MRI imaging in neuroscience Dr. Thom Oostendorp Lab class: 2 hrs 1 Introduction In tomographic imaging techniques, such as MRI, a certain tissue property within a slice is imaged. For each voxel (volume element) of the slice the measured valued is stored in the computer. The value is translated to the gray value of a pixel on the screen. In the interpretation of the image displayed on a computer screen two aspects are important: What image property is imaged? How are the measured property values translated to gray values on the screen? In this lab class both aspects are studies. First we will discuss how in general measured values stored in the computer are translated to an image. Next, we will see that in MRI not a single property is measured, but a combination of three properties. We will see that the relative weight of these properties depends on the device settings. Subsequently, it will be demonstrated how diffusion on a microscopic scale can be imaged using MR imaging, and how that can help clinical diagnosis. Finally, we will see how fmri images are constructed from a sequence of images recorded while a subject is alternatingly performing or nor performing a mental task. 2 Digital images: window level and window width Digital images Like pictures from a digital camera CT, PET and MR images actually consist of an array of number in computer memory. In order to view the image these numbers have to be converted to pixels of a certain color and brightness on a screen. For a picture recorded by the digital camera the numbers represent measured color and brightness, so it is only natural to show the recorded color and brightness value on the screen. In medical images there is no such natural choice: for CT for instance the numbers represent the property extinction coefficient for X-rays (µ), and the way we convert this value to an image is essentially arbitrary. It is customary to image one dimensional properties such as µ to gray values. Sometimes it is customary to image small values darker and large values lighter, as in CT, and sometimes the other way round (as in PET: check this in Medische Fysica). Note that there is no natural reason to choose the one or the other. Windowing Now let s use an MR image to look in some more detail at the way how signal intensities are translated to gray values. Go to the folder Practicum Programma s, and from there to the folder DIF2. Start the program MriSim. Within this program open the file mrt1.mri. A transvers MR-image of the head will appear (the same one as in figure 10-17 of Medische Fysica).
1. The tissue property that is imaged in an MRI-pixel is the signal strength of the radio waves emitted by the tissue in reaction to the excitation pulse. Are in this image small signal values displayed in a dark or in a light color? Hint: base your conclusion on the way the air around the head is displayed. If you move the mouse over the image (without clicking!) you will see at the left side of the status bar below the screen the signal strength of the pixel on which the mouse cursor is positioned. 2. What is (more or less) the signal strength in the brain, the subcutaneous fat, the tumor, the bone and the cerebro-spinal fluid in the ventricles? Check whether a brighter pixel corresponds to larger signal strength. The way signal strength is converted to gray values is displayed in figure 1. An obvious choice for the way to convert signal strength to gray value would be to make the pixels with zero signal strength black, those with the largest signal strength white, and those in between a brightness that corresponds to their signal strength. In the example of figure 1 this would mean that signal strength 0 would be mapped to black and signal strength 2000 to white. Figure 1. Conversion of signal values to gray values. If, however, all interesting regions have signal strengths between, for example, 250 and 1500, the result would be that they are all imaged in similar gray tones. It would be much better to use the available contrast in the region of interest, at the cost of the less interesting regions with low or high signal strength. This is achieved by windowing. In figure 1 all signal values below 250 are converted to black, and all signal values above 1500 to white. The range between 250 and 1500, in which signal values are translated to gray values of increasing brightness, is called the window. The width and position of the window can be changed, in order to optimize contrast in interesting structures. In figure you can see that the window width is the difference in signal strength between the largest signal value that is displayed in black and the smallest signal
value that is displayed in white. In this example that is 1250. The window level is the signal value exactly halfway the window, i.e. the signal strength that is displayed at 50% gray (in this case 875). Click on button in the toolbar. A dialog window will appear by which you can change the level en width of the window (figure 2). The window is indicated by the white region in the graph. Furthermore a histogram is plotted that indicates how often certain signal values occur in the image. You can change the level and width using the sliders, but also by dragging the center of the double arrow (for the level) or the arrow tips (for the width). 3. Change Level and Width, and observe what happens. Explain what you see. As you can see, manipulating the window can increase contrast between certain signal values Figuur 2. Window level en width (and hence certain structures), at the expense, of course, of the contrast in other areas. In clinical practice radiologists continues adept window width and level to get an optimal view of the detail they observe. To do this they don t drag sliders, but drag the mouse over the image. The convention is that left-right dragging changes window width, and top-bottom dragging the window level. This also works in MriSim; try it. Note that the values of the window level and width are also cited in the status bar (at the right: abbreviated to wl en ww). 4. Try to find a setting for van level and width in which the edge of the lesion is completely white, and the brain is completely black. Examine the curve defining the conversion form signal value to gray value, and explain why in this setting the edge of the lesion is white and the brain is black. Is it possible to make the edge white and the subcutaneous fat black? 5. Find a setting in which the edge is white, the brain gray and the ventricles black. 3 T2-relaxation an echo time In the previous section we studied how an image can be adapted to our desire. The underlying signal values are fixed, of course; the only thing that was manipulated was the conversion from signal values to gray values. In this and the next sections we will see that the signal values themselves reflect a combination of three tissue properties, and that the relative weight of each property in the combination is determined by the settings of the MR scanner. As a consequence an MR scanner can produce quite different images of the same structures. Conversely, it is important to know what the settings were in order to correctly interpret an MR image.
