FUNDAMENTAL
NEUROSCIENCE FOR NEUROIMAGING
PART 4: Fundamental Neuroscience for Neuroimaging
Experimental Design
In the previous module, we've talked about the basic design of an fMRI experiment. In this module, I'd like to talk a little bit about things to consider when you're designing an fMRI task. The goal of the fMRI experiment typically is to use stimuli to induce a certain psychological state that you are interested in. Then use the MR signal to measure brain activity associated with that psychological state to be able to draw conclusions about where and how in the brain this psychological process is supported. In the previous modules I've talked about this example, which is a very simple example of an fMRI experiment. Here, obviously the psychological state of interest is motor control of a finger tapping. We compare an active finger tapping condition with a passive rest condition and the instructions here merely at the beginning of each block of tapping is to start tapping your finger or stop tapping your finger at the beginning of a rest state by comparing brain activation between those two conditions, we can localize and examine the brain activation that is responsible for performing this task. But obviously, we can do many more complicated types of design using more sophisticated cognitive processes. Memory executive function, vision, all types of domains that we've talked about previously. I'd like to talk a little bit about factors that you should consider when designing an experiment and they can be grossly subdivided into technical limitations, psychological state considerations, and finally the statistical design of your paradigm of your experiment. So in talking about technical limitations, we have to consider the constrained circumstances in which we collect fMRI data. Obviously, the subject is positioned in fairly tight quarters in the bore of the scanner and often the stimuli are presented using a projector outside of the MRI room as to not interfere with the magnetic signal and project the image from outside of the room onto a screen that's within the bore of the fMRI scanner and the subjects can see that screen through a mirror that's mounted on top of the head coil, as you can see a little bit from the image here on the right hand side. The subject is then given button boxes in their hands either in both hands or on one hand, which are often limited to only three maybe four
buttons in each hand. So within that constrained environment, we have to remember that this visual stimuli on the screen typically need to be fairly large to be able to be seen through this mirror and that our response options are limited to three or four options of show because that's the maximum number of buttons a participant has available. We also should consider the stimulus timing and the length of the stimulus. So we've talked about block design and event related design before. When we do an event related design we have to make sure that the timing or the amount of time between two stimuli is somewhat varied. If we don't do that, we get a fixed interval and not enough variability in the MR signal to be able to what's called deconvolve the signal and assign a single hemodynamic response to an individual stimulus. Instead, we need to vary the spacing between the stimuli so that we have more variation in the overall signal, which allows us to reliably assign a hemodynamic response to the onset of a stimulus. So we need to consider the experimental design in terms of timing of the presentation of the various stimuli that we want to include in our experiment. We also should consider the number of stimuli that we can present at a time. Typically, an experiment is designed or subdivided into individual runs and a run will consist of a certain number of stimuli. Runs should typically be limited to four to six minutes or so. And this is because of scanner drift. Due to thermal energy and the continuous operation of the scanner, the scanner will heat up ever so slightly. This increased thermal energy will change the magnetic field, the measurements that can be taken from the MR scanner, which potentially could influence the results if you were to compare stimuli at the very end of your run from stimuli at the beginning of your run. So typically runs are limited to four to six minutes or so and then stopped so that the scanner has time to cool back down to normalize again if you will before the next run is started. So when you design your experiment you have to make sure that your stimuli, the sequence of stimuli can be subdivided into these individual runs. One should also consider the overall length of the experiments. You can certainly run a number of runs one after another. But the participant is in a somewhat uncomfortable environment, it is difficult to communicate with the technician due to the loud noise of the scanner and they're constrained in this small scanner tube. So typically you want to limit experiments to about 60 minutes or so or less in order to give the participant the opportunity to come back out of the scanner and distraction and be a little bit more comfortable. So both in terms of the number of runs and the overall length of the experiment, one should consider the physical constraints and the technical limitations of the scanner setup. Another technical limitation to consider is subject motion. Obviously when we're doing measurements at the millimeter scale, any type of motion by the subject will significantly interfere with our measurements. On the right hand side, you can see the green, blue and red lines indicating the amount of translation or the amount of shifting that this particular person did in the MRI scanner and you can see that sometimes it is several millimeters. If our voxel size is about a millimeter to a millimeter and a half, that could significantly change the measurements that we're trying to get from our experiment. The bottom side shows you a similar problem with rotations, so any type of rotation of the volume within the magnetic field will interfere with our measurements. The very top shows an example of motion in a structural scan which causes blurriness and the same is true for functional MRI scans that will have this blurriness as a result of motion that will decrease your ability to detect signals related to your active and your rest conditions. So it's very important to control for motion as much as possible which typically means constraining the participants using padding next to their head, which makes the overall experience unfortunately less comfortable. So again the length of your overall experiment should be considered and reduce the likelihood that motion becomes a significant factor in your experiment or in your results. When we're talking about psychological state, there's a number of things to carefully consider when you're designing your experiment. The key obviously is to induce the psychological state that we're interested in. So with our design, we have to ask the question if we're successfully doing that. For example, is the subject effectively engaged in the intended task? Imagine a situation in which a participant is presented with a stimulus and asked to determine if they've seen this before in the context of the experiment i.e. is the item new or is it old. Here I'm showing an example of a Rubik's cube and imagine the total trial length is about three seconds or so when we're measuring the signal associated with this recollection. But this particular participant is able to retrieve that information within the first 500 milliseconds or so. After 500 milliseconds, the participant presses the correct button, we then still have about 2,500 milliseconds before this trial is over. And it is very likely that the participant starts to think about things not related to the experiment. For example, he could recall that he was given a Rubik's cube for his birthday when he was about four or five years old, the person is wondering where that Rubik's cube is right now, whether or not he would still be able to find it; all things that are not relevant to the psychological state that you're trying to measure. So when designing an experiment it's very important to consider how long the psychological state lasts in relation to the length of your trial. The second thing to consider is whether the subject is employing a shortcut or a strategy to solve the problem. I'm showing an example here of a Go-no-Go task in which a participant is shown a series of letters on the screen. The participant