems. And so it's important to understand in what way the MRI camera is distorting or corrupting the image. So in order to understand that, we need to define some basic concepts about image space. Image space is an array of numbers that represents a signal intensity with X and Y coordinates, just like a picture on your phone. And so each cell of the image is called a picture element or volume element, voxel, and the signal intensity has a physical interpretation. If we look at this particular image, it has some number of columns, which is represented by some number of pixels in the X direction, and then some number of pixels in the Y direction. These are intrinsically related to properties such as the field of view and spatial resolution. The field of view is the image dimensions. They have units of distance. So for instance, a typical value for the field of view of a picture is in 240 to 400 millimeters or centimeters. And if we divide the field of view by the number of pixels, then we have the spatial resolution. So in this example, if we divide 240 by 100, then we're gonna have the, or 200, then we have the spatial resolution of 1.2 millimeters. So k-space. K-space is an array of numbers that represent the signal intensity in spatial frequency space with frequency coordinates kx and ky. Unlike image space, the pixel values in k-space don't have a simple interpretation. So we need to be able to understand what this means. When you look at the k-space on the right, you can see that you can represent the coordinate system with coordinates kx and ky instead of x and y. And that if we look at any particular pixel in k-space, we can represent it as a intensity I, and we're gonna try to learn what that is. So there's a relationship between image space and k-space, and that is the image reconstruction process. The MRI data is collected in k-space, and then through image reconstruction, we are making a picture that we recognize. And there are many algorithms to be able to do this. The one that is most commonly used is called the Fourier transform. So there are important definitions about k-space. The first is that if we traverse k-space and we go to the very edge of k-space, and the way we do that is using magnetic field gradients, then the maximum sampled point in k-space, we call k-max, the edge of k-space. It could be in the kx direction, then it would be kx-max or in the ky direction, ky-max. And the relationship between k-max and the spatial resolution are inverse. So if you go further out, then you're able to decrease delta x, which is increasing your spatial resolution. Similarly, if the sampling density in k-space affects the field of view. So again, these are inverse properties. As you increase delta k, you decrease the field of view, or decrease delta k, then you increase the field of view. So why should you care? Well, k-space sampling, the way we sample it, determines fundamentally the scan time, the breath hold duration, and all other important concepts that the patient has to be able to tolerate during the MRI exam. There are trade-offs between the spatial resolution, field of view, and temporal resolution. As they say, there's no free lunch. So when we try to compute the scan time in an MRI exam, most simply, the scan time is proportional to the time it takes to collect a column, delta t, times the number of columns, or phase encode directions. So typically, this delta t is in milliseconds. So if we're collecting, for instance, 100 phase encoding steps, or 100 columns, then the time it takes to collect this image is gonna be on the order of several hundred milliseconds. Depending on how we sample k-space, or if there were problems in the acquisition of k-space, image artifacts will arise. And so it's important to understand and recognize the relationship between aberrant signals in k-space and the way the image artifacts appear. So what is k-space? It's really the amount of frequency content in the image. So think about it like this. If you were to position your cursor directly in the center of k-space and measure the signal intensity, that would tell you precisely how much of a DC offset of the image, like your brightness setting on your viewer. As you move away from the center of k-space in one direction, in this case, we're going in the kx direction, you can see I'm able to represent the pixel here with a wave, or a spatial frequency. And it's a spatial frequency because it's a wave through space, not a wave through time. And you can see here it has some period, and it has some amplitude. And the further away I move in that direction, I'm able to make waves of a higher frequency. And so the wavelength is getting shorter. And you can see that depending on where I go in k-space, I'm able to make waves of different periods and different angular frequencies. So here you can see I'm moving at a diagonal distance away from the center of k-space. And you can see I'm getting these diagonally shaped waves of a particular period. So I think you understand that one pixel in k-space is representing a wave in image space. So now here comes the trick, that images are sums of these frequencies. So you may know that if you play notes on a piano and you combine them in just the right way, you get a very rich sound. It's the sum of all of these