As mentioned earlier the signal values in MRI represent the signal strength of the radio waves emitted by the voxels. As only hydrogen nuclei are excited by the excitation pulse, the signal strength of a voxel is among other things determined by the hydrogen density. In reaction to the excitation pulse the excited nuclei precede around the direction of the magnetic field. One by one the excited nuclei return to the preferred direction parallel to the field, and doing so emit a radio wave (see figure 10-14 in Medische Fysica). After some time the spins that have not yet return get out of phase, and so do the radio waves that are emitted. This causes the signal to weaken. Of course the signal also weakens as nuclei are returning to ground state, but this effect takes much longer than the reduction caused by dephasing. As a consequence the strength of the signal emitted by a voxels depends on the echo time TE: the time interval between excitation and measurement. We will now study the way the signal weakens after excitation. In reality this can only be done by making a series of MR-scans with different TE values, but the program MriSim can simulate an MR scanner. Open the file meningioma.mri. This file does not contain an actual MR recording, but the data the program needs to simulate the MR scanner. When you open the file the program will ask for TE Figure 3. Setting TE en TR. an TR values (figure 3). Set TR at 2100 ms (we will study TR in the next section), and set TE to 10 ms. After you clicked OK the corresponding MR image for those settings will appear. Tip: do not maximize this image, you will have to observe many such windows together. Next, click button in the toolbar. The dialog window of figure 3 will reappear. This time, choose TE=50 ms. A second MR image will appear. In the title bar of these images you can see that the first one was recorded (actually: simulated) at TE=10 ms, and the second one at TE=50 ms. 6. When you move the mouse over the image the signal value of the pixel the mouse cursor is pointing to is displayed in the status bar of that image. Simultaneously, in the status bar of the other image the signal value of the same pixel in that image is indicated. Note that for all pixels the signal value is less at TE=50 than at TR=10 ms. Why is that (see figure 4)? 7. Does the signal value decrease with the same rate for all tissues? 8. As the time between excitation pulse and measurement increases, the MR-signal decreases. Because of this, one might expect that the second image is darker than the first one, but that is not the case. Why? Hint: note the values of window level and width in the status bars.
Figure 4. As a result of dephasing of the spinning nuclei after the excitation pulse the intensity of the emitted radio waves decreases. TE is the moment at which the signal is measured. T2 is the time it takes for the signal to decrease to 37% of its initial value. At question 8 you have observed that the program automatically chooses appropriate values for window level and width for each image, and that these values are different for each image. You can enforce that all images use the level and width of the active image (the one on which you have clicked most recently) by clicking the button in the toolbar. If you click the button again, each image will again use its own level and width. Use this setting for the rest of this lab class. At question 7 you (should) have observed that the signal strength does not decrease with the same rate for all tissues. We will now study this in more detail. 9. Create a series of MR-scans with TE-values 10, 20, 50, 100, 200, 500, 1000 en 2000 ms. Use for TR always the value 2100. 10. Explain why for high TE values the brain has disappeared, while the ventricles are still visible. 11. Use Excel to plot in one graph the relation between TE and signal strength for a pixel in the ventricles, one in the brain, and one in the edema around the tumor. Choose for an XY-scatter plot, otherwise the x-axis will not be correct. It is advisable to add gridlines, so you can read the values in the graph. to do that right-click on the x- or y-axis and choose Add Major Gridlines and Add Minor Gridlines. Apparently the rate by which the spins dephase is a tissue property. The relation between signal intensity and time is a negative exponential function: e / (1) with I the intensity of the emitted signal, I0 the intensity immediately after the excitation pulse, and T2 a tissue property that indicates how fast the signal decreases for that tissue. You will recognize this type of relation from for instance the extinction of radiation by matter (Lambert-Beer s law) of the decrease of activity of a radioactive source.