is asked to press the button as quickly as they possibly can for each letter that they see except for the letter X at which point they should not press the button. So this is a task that measures response inhibition, the ability to not press a button when you've been doing that for a while. Now imagine that there's a certain order to these stimuli. For example, the X occurs as every fourth stimulus or the X occurs after each S presentation, the letter S is seen. This makes the sequence predictable at which point the cognitive state is no longer about response inhibition but about anticipation and predictability. So when organizing the sequence of your stimuli, one should consider very carefully the shortcuts or strategies that can be employed to solve the question at hand. How long does the psychological state last is a very important factor to consider. The complexity of the cognitive task should be considered. Some cognitive problems are much more involved and complicated and take more time to resolve than others. If you are then comparing a simple cognitive task with a complex cognitive task, the amount of time that the person needs to solve that problem can be different, which could influence your results. This is particularly true for emotional responses which are typically slow. If your goal is to induce an angry state or a happy state or a sad state, those typically take longer than cognitive processes and they're more difficult to switch, so you can imagine after you've induced an angry state, it is much more difficult then to use another stimulus to then induce a happy state. You need a certain amount of time to pass between these types of trials in order for that to go back to normal. So it's very important to consider how long the elicited psychological state lasts in the context of the trial length that you're designing. Finally, does the stimulus reduce or induce unintended psychological states? You can imagine a situation in which a person is engaged in a complex cognitive task and maybe they don't know the answer to each of the questions being asked or that they feel that they're not performing the task very well, that person might get frustrated, might be anxious about whether or not they're doing the task right or whether or not they will be judged for not doing well. And at that point you're essentially measuring a state of frustration or irritability or even anger rather than the cognitive process that you're interested in your experiment. So it's important to consider the unintended psychological states that could be induced from your experimental design when you're designing your experiment. Moving on now to some of the statistical considerations that we should keep in mind. Obviously, the important questions are going to be what are the dependent and independent variables here? Will independent variables have multiple levels and how will the stimuli be organized? Will we be using a block design or an event-related design or maybe a combination thereof. Very commonly used experimental designs in fMRI studies are the subtraction design, which we've talked about a little bit before in the previous module, a factorial design in which we include multiple levels of an independent variable, a parametric design in individual differences approach or an outcome measures design, and we'll step through each of these to give you a little bit of an example for each of these types of design. But the fundamental trade-off for all of them is always that when you use fewer conditions in a simpler experimental design, you have better power but you have less ability to be specific about your conclusions and less able to generalize your results to other types of conditions. When you use a more complicated design with multiple levels or more conditions, typically it results in worse power but you do have a better potential to make inferences across your different types of stimuli. So that's the trade-off when you're designing your experiment that one should keep in mind. We've talked about the subtraction design before. Subtracting activation during a controlled condition from activation during the active or experimental condition, and the idea is that the difference between the two shows activation in an area that is important or supports the cognitive state or the psychological state that you're trying to measure. This works particularly well in cognitive processes where the cognitive process is considered to be categorical. So for example, I'm showing an example here of the famous faces task in which the experimenter is interested in the brain area and manner in which the brain supports recognition of familiar faces. You can imagine an example where familiar faces are shown and the activation during those familiar faces is compared with activation in response to unknown faces and again the difference should then relate to our ability to recognize familiar faces. In a factorial design, we allow for multiple levels of a single factor to be included in the experiment. And this is particularly important when we're interested in the interaction between two types of conditions. So here I'm showing an example of stimuli where we use curvilinear spaces of curvilinear environments and rectangular environments that are either open spaces or more enclosed spaces and they're also varied by the level of high ceilings and low ceilings. So here we have multiple factors that are embedded in these types of stimuli and full factorial design allows us to look at activation that is important for the interactions between these types of conditions. Now the downside for the full factorial design is that sometimes the results that come from these interactions are very difficult to interpret. But if the interaction is of particular interest, a factorial design is the only way to approach your research question. In a parametric design, localized activity varies as a result of difficulty or cognitive demand. So typically what you do is you use a task that has multiple levels of difficulty such as the N-Back task shown here at the bottom of the screen. In the N-Back task, a participant is shown a black diamond that has a number in it as you can see here and in the 0-back condition, the participant is asked to press the number corresponding to the number that they see on the screen. So in the far left stimulus, the person would press the button number one. In the 1-back condition, they see a similar series of black triangles with numbers in them. And during this condition, the participant is asked to press the number corresponding to the number they saw on the previous slide. So for the first slide, they would not press anything but when the second slide comes up they would press the number corresponding to the one on the first slide. And you can imagine, you can increase the working memory load for this problem by increasing the number of 2-back steps if you will. So in the 2-back condition, you have to press the number corresponding to two slides ago, again increasing the working memory load. The problem here is that the downside of this type of experimental design is that psychological states may not always be linear as is assumed in this particular case. You can imagine a situation in which if a 4-back or a 5-back condition is employed, participants are simply not able to solve that problem and activity would be non-existent rather than higher because of the 4-back or 5-back condition. So there is a point at which there's optimal arousal in response to the difficulty after which things are likely to drop off and could interfere with the outcome of your experiment. In an individual differences design, individual behavioral performance is considered and correlated with the activation that is generated from the fMRI results. So here I'm showing an example of our own work where brain activation caudate in the hippocampus was established or was measured during a spatial navigation task and the individual subject's behavior performance on the recall tasks was correlated with their activation. So each dot here represents an individual subject and you can see a nice correlation between behavioral performance or task success if you will, and the level of activation in these two brain structures. So here we're making a connection between the behavior that an individual subject is displaying and the brain activation. And finally in terms of outcome measures design, we can use fMRI experiments to test the effect of a particular intervention. This is referred to as a