frequencies and harmonics and overtones. Similarly, when you represent k-space this way, we're telling the computer precisely how to add waves of different frequencies together to make a rich looking image. And it's really surprising and kind of incredible that when you add pictures of waves like this, you can create an image. And it doesn't take a lot of these types of representations before you get something that begins to look just like a person or a brain or a heart. So here's an example. What I'm gonna do in this example is I'm gonna play a movie. And we're gonna go through this process of encoding k-space using our scanner. And so I'm gonna be collecting data by collecting columns in k-space. And I'm gonna show you what a picture looks like when I collect only a subset of the total data. So we're gonna get a sense of what the information in k-space is actually telling us about the image. So here you can see I've collected now just a small fraction of k-space using my MRI scanner. So you can see these are represented by the vertical columns on the left where I've collected these signals. Now, if I'm adding the information about those waves that I showed you in the previous slide, you can see I can make a picture. But the picture doesn't quite make a lot of sense yet. You can begin to see that I'm starting to see the edges of whatever the object that's represented here is. And I think you can appreciate as you look closely at this, maybe this is a person. You can see here there is a hair on top of this person. You can see the eyes begin to be represented and you can see the tie. And maybe there's a intricate pattern on the tie. And so as I am collecting the edge of k-space, I'm collecting the high spatial frequency components of the image. All those waves that had very short period wavelengths, high frequencies. And when I add them together, what I get is information about the edges of the object. So as I continue to collect data in k-space, you can see I'm collecting more low frequency information as I'm moving through the center of k-space where those waves are getting lower in period. And now I think you can appreciate that it truly is a picture of a person. It is a tie. You can see the lapel. There's a pen on the lapel, the hair, the eyes. And here I'm moving through the center of the image and now I'm representing a person here. And so there are a couple of key points. I've only collected half of k-space, but yet I'm able to make almost a clear picture. So this tells you something about k-space that k-space has this symmetry involved where if I collect part of k-space, I know that the other half of k-space also has similar information in it. And so that's one way we can use to accelerate our scan by not repeating the process. But you can see as I continue to collect data, the image improves in signal to noise ratio and some of those artifacts that you began to see are beginning to disappear. So, okay, so we're not encoding real people, or pictures of people. We're encoding pictures of their organs. So let's take a look at what happens when we do that. Here's an example of the heart. So I'm gonna show you what it looks like when we take a picture of the short axis view in the heart. And again, we've begun to collect the edges of K-space first, and I think you can appreciate here, the anterior chest wall is beginning to show up on this image. And as I collect a little more information, you can begin to see vessels in the lungs. You can begin to see the myocardium, represented by the epicardial surface, and the endocardium begins to have that donut shape. And then as I collect more data, and pass through the center of K-space, the full image appears. So, it takes some time to collect K-space. The scanner takes some time to move from left to right, scanning K-space, and then forming an image. And so the time it takes to do that process is gonna fundamentally be affected by motion that occurs while that process is occurring. And so, for instance, if you were to take a picture with your camera, the similar parameter on your camera is kind of the shutter speed, or the time that the lens is open, capturing light. And so if you're trying to take a picture of something that's moving quickly, like a tennis ball, or a person playing soccer, a bunch of birds in flight, it's important to have that shutter speed as fast as possible. The time that the lens is open needs to be very short. Otherwise, you're gonna get this blurring. You're not able to represent these motions that are happening very quickly. But one thing that you also begin to see is that as you reduce the shutter speed, while you're able to resolve the object, the object's signal to noise begins to drop. So there's some important timescales of physiologic motion. For instance, if you consider just the chest, we're in the 10 to 50 millisecond range. We're seeing myocardial motion, pulsatile flow in vessels. MRI takes on the order of 300 to 500 milliseconds. So if we just make a picture of those things, we're gonna get a blurry image, like those birds in flight. And other motions, like respiratory motion, have a longer period. So it may be that this is adequate for being able to capture respiratory motion. So consider, again, the scan time is proportional to the time it takes to collect a