The T2 value can be determined from the plot. From equation (1) follows that, when the signal is measured at t=t2 after the excitation pulse, the signal intensity e / e. Consequently, T2 is the moment at which the signal has decreased to 1/e (37%) of its initial value (also see figure 4). 12. Estimate from your plot the T2 relaxation time for the cerebro-spinal fluid, the brain and the edema. Compare the values you found with those in table 10-2 of Medische Fysica. Note: the actual value of T2 also depends on the magnetic field strength of the scanner, so you should not expect a strong quantitative correspondence. 13. Estimate from your plot the TE-value for which the contrast between edema and brain is optimal. Do an MR-scan at this value, and check whether the edema can indeed be easily distinguished form the surrounding tissue. 14. What is wiser: make an MR-scan using random settings for TE an use window level and width afterwards to optimize the contrast between edema and surrounding tissue, or first choose an appropriate value for TE? It is possible to watch the decrease in signal strength (the relaxation) in an animation. Click in the toolbar on the button. A window will appear in which the emitted signal intensity is animated. Check that the cerebro-spinal fluid extinguishes much slower than the brain, and that the edema extinguishes a little slower than the brain. 4 Diffusion-weighted imaging In MR Imaging gradients in the magnetic field (i.e. a variation in strength along a certain direction) are used for position coding. During excitation, a gradient is applied to select a specific slice, and during recording another gradient is applied to select lines within the slice (see Medische Fysica ). In anatomic MR imaging no gradient is applied in the time interval between the excitation pulse and recording. In diffusion weighted imaging (DWI) two gradients in opposite directions are applied in that time interval: As the precession frequency increases with field strength, the gradient will cause the precession frequency of the nuclei within the voxel to vary, and hence the spins to dephase faster, which in turn will lead to a smaller signal during recording. However, as the opposite gradient is also applied, the effect is canceled if all nuclei are stationary.
On the other hand, if the nuclei move randomly within the pixel (as the result of diffusion), the second, opposite gradient will not completely cancel the first one, as explained in the picture below. In this example a nucleus at the bottom happened to be moving about, while another one at the top happed to be stationary. As a result, these two nuclei don t end up at the same phase after the application of the double gradient. The total effect of diffusion of all nuclei is that the dephasing increases with stronger diffusion. Load the file dti.mri. Make an image with TE=100 ms and TR=8000 ms. The result is a standard T2-weigted image. Now go to the Simulation menu item, and choose Allow gradient double pulse. Press the -button to make a new image. You can now choose between different diffusion imaging techniques to use. Choose single direction. This means an image will be constructed that results when a gradient double-pulse is applied between excitation and recording. Use the dial to choose a frontal direction (toward the nose), and click OK. 15. Why are the ventricles black in this image? Make a new recording, but now with the gradient double-pulse in the left-right position. 16. Observe that the corpus callosum (the structure connecting the left- and right side of the brain) is much darker in the image with the left-right gradient doublepulse. Why is that? When there are a lot of fibers in a certain direction in a voxel, there will be more diffusion in the direction of the fibers than in other directions. Diffusion Tensor Imaging (DTI) uses this property to visualize fiber direction. A DTI-image is the result of many different singledirection gradient images. From these images the diffusion tensor, is a 3x3 matrix that describes the amount of diffusion in all directions, is constructed. Make a DTI-image for dti.mri. In the resulting image lines are displayed that indicate how dominant the diffusion is in a certain direction. If the diffusion is equally strong in all directions, a point is displayed. If the diffusion is stronger in a certain direction, a line in that direction is displayed. The longer the line, the more the diffusion in that direction dominates. At the background of the lines, the Apparent Diffusion Coefficient is displayed: the reconstructed average diffusion in all directions. 17. Observe that the ADC is large for the ventricles. Is that what you would expect? Is there a dominant direction for the diffusion in the ventricles?