within-subject design in which a person gets into the scanner twice, once before an intervention and the second time after the intervention and we can use the difference between the pre and the post test to examine the effect of that intervention. And these interventions can span a wide range of things. Obviously, drug development is an important factor where we look at the effect of a particular drug on brain function and on cognitive function. We can also use practice or training as an intervention to see what the effects are. There have been a number of studies that used exercise as an intervention method to see what effect exercise has on the brain. Or you can talk about other types of clinical intervention like deep brain stimulation which is often used in Parkinson's disease to control the motor symptoms, transmagnetic stimulation where an external magnetic field is applied to a specific area of the brain to try to alter the electrical activity in that area of the brain or ECT, electroconvulsive therapy, which is commonly used for depression. So you can do a pre scan, do the intervention, and then do a post scan to determine using fMRI what the effect is of those types of interventions. Finally, briefly because we've already talked about this, it's important to consider how this stimuli will be designed within your experiment. We've talked about the block design in which a series of stimuli are organized within a block and that is compared then with the rest condition. The N-back is an example of a block design because you need to provide the instruction at the beginning of a block and then have a series of stimuli that all have the same instruction before you move on to the 2-back or the 3-back. But you can in other situations use an event related design if you have questions about the individual stimulus. For example, the recollection of the Rubik's cube that I showed before or do a hybrid between the two, where pseudo randomly organized events are organized in blocks so that you can analyze the data either as a block design or an event-related design depending on the question that you have. What I have hopefully made clear is that there are a number of factors to consider when designing your experiment. I've only highlighted a few and there's many more that one should think about when designing an experiment, which takes quite a bit of research and practice and experience to be able to do. Careful experimental design is critical and selection of the experimental control conditions is critical for the outcome of the study. So one should really focus a significant amount of time considering all these factors and testing out their experiment before employing it in the fMRI scanner and the MRI scanner. In the next module, I'm going to talk about a specific case of functional magnetic resonance imaging which is called resting state connectivity analysis.
Functional Connectivity MRI Studies
In the module on fMRI experiments, we saw that an fMRI experiment is basically taking a measurement from every single voxel in our volume. And looking at the bulk response within that voxel over time. So here we see an example of that time series that we then obtain from each and every single voxel in the brain, over time. The goal then is to look at that time series and compare it with the on and off conditions of our experimental task. We've talked about the finger tapping experiment where we comparing period of tapping with periods of rest. And the goal then is to look at the time series activation of the bolt signal. And look for areas of the brain where during the tapping condition the activation is high and during the rest condition the activation is low. If we find such a brain area, the assumption then is that that brain area shows a high degree of correlation between the task conditions and the activation. And therefore must be critically important or at the very least involved in the performance of that task. Functional connectivity MRI is essentially a special variant of fMRI. It is based on the premise that the brain activation is observed under any condition, even in the absence of any external task demands to or stimuli. Examining the brain during rest when it's not engaged in any type of cognitive or motor task can be very useful in assessing the functional organization of the brain. Completely separate and independent from cognitive or behavioral performance. Functional connectivity MRI is based on the Hebbian principle that states that when cells fire together or fire in synchrony, they must wire together. In this particular case, the assumption is that when two cells or two fractals in the brain show a similar pattern of activation, a correlated pattern of activation. That those two voxels must be functionally connected, that they must contribute to the same task or be part of the same brain network. So a functional connectivity MRI experiment then collects time series from every voxel in the volume. And looks for correlations between the time series, in different locations in the brain. So here I show an example of three different voxels from which we measure a time series. If those time series show a high degree of overlap, the assumption is that these three voxels are part of the same brain network, or are at least functionally connected to each other. Now, the correlation could obviously not only be a spatial correlation, but also a temporal correlation. If there's a small time offset between the time series activation, we can still consider those to be functionally correlated, j
ust separated in time. So in a functional connectivity MRI experiment, you essentially collect over time a number of time series for each voxel, for each location in the brain, over time. And look for correlations both spatially and temporally to determine what areas of the brain show high degrees of correlation of activation. And therefore are considered to be functionally connected with each other. In this analysis, we can take what's called a voxel to voxel connectivity approach. Where we look at the time series from a single voxel in the brain, and look to all other voxels in the brain for a correlated levels of activation. So we pick one single voxel in our volume as the source. And then do correlations of every other voxel in the brain, their activation with that source voxel to look at voxel to voxel connectivity. So on the right-hand side here, you see an example of such a resulting correlation map. Where the color of the voxel indicates the degree of correlation with your originally selected voxel. An alternative approach is to use a seed based connectivity analysis. In a seed based connectivity analysis a region of interest is selected, either based on anatomy or functional task performance. The area is defined by usually overlaying a sphere, and then taking the average measurement of the time series within that sphere. So usually a sphere consists of a number of voxels that have time series associated with it. You take the average of that time series within that region of interest, within that seed. And then you're looking for other areas in the brain that show activation that is correlated with the activation in your seed region. On the right hand side, I'm showing an example of functional connectivity analysis based on the seeds on the left motor cortex and the right motor cortex. And clearly, you can see here, a very strong correlation and activation between these two pieces of cortex. The bottom graph shows a measurement from the left motor cortex, compared to the left visual cortex. And here, as expected, we see a low degree of correlation between the time series. And the assumption there is that these are functionally not, or at least functionally less connected with each other. And that the left motor and the right motor cortex show a high degree of functional connectivity. You can base these seeds on anatomical regions by overlaying an atlas on top of your structural scan as we've seen in one of the previous modules. Performing a segmentation of the structural volume, identifying brain areas or structures of interest. And then calculating the average time series for that brain region, that anatomically defined brain region. The blue and red arrow here indicate the left and the right motor cortex respectively. And the top right shows the time series, the average time series that is observed in terms of functional connectivity between these areas. The very bottom then shows a scatter plot of all the individual measures from all