column and the number of phase encoding steps that I'm gonna use. So in this example, it takes about 2 1⁄2 milliseconds to collect a single column. And I'm gonna use 144 steps, or 144 pixels, to represent k-space. So the minimum time to collect one image is 360 milliseconds. You can see that's already too long to be able to capture a picture of the moving heart. So what we want is to obtain a desired temporal resolution of 40 milliseconds to see the heart moving clearly. But unfortunately, we can't do it with this approach. So there are methods that we need in order to encode k-space faster. The first thing you can do is just simply collect data that are lower resolution with less k-space lines. So here you can see that I've collected the data using a snapshot of k-space. And the time it takes to collect that k-space data is gonna be reflecting motion degradation in these images. So you can see in the image on the left, because the time to collect the image was 6 1⁄2 seconds, you can begin to see motion degradation related to respiratory motion in the lungs. And you can see these artifacts, they look like ghost artifacts appearing near the liver and diaphragm. But if I make the time period to collect this data shorter, then these artifacts go away and I can clearly see the image. So there are some approaches that we can use to reduce scan time and preserve spatial resolution. One approach is to just acquire a subset of k-space. So if, for instance, I collect the center of k-space and make a picture, then I'm gonna make a lower resolution image, but it will have a full field of view. Alternatively, I could skip data. So I could skip every other line of k-space. So you can see here, because that's increasing the delta k-spacing in k-space, I'm getting a fold-over artifact or aliasing because the image field of view is smaller than the object that I'm trying to image. So there are a number of different approaches that you may have heard of to be able to solve both of these problems. That is, create missing k-space data from other ways than sampling it directly. So one way of doing that is using parallel imaging, the multiple sensitivities of different radiofrequency coils, or compressive sensing, which is a reconstruction method, or machine learning. These are all methods that we can use to recover high spatial resolution information at a full field of view. So you can see that frame rate and temporal resolution are two sides of the scan. So when you look at, for instance, at this picture, it's showing a picture of the fetus in utero, and you can see here that the image has very high spatial resolution, but you can see that I'm missing information about how the fetus is moving, that it appears that the fetus is just stepping from one location to another. So there's some missing frames, the frame rate is low. Here we have, on the other hand, the frame rate is high, but the spatial resolution is low, and you can see the fetal heart beating. And so the important concept here is that there are trade-offs between the frame rate and the spatial resolution and the signal to noise of the picture. Cine Magnetic Resonance Imaging is an approach that we can use to visualize the heart beating, and it makes these quite exquisite-looking pictures. And one of the approaches that we use to resolve that issue of scanning in a time so that we can capture that information is to be able to use this segmentation-based approach. So what we do is we collect only a portion of k-space each heart beat. Those movies that I showed you on the previous frame, they didn't represent a single heart beat, they represented data that was collected over 10 to 15 heart beats, where we collected only a subset of k-space data each heart beat, and then combined it together to make a single picture. So there's some trade-offs between temporal resolution and scan time. So if you look here at the image on the left, you can see the temporal resolution is 30 milliseconds, but it took nine heart beats to collect this data. The image on the right was collected in just a single heart beat, but it took a temporal resolution of 150 milliseconds to collect. So there's gonna be a loss of temporal fidelity in the image on the right compared to the image on the left, though if your patient can't hold their breath, then maybe you would consider somewhere in the middle. There are also trade-offs between spatial resolution, signal-to-noise ratio, and scan time. If you look at these pictures, you can see that as I improve the spatial resolution, the voxel size is decreasing, the time it takes to collect the data in terms of number of heart beats goes up. But when you look at these images, you might say, okay, it's not really clear that the image on the right, which has the highest spatial resolution, took the longest time to collect, is necessarily the best-looking image. You might say the third one from the left might be. And you might say that because as you look at the one on the right, the signal-to-noise ratio is decreasing, and that's because the signal-to-noise ratio is proportional to the size of our voxel. So there are these important trade-offs to consider. You don't wanna just make a decision to increase the spatial resolution without consideration of what is happening to the signal-to-noise