18. Can you determine the fiber direction in the corpus callosum, and some other structures? You can make the image larger to observe more detail. In clinical practice diffusion MRI is used to diagnose acute ischemia in the brain. In the early phase, ischemia in the brain doesn t have any structural effects that can be observed in anatomical MRI. However, ischemic cells swell, which diminished the extra-cellular space, and hence the diffusion of fluid. Open the file infarct.mri and make an image with TE=100 ms, TR=8000 ms and no diffusion imaging. The patient from whom this image was made was suspected for suffering a stroke. 19. Can you see any signs of stroke in this image? Now make a Diffusion Weighted Image (DWI). The DWI is the average of several single direction diffusion images. 20. Can you now see signs of stroke? Finally, make an ADC image. 21. What is the relation between the DWI and DTI images? 5 Functional MRI Open the file fmri.mri. As you can see, this is a brain slice. Actually, the file contains a series of 99 consecutive scans. They are T2* weighted images (measured 25 ms after the excitation pulse, no180 pulse applied). An image was acquired every 2 seconds. As the scans were recorded in so little time, the resolution is very low (there was not enough time to scan the slice from many directions). During the recording the following experiment was performed. Alternatingly, in 20 s interval, the subject either did not move his fingers, or made rhythmic movements with his thumb and index finger. By using the arrow buttons on the keyboard you may scroll through the next and previous recording. Below the image the time instant of the image is displayed. You may also display the data as a movie by clicking the play button in the taskbar. Observe that there is not that much difference between the recordings. When you point the mouse to a pixel, a window appears that plots the recorded signal intensity as a function of scan number. For most pixels this is noise, but for a small area in the brain that is active with the task the subject is performing there is a (tiny but significant) difference between signal values between the on- and off-state. 22. Try to find the brain area involved in the task; you will see that this is not easy to do. Click the button in the toolbar. You can choose from a number of ways to display the data. Choose Show fmri Statistics. You will then see an map of the t-value of the pixel. The
t-value is the average difference in signal intensity between on- and off-state, divided by the standard deviation. The larger the t-value, the smaller the p-value (i.e. the change that the difference is purely accidental). If you point the mouse to a pixel the t-value is displayed in the status bar, together with the p-value. In this mode it is not difficult to identify the brain area that is involved in the task. Is the result significant? Note that many other brain areas also light op, probably by coincidence. In Show Both mode the original MR image is shown, while pixels whose p-value is below a certain value are colored. 23. Does the colored region match your expectation? Did the subject move his left or his right hand? Note that the image, as convention demands, must be viewed from the feet up. 6 T1-relaxation and repetition time If there is time left, you can use the rest of the lab-class to catch up on what you learned before about T1, T2 and hydrogen density. An MR scanner always produces many subsequent excitation pulses, as the slice has to be scanned from many different directions (see section 10.3.2 of Medische Fysica). It turns out that the signal strength depends on the time interval between the pulses: the repetition time TR. The longer the interval between the pulses, the stronger the signal. This is caused by the fact that spins can only be excited if they are parallel to the magnetic field (the ground state). When TR is short, not all nuclei will have returned to the ground state, en consequently the signal will be small (see figure 5). Figure 5. As the time between the excitation pulses shortens, the signal intensity decreases. T1 is the repetition time for which the signal intensity is at 37% from its maximum value.