the individual subjects of the correlation between these two brain structures. Showing a high degree of correlation and activation between the right and the left motor cortex. A slightly different approach is by looking at functional connectivity data and trying to determine how many independent components or independent networks of correlations can be observed. This is referred to as independent component analysis of resting state fMRI data, or functional connectivity MRI data. So you're looking for networks of brain areas that show a high degree of correlation. That are dissociable from other networks that in and of themselves show high degrees of correlation. And you're looking then for the maximum number of independent or orthogonal network components that you can discover that away. Ones that are very commonly observed are listed here. And they include higher order visual motor processing, lower order visual processing. Motor control as you would expect to see in a button pressing task or a finger tapping task, a vigilance network which has to do with attention or attending to the task at hand. Error monitoring and inhibition, response inhibition, often shows a network of brain regions associated with that. And finally brain areas that have been associated with visual monitoring. A very specific example of one of those networks that you can observe using independent component analysis Is called the default mode network. The default mode network consists of a number of brain regions with highly correlated activation when a person is not engaged in a task but instead they're resting. Very often, you will see involvement of the posterior singular cortex, as you can see on the right-hand side image. The medial pre-frontal cortex as well as areas in the temporal lobe. The posterior singular cortex is very often considered a critical hub of the default mode network. This default mode network is relevant, because it tends to be less active during the performance of an external task. So again, during rest activation in this network is high when a person is engaged in an externally started initiated task. Activation of the default mode network tends to be lower. The default mode network is developmentally established. In very young children, this correlated activity is typically not observed. But as children get older particularly between the ages of 9 and 12, we see this correlated activation emerge in the brain in the same brain areas. And that persists throughout adulthood. There's also great similarity between the rodent and the primate brain In regions involved in the default mode network. Or brain activation patterns that we see during rest. The areas that you see here in the rat default mode network are very similar to those observed in the monkey, which again, show a great degree of overlap, with the human default mode network. So there's great similarity and overlap between the rodent and the primate brains in the default mode network. And it's a very robust observation, it's a very robust finding. The default mode network is a functionally connected network. But in the case of the default mode network, it also shows a great deal of overlap with structural connectivity. The default mode network is thought to support comprehension of information and learning and memory. But it is also thought to be critically important for and for the neurological basis of self, self reference ability, our ability to obtain autobiographical information, etc. It's also thought to support our ability in thinking about the past and the future. So it allows us to use information from our memory to think about the past, but also extrapolate us to what might happen in the future. And finally it's also thought to be in support of the theory of mind, social cognition as well as emotion. So our ability to refer to ourselves, to evaluate our own performance. To evaluate our own standing relative to those around us in terms of social cognition and to imagine the perspective of the other person. So the default mode network seems to be involved in a number of different critical functions. And a lot of study is still continuing to figure out exactly how the default mode network supports these behaviors and these abilities. In terms of its relation to tasks, the default mode network seems to be correlated with successful memory encoding when the default mode network is deactivated. So, again, the default mode network tends to be active during rest and needs to be deactivated in order to engage in an external stimulus or an external task. The level of deactivation of the default mode network seems to be correlated with successful encoding. Deactivation of the default mode network also seems to be correlated with the task difficulty. In that the more difficult, or the more demanding the external task is, the greater the deactivation of the default mode network. And finally, activation of the default mode network after we learn a particular piece of information tends to improve retention and later recall off that information. So in terms of connectivity studies, the basic premises that we collect function MRI data during rest. Typically this is done by asking the participant to stare at a plus sign on the screen, not fall asleep, but also not think about anything specific. Try to be as restful as possible. We then use the resulting data to look for correlation patterns between time series obtained from each of the voxels within that volume. And use either brain structures or clusters of activation resulting from task-based fMRI to look for brain networks associated with that structure. Or that cluster of activation resulting from the task. We can assess connectivity as a function of task-based performance. And finally, we can compare connectivity between participant groups. And just to give you an example of those last two categories. Here I'm showing an example of a study where they compared connectivity with the superior temporal cortex between cases with Alzheimer's disease, cases with frontotemporal dementia and healthy control subjects. And what we see here, we see altered functional connectivity of the superior temporal cortex with the cuneal cortex to supracalcarine cortex. The intracalcarine cortex, and the lingual gyrus as a function of Alzheimer's disease or frontotemporal dementia when compared to healthy control subjects. In this slide, I'm showing an example of relating connectivity measures with task performance outside of the scanner. So by observing differences in default mode network deactivation or changes in default mode network deactivation between young and older adults. We see that that's strongly correlated here with task performance. In that younger adults show a greater deactivation in the default mode network, which is then associated with better performance on the memory task. Older adults show less deactivation of the default mode network, which is then associated with poor performance on the task. So here's an example where functional connectivity data is used to compare two groups, old and young adults, as well as correlated with behavioral performance on a task. Connectivity studies are used in a great number of different contexts. And connectivity and default mode network changes have been reported in a wide range of diseases and disorders. Including Alzheimer's disease, autism, depression, schizophrenia, aging, epilepsy, Parkinson's disease, obsessive compulsive disorder, and anorexia nervosa. So, all these studies, essentially look at brain areas that are critically important or associated with the disease. And then look for network changes that result from connectivity measures and connectivity differences between those brain areas and others in these patient groups. These types of data are also used to examine individual differences. There have been a number of studies that have shown that individual variations in default mode network deactivation or individual variations in functional connectivity between critical brain areas is associated with individual differences in task performance. And in some cases even personality or other types of behaviors. So functional connectivity studies are a special variant of task based fMRI studies. And significantly contribute to our ability to study the brain using these different approaches. And functional connectivity studies are particularly important for network approaches. In the next module, we're going to talk about structural connectivity studies based on diffusing tenture imaging.