ratio of the picture. So, to summarize, MRI data is collected in case-based. Images are reconstructed from the case-based data. Case-based sampling requirements determine the field of view, temporal resolution, spatial resolution, total acquisition time. And you need to understand these trade-offs between resolution and acquisition time in order to get diagnostic quality pictures in your patients. So thank you, and I really appreciate your time, and Nicole, for the invitation to join. So I'll be covering managing common artifacts in MRI. Okay, and I'd like to start off, of course, with acknowledgments, acknowledging a number of my colleagues at Stanford, as well as a few of my colleagues at different institutions who have contributed examples to this presentation. Okay, so if we think at a bit of a higher level to start off with, we really have sort of two time frames in MRI where we can do work to manage our artifacts. So the first one is, the first time frame is before patient scanning. And so this involves setting up sort of our baseline protocols. So this includes choice of field strength, what coil we're gonna use, our choice of sequence. And then also, we also have a baseline parameter that we set up, right? The first thing that we pull up when we're gonna scan a patient. Okay, so then there's a second time frame in which we can manage our artifacts. And that time frame is during the patient scan itself. And so I think about that as the time frame from when the patients walk in the imaging suite to the time then when we're done with the scan. And so then our options in the second time frame include how we set the patient up, their positioning, some pre-scan and calibration functions we do. So there's certain things that are patient-specific, certain measurements that we need to do on a patient-by-patient basis. And then finally, of course, we do have an option to modify the scan parameters that we set up, those baseline parameters. So as I go through the artifacts in this talk, I'll definitely be referencing artifact management aspects from sort of both these time frames. But I hope to convey, really kind of focus on the second one a little bit more, what we can do to manage these artifacts, what our options are when we're at the scanner or when we get sent an image that has an artifact and we need to manage it while the patient is on the table. All right, so the first artifact that I'll discuss is low SNR. And so I have two examples of images that were acquired and then when we looked at them, they had low SNR. Okay, so low SNR in this case, I'm really talking about low SNR across sort of a large region of the image or the whole image itself. And we just saw a very lovely description of the trade-offs and this is a fundamental trade-off that we learn and both physicists and clinicians learn of MRI. And it's the trade-off between SNR, resolution and scan time. But this isn't the first thing I want you to think about when you see a low SNR image. So in the clinic, when you see a low SNR image, particularly if it's across a large region of the image or across the image itself, check your coil. So yeah, check your coil. Okay, so oftentimes, the coil, in fact, might not be plugged in, right? In day-to-day going in and out different patients, sometimes we just forget to plug the coil in. And then also, as our coils become more and more sophisticated with more and more elements, we often have to choose a subset of the MRI coils. So and that's exactly what happened in the two examples that I have on the slide, is that the wrong set of coil elements was chosen. And so if you look down at the bottom here, or I guess I can't see the pointer, but on the left-hand side, where the SNR was lower in the superior region, we needed to choose the full set of elements, and only the upper ones had been chosen. On the other side, obviously, this is a head and neck scan, but the first image was chosen with the lower limb elements chosen. Okay, so, but also always keep this fundamental relationship in mind as well. So if the coils are connected, and you still are seeing low SNR in your image, then you can think about this trade-off that we had just, we heard about in the previous talk. Again, SNR is promotional to the voxel size and the square root of the image time. And one thing to note here is that you're gonna get a little bit more bang for your buck if you increase the voxel size, if you can do that, since it's directly proportional to that, but in terms of the image time, it's proportional to the square root. Okay, so low SNR, the first thing to think about is really check your coil, check your coil elements that are chosen, but also keep that fundamental relationship in mind. Okay, so now we'll move on to a FAT artifact. So as we heard already in the previous contrast talk, FAT is problematic in MRI. It has a long T2, a short T1, and so it's something that we often want to remove from our MRI images. Now, there's a few different types of FAT artifacts, chemical shift artifact, we can have out-of-phase artifacts that have to do with FAT, but here what I wanna focus on is when we have FAT suppression failure. So you see and you run an image or it gets sent to you, and you see at the top that you have just a failure of FAT suppression in one of the