24. Open the file brain_15.mri, and make a scan with TE=10 ms and TR =100 ms. Next make scans with TR=200, 500, 1000, 2000, 5000 en 10000 ms, while keeping TE at 10 ms. The fraction of nuclei that still is not in the ground state decreases negative exponentially. Consequently, the signal intensity increases with repetition time: max 1e / (2) In which T1 is a tissue property that indicates how fast nuclei return to ground state for that tissue (see figure 5). When the repetition time happens to be equal to T1 we see max 1e, or 1-0,37 = 63% of the maximum value. From the images you have just constructed for different TR values you will have noted that the T1 relaxation time is different for different tissues. 25. Make a plot of the relation between signal strength and TR for gray matter (at the surface of the brain), white matter (more to the inside), and cerebro-spinal fluid. Estimate the T1 relaxation times for these tissues (note: not all tissues may have reached maximum signal strength. In those cases try to extrapolate how high the graph will end). Compare the results to the data in table 10-2 of Medische Fysica. The effect of T1 relaxation can also be shown by animation. Start the animation again by clicking the button in the toolbar. Note that with a short repetition time the brain hardly lightens, while the subcutaneous fat does (you may have to turn down animation speed to see the effect). Further note that the cerebro-spinal fluid lightens stronger as repetition time increases, while the subcutaneous fat reaches it maximum at relatively short TR. 7 T1-, T2- and ρ-weighted images In this lab class we have seen that three tissue properties contribute to an MR image: hydrogen density ρ and the relaxation times T1 and T2. We have also seen that the way the relaxation times contribute depends on the repetition and echo time of the scanner respectively. If we combine all this we find for the emitted signal intensity:, ~ e / 1e / (3) Figure 6 shows the signal intensity as a function of TE and TR for four different tissues. This graph might help some people in understanding how it all works, but ignore it if it doesn t help you.
T1=500, T2=50 en ρ = 100% T1=250, T2=50 en ρ = 100% T1=500, T2=100 en ρ = 100% T1=500, T2=50 en ρ = 75% Figure 6. Signal intensity as a function of TE and TR for 4 different tissues with different T1, T2 and ρ properties.
26. In the figure above the relation between TR and signal strength is plotted for a tissue with T1=600 ms. In the same picture, plot the relation for another tissue that has the same hydrogen density and T2 as this one, but has a T1 of 800 ms. Hint: first mark three points: the initial value, the final value and the value at TR=800 ms. 27. In order to see optimal contrast between these two tissues, is it better to choose a short, intermediate or long TR? 28. Suppose that one would make an image with long TR, much longer that the T1 of all tissues involved. If there is a contrast between tissues in this image, can that contrast be the result of differences in T1? If not, what else may cause this contrast? 29. In the figure above the relation between TR and signal strength is plotted for a tissue with T2=100 ms. In the same picture, plot the relation for another tissue that has the same hydrogen density and T1 as this one, but has a T2 of 150 ms. Hint: first mark three points: the initial value, the final value and the value at TE=150 ms. 30. In order to see optimal contrast between these two tissues, is it better to choose a short, intermediate or long TE? 31. Suppose that one would make an image with short TE, much shorter that the T2 of all tissues involved. If there is a contrast between tissues in this image, can that contrast be the result of differences in T2? If not, what else may cause this
contrast? The conclusion of the previous six questions is: If we want T1 to be important in the image, we need to choose an intermediate value for TR. If we want T1 not to be important in the image, we need to choose a large for TR. If we want T2 to be important in the image, we need to choose an intermediate value for TE. If we want T2 not to be important in the image, we need to choose a small for TE. Note that small etc. for TR is measured compared to the T1 values of the tissues, and for TE compared to the T2 values. From this we can conclude that when a short TE is chosen, together with a long TR, both T1 and T2 will not be important in the image. Consequently, only difference in hydrogen density will play a role in this image. Such an image is called a ρ-weighted image. 32. Again open meningioma.mri, and make a ρ-weighted image. Do you think this image is well suited for diagnostic purposes? From question 23 we may conclude that the hydrogen density is not very different among tissues; the variation of T1 and T2 values is much larger. Hence, the relaxation effects are not disturbing influences for which you might want to compensate, but rather provide a very useful contribution to diagnostic images. 33. The left image in figure 10-17 of Medische Fysica is called a T1-weigted image, and the right one a T2-weighted image. Explain the names. Hint: note the TE and TR values in the caption. 34. The left image in figure 10-17 of Medische Fysica is called a fat image, and the right one a water image. Explain why fat is particularly light in a T1-image fat, and water particularly light in a T2-image. Some background information Classic MRI MR imaging is the common method to image the geometry of the brain. In principle MRI images hydrogen density, but the image is weighted with the T1 an T2 properties of the tissue. The strength of these weights depends on the TE and TR setting of the MR device. The MR device generates radio pulses. As a result the axis of the spinning hydrogen nuclei (this spins ) will flip and start to rotate around the axis of the magnetic field in the MR device. This movement is called precession. While falling back to the ground state, the nuclei emit radio waves which are measured by the device, and that are the basis of the image. During precession, the spins will dephase and as a result the signal will decrease with time.