Diffusion Tensor Imaging
In this module, we're going to discuss diffusion tensor imaging. We have seen that in order to record an MRI signal from a volume, we must first excite that volume using a radio frequency pulse. The radio frequency pulse is referred to as the "Larmor frequency," which in turn depends on the gyromagnetic ratio of the particle that we're interested in imaging. On the right hand side, I'm showing the table representing the gyromagnetic ratio of the various particles that can be observed in biological matter. Hydrogen is the most commonly used particle for structural and functional magnetic resonance imaging, and that's because it has an unpaired proton which gives it magnetic properties, and therefore, ideally suited to disturb the local static magnetic field. The second part is that it's ubiquitously available in the brain. It's everywhere in the form of water and fat tissue, which makes it possible for us to image the entire volume that we're interested in. If hydrogen was not available in part of the brain, we would not be able to create an image of that area of the brain, giving us a problem in representing the whole volume or the whole image that we would like to create. So, the fact that it's ubiquitously available throughout the entire brain makes it a perfectly suited particle for magnetic resonance imaging. Diffusion tensor imaging focuses on water diffusion throughout the brain. I'm showing an example in the far left of a dye being injected into water. And as you can see, initially, it diffuses freely throughout the water. If it diffuses equally in each and every direction, that is referred to as "isotropic diffusion," as can be seen in the middle schematic. But at a certain point the dye hits the glass container in which the water is contained, and it can't diffuse any more in that direction, any further in that direction. So, when there is constraint diffusion that has a directionality to it, as can be seen in the far right schematic, that's referred to as "anisotropic diffusion." So, if there's a directionality to it, it's referred to as "anisotropic." If there is no directionality to it, it's equal in all dimensions refer to it as "isotropic.". In the brain, we can imagine a situation in which grey matter, which consists of cell bodies, if we take a measurement of global diffusion throughout gray matter, that the water diffuses equally in almost all directions, which is isotropic diffusion. But if we focus on areas with a large number of axons, you can imagine a situation in which water diffusion is constrained along the length of the axon. So by taking measurements from the brain of isotropic and anisotropic diffusion,
we can make an estimate of the location of gray matter and axons, or white matter. Diffusion tensor imaging then employs gradient pulses which cancel each other out in the case of static water molecules, but it causes a lack of signal for diffusing water molecules, appearing as darker voxels on the brain images that you can see on the right hand side. By creating images from multiple different directions, a three dimensional volume can be created representing the diffusion model of water throughout the brain. The displacement of molecules, of water molecules, is defined by three eigenvectors, which in turn have three eigenvalues. They essentially represent the diffusion in three dimensions, in three directions. The largest eigenvector represents the direction of diffusion. So in the right hand side, you can see in the isotropic case, the eigenvectors are equally long. The water is diffusing in all three directions equally. In the anisotropic case, one of the eigenvectors is clearly much longer, much greater than the other representing the diffusion of the water molecule in that particular direction. Diffusion tensor imaging then, takes a measurement from every single voxel in a volume, according to a specified voxel grid, that we've now seen on several occasions. But instead of creating a white or a gray matter of value, we now create a value for the directionality of waterflow within that voxel. We measure the largest eigenvalue for that particular voxel, which in turn represents the direction of water flow through that voxel. By taking a series of measurements, and measuring the eigenvalues for each of those voxels, we can then create a three-dimensional volume of eigenvalues, and thereby a three-dimensional volume of the directionality of waterflow throughout the brain. On the top right hand side, the direction is color coded by red, green, and blue, so that you can easily see the direction of waterflow through these voxels. The bottom is an example of where those voxel- The vectors are still superimposed on each voxel, so you can easily see the direction of water flow in a zoomed in area of that volume. If you then take a measurement of each voxel relative to the voxel right next to it, you can determine the overall direction of fiber bundles or axons, running from one area of the brain to another area of the brain. So, here you see a zoomed in section of these voxels for which we now have measurements. And by determining the eigenvalue for each voxel relative to the one before it then after it, we can figure out an overall vector of direction of the water flow through that larger volume. This then is used to create fiber bundles or connected voxels, where the diffusion goes into the same direction for the entire volume of acquisition. So, here you see a typical diffusion tensor image, where connected fibers are determined probabilistically and color-coded based on whether they move from left or right, whether or not they move. So in the case of red on the right bottom side, whether they go anterior to posterior, front of the brain to back of the brain, as can be seen in the green fibers, or superior to inferior, top of the brain to bottom of the brain, as indicated by these blue fibers. And here you can very clearly see in red, the corpus callosum, which is obviously a very strong significant fiber bundle that connects the left hemisphere to the right hemisphere, representing that communication between the brain halves. Diffusion tensor imaging can then be used for a number of different types of studies. Studies use fractional anisotropy, which is a fractional measure of diffusivity. It basically takes the three dimensions of anisotropy down to one single