regions. Okay, so here in terms of the physics, what I want to go back and review, or what we should focus on is this difference in frequency between water and FAT. Of course, we use the frequency of water to see water molecules in MRI. So FAT has a different resonant frequency than water, which is very useful. So three of our four primary methods for suppressing or removing FAT from our images in MRI take advantage of this difference in frequency. So we're using a chemSAT pulse, we suppress signal at the FAT frequency. Water excitation, we do the opposite, we're exciting just the water frequency. And then Dixon or multi-echo methods take advantage of the difference in frequency and the phase that's accrued, the difference in phase that's accrued because of the difference in frequency. So the inversion pulse, using a short T1 inversion pulse to suppress FAT is the only FAT suppression method that isn't somehow dependent on this difference in frequency. Okay, so of course, in vivo, our spectrum does not look as nice as it does when we present it theoretically. And so the in vivo spectrum, I have just an example of a spectrum that was taken in vivo. And so one thing that you can see is that we will get broadening of these peaks due to background V0 inhomogeneity. So, and it's important to remember that this environment is going to vary patient to patient. The V0 inhomogeneity is affected by the patient when we put in the scanner, and then it varies based on what the composition of the patient is. Okay, so back to managing this artifact. What I'd like you to do is if you see a FAT suppression failure, this is one thing to keep in mind, maybe one of the first things we should think about. So one thing to do is first to confirm that your center frequency is on the water peak. Then the second thing to keep in mind is to think about the shims. So shim volumes, the specific, the shim volumes are utilized to correct V0 inhomogeneity. So we really should use the shims in most of our acquisitions. And again, these are really, oftentimes they're pre-scanned, they're pre-scanned measurements, there's a way to check these. And remember, they're patient-specific. So I just have a couple examples here. So this is an example where the center frequency was on FAT, not on water. There's silicon implants here, and there's FAT, there's not a lot of water in this image. And so when the scanner found the center frequency, it found FAT instead of water. So if you see an image like this, one thing you can go back and check, did my scanner actually find the center frequency correctly? And this is an example of varying your shim. So in the breast on the left, it was a smaller shim volume. And you can see that there's still a decent amount of FAT showing up near the chest wall. By increasing the shim volume, having it cover a larger area of the breast, we see an improved shim. Okay, so in terms of a FAT suppression failure, I think a good thing to keep in mind is initially think about checking that spectrum. Think about checking your center frequency and double-checking that your shims are applied and that they're big enough, they're covering a region of interest. Okay, so we're gonna move on now to aliasing. So aliasing, again, is a very common artifact in MRI. I have three different examples across the top of the slide. On the left-hand side, we see aliasing actually in two directions. The arrows at the bottom are pointing to in-plane aliasing of the arms into the tissue or area of interest. And then we also see through-plane aliasing of the hips. In the center, maybe a bit more straightforward, we just see the hips on the side aliasing in. And then in the breast, we have a through-plane aliasing. You can see a slight outline from the breast that's wrapping into the image that we're interested in looking at. So aliasing artifact generally looks like we saw it in the previous talk. It looks like this fold-over, this replication of your anatomy in your image to a different, the severity to a different degree. It can occur both in-plane and through-plane. One thing to note is that will occur in the phase encode direction, not in the frequency encode direction. Okay, so again, if we're sitting at the scanner, what are some things that we can do or what are our options to manage this? So the first thing that we wanna do is increase the field of view or increase the number of slices. So this might seem pretty obvious, okay? Where we have wrap-over, especially maybe the case in the middle, we think that that makes sense, right? Let's just increase our field of view and then the artifact won't wrap over anymore. But one thing that's interesting to think about, though, is that it happens in the phase encode direction, but look at the frequency encode direction the other way. We're not getting aliasing in that direction. And so sometimes you'll hear to manage aliasing, flip the frequency and phase encode, and this is an option, but also a little bit more prospectively, remember to align your phase encode with the short axis of your anatomy because that's where you need the field of view to be outside your anatomy. Okay, so I'll talk a little bit about motion. All right, and as we just heard in the previous talk, motion is just very impactful in MRI. MRI is sort of a slow imaging