This dephasing is not equally fast for all tissues. The T2 relaxation time of a tissue is the time it takes for the signal to decrease to 37% of its initial value. Depending on the moment after the radio pulse the signal is measured, the echo time TE, tissues will be imaged brighter or darker. MR images are constructed from the signals generated by many repetitive radio pulses. If the interval between pulses (the repetition time TR) is short, many of the spins have not yet returned to the ground state, and consequently only a small signal is generated. The figure below shows the relation between TR and signal strength.
T1 is the repetition time for which the signal is within 37% of its maximum value. Again, this T1 is a property that is different for different tissues. As a consequence MR images will look different when recorded with different TR. Diffusion Tensor Imaging In MR Imaging gradients in the magnetic field (i.e. a variation in strength along a certain direction) are used for position coding. During excitation, a gradient is applied to select a specific slice, and during recording another gradient is applied to select lines within the slice (see Medische Fysica ). In anatomic MR imaging no gradient is applied in the time interval between the excitation pulse and recording. In diffusion weighted imaging (DWI) two gradients in opposite directions are applied in that time interval: As the precession frequency increases with field strength, the gradient will cause the precession frequency of the nuclei within the voxel to vary, and hence the spins to dephase faster, which in turn will lead to a smaller signal during recording. However, as the opposite gradient is also applied, the effect is canceled if all nuclei are stationary. On the other hand, if the nuclei move randomly within the pixel (because of diffusion), the second, opposite gradient will not completely cancel the first one, as explained in the picture below.
In this example a nucleus at the bottom happened to be moving about, while another one at the top happed to be stationary. As a result, these two nuclei don t end up at the same phase after the application of the double gradient. The total effect of diffusion of all nuclei is that the dephasing increases with stronger diffusion. In diffusion tensor imaging (DTI) DWI images are constructed with gradients in many directions, in order to determine the fiber direction per voxel. Below a DTI image is displayed. Lines indicate the direction that has the strongest gradient. The length of the lines is used to show how large the difference in diffusion in different direction is: if there is only a dot, the diffusion is equally strong in all directions, and there is no dominant fiber direction within the voxel. Not that one needs a 3D-image to see diffusion in all directions. Instructions part 3: functional MRI Standard MR images show the anatomy of the brain. Functional MRI shows the activity of the brain. In order to understand fmri we first need to understand another tissues T2* (not to be confused with T2). Actually, the signal drops much faster after a radio pulse than indicated by T2. This is because of local in homogeneities in the magnetic properties of the tissue causes faster or slower precession within a voxel. The actual decay time is called T2*.
In standard MR imaging, the T2* effect is compensated by giving a radio pulse at ½ TE that reverts the direction of the precession. As a result the effect of the local in homogeneities is compensated precisely at TE. That is why the T2* effect is not important in clinical brain images. Because oxygenated and de-oxygenated blood have different magnetic properties, they differ in their T2*. fmri used this difference by making so called T2* weighted images: The difference in T2* is quite small. As a result, the only way to image the difference in tissue oxygenation in regions of the brain (which is what fmri does) is to make two images in different conditions: one while performing a task and one while not performing that task, and then subtract the images. Actually many images are made in both conditions, to get a better signal-to-noise ratio.
The pictures below show how the concentration of hemoglobin changes after neural excitation, and how that effects the signal strength in T2*-weighted images (the so called BOLD signal; BOLD=Blood Oxygenation Level Dependent contrast). This figure shows the typical variation in signal intensity for a voxel that is involved in the task. Note how slight the difference between on- and off-condition is. Statistical analysis is used to identify the voxels for which there is a significant difference between the two conditions. These locations are marked in anatomical MR images of the brain, and are thought to indicate brain regions that are involved in the task at hand. As an example the figure below shows the result of an fmri study where both moving and stationary dots where presented to the subject. The results show that area V1 and V2 in the brain respond to both stimuli (but a bit more to the moving images). V5 and hmt on the other hand respond only to the moving dots. This corresponds to the knowledge that these regions are involved in the detection of movement.