measure, a fractional measure, representing the directionality of water flow within that voxel. A value of zero represents isotropic diffusion, water moving in every which direction equally, whereas one represents perfect diffusion in only one direction. It's typically used as a measure of fiber density, with a greater number being greater anisotropy, meaning stronger directionality and thus fiber density. It's also used as a measure of brain connectivity because of that, and also as a measure of white matter integrity. The greater your anisotropy, the greater, the denser the white matter is there, and the better the white matter integrity, at least, that's the assumption. So, you can then use that FA value to look at group differences in fractional anisotropy. On the far left hand side, I've shown an example of an FA map derived from DTI data on which a segmentation atlas is overlaid, so that you can identify areas of the brain that you may be interested in analyzing, and then you can create a mean FA value for that region of interest. On the right hand side, I'm showing you an example of a study where they took exactly that approach. They took the corpus callosum and subdivided into seven segments, as you can see in the middle figure, and then took a measurement of FA for those seven segments comparing healthy controls with patients with multiple sclerosis. And multiple sclerosis, obviously, is a white matter disease that reduces the integrity of white matter, so the hypothesis was that in these seven segments of the corpus callosum they could see differences between the control and the patient groups, and the bar graph on the top right shows exactly that. There are certain segments of the corpus callosum where patients with multiple sclerosis have reduced FA values compared to the control subjects. We can also use FA measures to make correlations or to look for correlations with particular symptoms. Here I'm showing a study in patients with obsessive compulsive disorders. On the top left, you see the brain areas in which they've seen significantly different fractional anisotropy in certain brain regions in the OCD patients relative to the control subjects, indicated by the blue arrows, including the midbrain, the lingual gyrus, and the precuneous. And the top right shows correlation coefficients between the FA values within those brain areas and the YBOCS score, which is essentially a symptom severity checklist. So, in this particular case, the midbrain FA values and the precuneous FA values show a negative correlation with the number of symptoms that these patients with OCD are experiencing. On the left hand side here, I'm showing an example of an FA correlation with past performance, in this particular case, in a study of boxing. Subjects who have been continuously boxing were given a memory task in which they were asked to learn a number of words. And after a 20 minute delay they were asked to remember as many of those words as they possibly could. And this was compared with the fractional anisotropy of the left ventral striatum. And we can clearly see in the boxing group that there's a strong negative correlation between the ability to retain that information or the ability to remember that information and their FA values in the ventral striatum. In the right hand side, I'm showing you an example of correlation with life events in that same box or study, where the number of bouts is positively correlated with DTI-derived measures. And the bottom shows that the number of years of boxing that a person has completed is negatively correlated with the FA value with the white matter integrity of a brain area that they were interested in. Diffusion tensor imaging can also be used and is extensively used for surgical planning purposes. On the left, I'm showing you an example of tractography analysis conducted on diffusion tensor imaging in the cases for surgical planning for deep brain stimulation, often in the case of Parkinson's disease. In C in the bottom left, you can see the location of the electrodes that are implanted for stimulation, which are indicated in red. And here, tractography is used to determine if the electrodes are essentially implanted in the correct fiber tracks that they're trying to stimulate. So, in D you can see the same thing. The yellow lines represent the electrode placement with the blue dots being the sites of stimulation. And here tractography is used to confirm that the electorate placement is correct and that they ended up in the correct fiber pathway appropriate for stimulation of this particular patient population. On the right hand side, I'm showing an example of using diffusion tensor imaging for surgical resection. In this particular case, as you can see, outlined in red is a tumor that is observed on the MRI scan. The green areas are highlighted fiber tracks coming from the thalamus, very critically important for brain function. And the imaging here is used to make sure that the resection of the tumor is not going to interfere or damage the fiber bundle that runs by very closely to it. So, they've taken some measurements to determine the distance between the area that needs to be resected in the area that's clearly critically important for these fiber bundles that are running through it. So, here are two very practical applications of diffusion tensor imaging being used for surgical planning purposes. So, overall, diffusion tensor imaging is a visualization method for the directionality of water flow throughout the brain, representing fiber pathways and connectivity pathways throughout the brain that are very different from the functional connectivity that we've talked about in one of the previous modules. This is purely based on structural connectivity between areas of the brain. So, it provides a measure of structural connectivity. It provides a measure of structural organization of the brain. And it can be used for group comparisons between various different groups based on disease or some kind of other factor. It can be used to correlate with symptoms or life events, as we've seen in the case of the boxer study. And it can also be correlated with cognitive performance, like memory or really any other type of cognitive performance. In the next module, we're going to talk about another example of a specific application of imaging in this particular case, spectroscopy imaging.