acquisition and there's a number of amazing methods being developed to try to compensate for motion, remove motion artifacts. But again, let's think about, we're sitting at the scanner, what are our options day to day to deal with motion artifacts? So motion artifact manifests as blurring or ghosting of the tissue. As I just said, it's, I think, always gonna be an issue in MR. MR is a slower acquisition. And as was mentioned earlier, one of the assumptions we make in MRI when we apply our imaging gradients is that everything's gonna be in the same place. And when we reconstruct the data, we can combine it all and assume that the phase we have is just from those gradients that we have applied. But of course, when things start to move around, that no longer is the case. Okay, so if we think about this time frame, this second time frame, here for motion, one thing to think about, and of course, this is prospective, as we're positioning the patient, is patient comfort and length of exam time really does make a difference. And I know that certainly, in many environments today, we are trying to scan as quickly as possible and get in as many patients as we possibly can. But patient comfort, positioning, padding, and then also keeping the length of exam, this is something that can really affect the degree of motion that you're gonna have to deal with in an MRI scan. Okay, aside from that, aside from preparing the patient, do we have any other options? So of course, we have respiratory and cardiac triggering. We can also do breath holds. You'll see a breath hold on the left, free breathing acquisition in the center. So one other thing to keep in your back pocket to keep in mind is, again, this directionality of some of our artifacts. And so, since motion also propagates in the phase encode directions, we do have the ability, again, here, to switch the frequency and phase. So this isn't necessarily going to eliminate the artifact. We're still gonna have a motion artifact. But perhaps it would be useful to moving the artifact away from an area that you're interested in looking at. And so you can see, in the example in the center, and then on the right, we've switched the direction of the frequency and phase encode, and the artifact manifests differently. And you're able to, some of the tissue that's obscured with the center image, we can now see clearly. So just something to keep in mind, this directionality dependence. One other option for managing motion artifacts when we're at the scanner, in both of these cases, we can see kind of a light ghosting across the image. One other thing we can do to remove motion is if that motion is coming from an area of the anatomy we're not concerned about, or we're not trying to get diagnostic information from, we do have an option to apply a suppression band. So in both these cases, we could apply a suppression band to get rid of the signal from that area, and then it wouldn't be causing ghosting across the image. Okay, so in terms of motion, there's always going to be sort of a challenge in MRI. Remember, don't forget, I guess, about the importance of patient setup, patient comfort, length of exam. And then there's a couple things we can do during the scan to help to mitigate, or at least control, the motion artifact. Okay, I'll talk a little bit now about parallel imaging. So parallel imaging is a very powerful method that we have in MRI because it allows us to accelerate. We accelerate by skipping lines in case space, and then we take advantage of our coil sensitivity profiles to reconstruct that missing data. The example on the right is a breast exam where we use a high degree of parallel imaging. It allows us to make what would be a 24-minute exam with no parallel imaging a four-minute exam. All right, so one parallel imaging artifact is essentially aliasing. It's due to the undersampling that we're doing. And in a way, we can think also about the aliasing artifact that we already talked about. And in this case, so we have the aliasing artifact we talked about previously. Our field of view isn't big enough. With parallel imaging, we're forcing undersampling. So we're forcing there to be aliasing in our image. So what happens is sometimes if there's motion, or for some reason, the reconstruction algorithm, it just isn't able to sort of fully reconstruct the image, we will have an artifact. And the artifact will, again, it's aliasing. It's gonna look similar to other aliasing artifacts that we have. The aliasing artifact, again, is present in the phase encode direction. And so the options to manage it, we can think again back to the other standard aliasing that we talked about earlier. Adjust the field of view. So adjusting the field of view can just move this artifact now. We're undersampling in a slightly different way, and so the artifact will be moved. There's also an option to reduce the parallel imaging factor if you think you've just sort of pushed the parallel imaging too much. There's definitely an option to just reduce the parallel imaging factor. And we can see on the right-hand side, by changing the field of view, we were able to remove that artifact. Okay, and then the final artifact that I'd like to briefly discuss are artifacts due to metal. Okay, so metal is problematic in MRI because of the large degree of off-resonance that it causes. So a large susceptibility difference. The artifacts, there are just a number of artifacts that can result in signal loss, signal pile-up, distortion. Fat suppression is difficult around metal implants. And just to give you an idea, when we were talking about fat, the difference between fat and water, the difference between fat and water is 440 hertz. The spectrum that I was showing earlier is 2,000 hertz, so two kilohertz. Around metal, it's tens of kilohertz, so it's 10 times the amount of off-resonance or susceptibility that we have between our fat and our water. So for metal, a lot of what we do is actually before we scan. And there's a lot of choices we sort of have to make in terms of developing our protocols for imaging metal that are gonna be consistent across all patients. And so these include the necessity to scan at a lower field strength. The susceptibility effects are not as strong at the lower field strength. We scan with spin echo versus gradient echo. And you can see an example, the spin echo in the center and the gradient echo on the right. And so the spin echo corrects for a lot of the defacing. With metal, this is where we want to use an inversion fat, a saturation method. And this is because the inversion pulse, as we talked about earlier, aren't dependent on that frequency difference. There's other trade-offs, though, with the inversion pulse. They introduce T1, or sorry, yeah, T1 contrast. So there's other trade-offs, but this is what we want to use with metal. Finally, at the bottom, we have a number of multispectral imaging sequences that have been developed specifically to image around metal. And also, imaging at a high bandwidth and at thinner slices is also something that we often do at the outset. Okay, so do we have any options with metal when we're at the scanner and we're looking at the artifact and it's still in the way of what we need to see diagnostically? So here, we can also keep in mind, again, the directionality. So the directionality with the metal artifacts propagates in the frequency direction. And so again here, we can hopefully take advantage or possibly take advantage of the directionality to scan in a different plane and maybe move that artifact away from what we're interested in. Okay, so today, we have covered six common artifacts in MRI. We have talked about some of the prospective things we do, but really hopefully highlighted some of the options that we have. Again, if you're at the scanner, the patient's on the table, and what are options to sort of manage some of these artifacts. And then some of the take-home points, so in terms of motion, and if you see that low SNR, really think outside of getting the image. Think about the patient's setup, and if you see low SNR, just automatically check your coil, check the subset of elements that you've chosen. So I pre-scanned calibration here. This is fat suppression. What I want you to think about is fat suppression, patient-specific information that we need. So again, if you see a fat sat failure, I think a good thing to keep in mind and to double-check, keep that spectrum in mind, check your center frequency, and check your shims. We did see some options for changing our imaging parameters, including increasing the field of view, using saturation bands. And then the last thing is the ability to manipulate the artifact. Not making it go away, but one of the things that's kind of striking as you go through all these different examples is the directionality dependence of the artifacts. And again, keep that in mind, and possibly something that you can use in a pinch if you can't see what you're interested in. So thank you so much for your attention.
Video Summary
In a comprehensive talk on MRI imaging, the speaker discusses the significance of MRI's ability to produce images with a wide range of contrasts and delves into the basics of T1, T2, and T2 star-weighted imaging, as well as more specialized contrasts like magnetization-prepared sequences and diffusion. The presentation explains how image contrast is crucial for identifying lesions, where high contrast can sometimes compensate for low SNR. The reasons behind contrast variations are explored, focusing on the magnetization properties that affect these differences. The talk then details the various MRI relaxation properties—T1, T2, T2 star—and proton density, discussing how each influences imaging and how these principles are applied in different MRI sequences like spin echo and gradient echo scans. The utility of gadolinium contrast for enhancing image visibility is also covered. Further, specialized image contrasts such as inversion recovery sequences and diffusion-weighted imaging are explained. Overall, the talk offers insights into understanding and manipulating various imaging parameters to achieve desired contrasts and tackle different scanning challenges, emphasizing practical solutions to common MRI artifacts.
Keywords
MRI imaging
image contrast
T1-weighted
T2-weighted
diffusion-weighted
magnetization-prepared
gadolinium contrast
spin echo
gradient echo
RSNA.org
|
RSNA EdCentral
|
CME Repository
|
CME Gateway
Copyright © 2025 Radiological Society of North America
Terms of Use
|
Privacy Policy
|
Cookie Policy
×
Please select your language
1
English