Magnetic Resonance Spectroscopy
In this module, we're going to discuss yet another specialized application of magnetic resonance imaging known as "magnetic resonance spectroscopy." We've seen on several occasions now that precessions or spins are in a low energy state when they are aligned parallel to the principal axis of the static magnetic field, or in a high energy state when their anti-parallel aligned to the principal axis of the static magnetic field. And we can change a spin from a low energy state to a high energy state by introducing electromagnetic energy. This is known as "excitation." We've also seen that the most efficient way to introduce that electromagnetic energy is by using the Larmor frequency, the radio frequency pulse based on the Larmor frequency formula, which basically states that the Larmor frequency depends on the gyromagnetic ratio of the particle that you're trying to image, as well as the static magnetic field. Precessions that are in a high energy state or that are excited from a low energy state to a high energy state actually causes small magnetic field changes at the nucleus, typically in the opposite direction of the magnetic field, the static magnetic field. This change is the local effective magnetic field, and actually causes the emitting frequency the resonant frequency of the particle to shift a tiny little bit, because we see that the Larmor frequency depends on the gyramagnetic ratio multiplied by the static magnetic field. By introducing that radio frequency pulse to excite that particle, we actually change the local magnetic properties, which then in turn changes the resonant frequency just a little bit. And this is referred to as a "chemical shift." Chemical shift is the change in resonant frequency that results from a small change in the local magnetic field, again, as a result of that excitation pulse that we've introduced. The size of the difference or the value of that difference of the resonant frequency gives us information about the local molecular group, which that nucleus is part of. So, magnetic resonance spectroscopy then is an imaging approach that aims to quantify the local presence of certain molecules, certain chemical compounds based on the shift in that resonant frequency. The chemical shift as expressed in parts per million. As we can clearly see from the Larmor frequency formula, the Larmor frequency heavily depends on the strength of the magnetic field. If we want to compare the frequency distribution or the presence of certain molecular compounds within a brain area across different studies, it would be
useful to have a metric that does not depend on the strength of the magnetic field, so that we can compare concentrations from a three Tesla scanner with concentrations from a seven Tesla scanner. Therefore, parts per million is used as the unit to express the quantity or the density of that molecular compound that's available in that area of the brain. And parts per million is calculated by dividing the change in the resonant frequency that results from the introduction of that excitation pulse by the frequency of the spectrometer, the frequency of the measuring device that is used to quantify these measurements. Now, note that one is expressed in hertz while the other is expressed in megahertz, so there's an order of magnitude of difference between the two, which results in the parts per million part. So chemical shift imaging, which is another term for magnetic resonance spectroscopy then, is nothing more than taking a measurement from a particular area of the brain. And rather than determine the location of water protons, which is what we typically do for structural and functional imaging, we take an entire spectrum measurement of the radio frequency that can be measured, that can be obtained from that area. And that's referred to as a "spectrum." The value of the difference of the resonance frequencies gives information about the molecular group that the nucleus is part of. So, by quantifying these peaks, by labeling these peaks of frequencies that we see, we can determine the local molecular groups that are available in that particular area of the brain. Again, this is expressed in parts per million. The frequency in the sample is subtracted from the frequency in TMS, which is used as a baseline compound, setting the range to zero, essentially. And that difference is divided by the frequency of this spectrometer, the device that is used to make these quantifications. And that's how we get one of these spectra distributions, as you can see on the right hand side. Spectra can be obtained from different types of nuclei, but typically or commonly protons are used, again, because they're highly abundant in the brain, and they're very sensitive to these types of changes. They're very sensitive to these types of measurements. When we look at molecules of interest, they typically have a very low concentration in the brain. But because we're using proton-based quantification, we know that water is also going to be one of those compounds that we can observe. Water, as we've seen, is very abundant in the brain. And the water signal in these spectra is much, much greater than the compounds that we're typically interested in. So the water signal must be suppressed, because it's an order of magnitude greater than what we would like to focus on. Water suppression is used for exactly that purpose. Chemical shift selective suppression or "chesss" is one of those approaches. And it essentially uses a very specialized pulse sequence to saturate the water signal and remove it from the read out. By doing so, we can focus the read out attention or the read out range on the molecules that we're interested in. as you can see in the right hand image. So, the spectra provides a detection of brain metabolites, molecules that we're interested in. And the area under the curve for a particular metabolite gives us the concentration in that local area of the brain. So, again, we're focusing on a sub-region of the brain, a small area of the brain. This voxel size is typically much larger than what we would use in structural or functional imaging, but it certainly doesn't encompass the entire brain. We're looking at sub-regions of the brain to quantify the amount or the density of metabolites that are locally available there. Certain spectroscopy pulse sequences are more sensitive to metabolites than other metabolites. So, typically, within a spectroscopy session, different pulse sequences are used to quantify these different types of metabolites. And as we've clearly seen from the Larmor frequency formula, the fields strength heavily determines the Larmor frequency and therefore our ability to detect these compounds. So, higher field strengths typically result in much better measurements, particularly for spectroscopy imaging. This type of imaging can be very useful and important, because changes in metabolites very often precede structural brain changes. For example, in the situation of nerve degenerative disorders or a stroke, we can see changes in the metabolite profile before we can see changes in the structural composition as we would see in structural MRI. Just to show you a few examples of common metabolites that can be observed with the spectroscopy imaging, NAA is a very commonly seen one. And you can see it's the highest peak in a normal brain. It's a marker of neuronal and axonal viability and density, so it's often used as a marker of structural integrity of the brain area that you're trying to image. And a decreased concentration of NAA is associated with white matter disease or malignant neoplasm. So, on the right hand side, I'm showing a typical example of one of those spectra. And we can clearly see that the NAA signal is the largest signal of all the metabolites that are quantified there. Creatine or CR is another common metabolite observed. As you can see that the second from the left peak is creatine. It represents molecules that contain creatine and phosphocreatine. It's considered a marker of energetic systems in intracellular metabolism. And reduce levels of CR are observed, again, in cases that show brain tumors. So, that's another marker that can be used in cases of tumors and cancers. Choline, CHO, it represents choline and choline-containing compounds. It's a marker of cellular membrane turnover reflecting cellular proliferation. Increased CHL seen in infarction and inflammation, but it is somewhat nonspecific, because you can see it both in infarctions situation or inflammation from all sorts of different sources. So, that can be the result of many different issues in the brain. Lactate is typically a very low peak in a normal brain, you see there on the right hand side of the NAA peak. It's a marker of a metabolism that results from cerebral hypoxia, ischemia, seizures and metabolic disorders, again. And it frequently occurs in cysts, normal pressure hydrocephalus, and, again, in certain brain tumors. So, again, it's another marker that can be used to assess brain health. Lipids, "lip" for short, is very difficult to detect. Typically, you see two peaks of lip that have to be compared and combined. It's considered a marker of cellular membrane breakdown and necrosis, as in metastasis or malignant tumors. So, again, you see this again as a change marker that's very common in brain tumors. On the right hand side, I'm showing a table of some others, and the reason that you would see changes in that metabolite marker associated with it. Clinically, spectroscopy is most commonly used for brain tumors and metabolic disorders. As we've clearly seen from some of the examples of the metabolites, some of them frequently co-occur or increased or decreased in the context of brain tumors. So, spectroscopy is somewhat sensitive to the presence of brain tumors, as well as metabolic disorders. Here in the example, you can see that the white square is superimposed on a tumor in this particular brain. And as we compare the spectra to the one that we've seen before from a normal brain, we can clearly see that the CHO peak is even higher than the NAA peak. And we had just seen that NAA typically is the highest peak in a spectra. So, here, we see a spectroscopy spectra that shows us that there is something abnormal happening in this particular brain area. And, again, in this case, it concerns a tumor. In research, there's many more applications for spectroscopy imaging. Here, I'm showing you an example where measurements were taken from three different areas in the brain, as indicated in the far left image by the white squares. And on the right hand side, I'm showing the resulting table where they've quantified the concentration of metabolites in these different brain areas, comparing them between patients with seizure disorder and a group of healthy control. And you can see the differences in the concentration of some of these metabolites, differentiating the patients from the control populations here. There's many other applications that can be used when it comes to research. So, clearly, I just showed you an example of comparing a patient group and a control group in the concentration of certain metabolites in particular areas of the brain. It can also be applied in terms of correlation with structural volume or the growth of a tumor. It can be used as a correlation with white matter integrity, and in some cases, it can be used in correlation with functional state. For example, like an APGAR score has been shown to be correlated with certain metabolite concentrations in a particular area of the brain. On the right hand side, I'm showing an example of a correlation between white matter integrity derived from diffusion tensor imaging, which we've discussed in one of the previous modules, and CR in the right temporal stem. And you clearly see a positive correlation between these measures here. So, magnetic resonance spectroscopy is a way to quantify particular molecules most often bring metabolites within a particular area of the brain. It's not so much about the spatial location of these things. The voxel size from which you take these measurements is usually preset. But it gives you a measurement of the chemical compounds that are locally available. So, this concludes our discussion of the special applications of magnetic resonance imaging that we've talked about that included functional connectivity analyses, diffusion tensor imaging, and magnetic resonance spectroscopy.
Conclusion
Well we've come to the end of the course. My goal of this course was to provide an overview of neuroscience topics that are relevant to the collection, analysis and interpretation of neuroimaging data. We've discussed a great number of different topics, including the structural and functional organization of the brain. Terminology that is commonly used in brain research, brain networks and how neurons communicate with each other, and how networks communicates with each other in the brain. Cognition and the different cognitive domains that can be assessed. We've talk about the principles of the MRI signal and how it is generated and measured. We've talked about experimental design and the design of neuroimaging studies and what to keep in mind when designing such studies. And finally, we've talked about several specialized applications of neuroimaging methods. Like function of connectivity MRI, diffusion tensor imaging as well as spectroscopy imaging. This course was by no means designed or intended to be a comprehensive review of these topics. It was merely designed as an introduction of the concepts that are important to understand the basics of these approaches, the basics of these methods. If you're interested in learning more, there are many great resources out there. And I'm pointing you to just a few examples of such resources. If you're interested in learning more about the neuroscience topic, I could highly recommend the book Principles of Neuroscience by Kandel and colleagues issued in 2012. It discusses, in much greater detail, cell and molecular biology of the neuron, synaptic transmission in there by communication from one neuron to another as well as the neural basis of cognition more broadly. If you're interested in cognition, Cognition the Thinking Animal by Willingham, published in 2006, discusses in great detail, cognitive psychology, the various cognitive domains that have been defined, memory, attention, executive function etc. As well as the methods that are commonly used in cognitive psychology to study those domains. If you're interested in learning more about brain organization and networks, Olaf Sporns wrote a book entitled Networks of the Brain, published in 2016. That goes into great depth about brain networks, organization, and the development of those brain networks as we grow. It discusses networks that are relevant for cognition and it also discusses networks that are relevant for certain diseases. Principles of Magnetic Resonance Imaging, published by Huettel and colleagues in 2014, goes into great depth about the basics of the MRI signal. But understanding pulse sequences for various purposes and the applications of magnetic resonance imaging. If you're interested in the analysis of fMRI data, Principles of functional Magnetic Resonance Imaging by Lindquist goes into great detail about how to acquire fMRI data, how to process fMRI data as well as various applications and a group analysis of fMRI data. In fact, this work was developed into a course that is available on the same platform. It's also entitled Principles of Functional Magnetic Resonance Imaging One and there's even a second course as a follow up. If you're interested in learning more about diffusion tensor imaging, the book Introduction to Diffusion Tensor Imaging by Mori and Turnier published in 2013. Goes into great detail about the basics of DTI imaging, the basis for the MRI signal there, the analysis and modeling of DTI data, and specialized applications of DTI imaging. Thank you for taking this course. If you have any suggestions on how to improve this course please leave them in the comment section. And I look forward to seeing you in one of our future courses.