good temporal resolution of approximately half a second or greater. And to the best of our knowledge, being exposed to a magnetic field, like a three tesla field, or even a seven tesla magnetic field, does not have any long lasting effects. In some cases, people report a little bit of headache or dizziness in a very strong magnet, like a 7T. But that's passing and usually gone by the time that people are done with the experiment. So there are no long-lasting effects of having an MRI, even having multiple MRIs over the course of your lifetime, unlike, for example, an x-ray examination or a CT examination. Critically, it also allows us to investigate local structural activity associated with the cognitive function, but also brain networks of the entire extent of the brain. So unlike single cell recordings, or other invasive techniques, it allows us to look at the brain as a whole functioning unit, rather than smaller subsections by itself. So it's a very powerful method that allows us to study neuroscience at different levels of analysis, from the smaller approximately one millimeter structures, all the way to whole brain networks. So how then do we do a typical fMRI experiment? One the left-hand side, I'm showing you an example that we've now seen many times of an actual slice through the brain, this is our volume. And the image size is going to partially determine our resolution, the quality of the MRI scan that we're collecting. We're going to layer over top of that a matrix, and we can determine the size of that matrix within certain parameters. If we choose a large matrix, the individual voxels or pixels here will be rather large, and the resolution of our resulting scan will be somewhat low. If we overlay a small matrix, we have a higher resolution, we have more information from that same image size, but obviously, it's going to take longer to acquire that amount of detail. So, depending on the needs of our experiment, we can choose our matrix size and determine the resolution, the detail of the data, that we're going to collect. Then if we look at one of these individual voxels, just three dimensional pixel, we can look at the activation at a certain point in time. So let's say that this person in this experiment is not doing anything particular, the activation from that area of the brain should be somewhat random. So at a point in time, we can take a measurement, which gives us a certain level of activation for that area, measured by the T2 star relaxation time, and we can do this repeatedly over the course of a time period. So every few seconds we take a scan, which is called TR, time repetition, the amount of time it takes the scanner to acquire a new image. For that one voxel, we then take a measurement of the T2 star decay, which gives us activation over time for this particular voxel. And this, at the bottom of your screen, what you see is referred to as a time series. So this is an activation level for that one particular voxel indicated in the image here, over the course of the length of the scan. Now, you essentially do this for every single voxel in your volume, so imagine that you do this for all the voxels that you see in this matrix here. And then in the third dimension, the extent of your volume gives you a great number of these time series, these activation patterns for each and every one of those voxels. So we collect a tremendous amount of data by just doing fMRI experiments. So let's look at a very specific of an fMRI experiment, in its simplest form, you can think about a finger tapping experiment. Let's say, as an investigator, you're interested in finding out which area of the brain is responsible for being able to finger tap, literally as simple as tapping your index finger and your thumb together. We would place a willing participant in the scanner, ask them to hold their fingers still for a period of time, and then start tapping for a period of time, and then we'll ask them to stop. We'll start to continue again, and rest again, and we do this for a number of occasions over time. We measure for each and every one of the voxels in our volume, the activity associated with that finger tapping task, which would give you a time series roughly as what you would see on the bottom. And the idea here is that if an area of the brain is responsible for that finger tapping, activation in that voxel, or in that group of voxels Will be higher for the tapping conditions than it would be for the rest conditions. And you can see some evidence for that in the time surge at the bottom of your screen. So initially during rest activation, it's somewhat low. When tapping commences, the activation rises a little bit. For the second block of tapping, you see a very robust response of activation, and towards the end, you see another sort of block of activation corresponding to the tapping on condition. So here you would see a situation in which you measure fMRI signal that is representative of the task that you're asking the participant to do. Now then, how does this activation that we see at the bottom here, this time series, correspond to the hemodynamic response function that we've talked about? Well if we zoom in a little bit at one of these areas of high activation, we essentially expect that while the person is continuing to finger tap, the area of the brain responsible for supporting that motor movement is continuously producing hemodynamic response functions where there's increased influx of oxygenated blood to that area. If you can concatenate those hemodynamic response functions, you can see a somewhat box curve, as you see at the bottom of your screen here, that is very similar to the on off condition of the experiment. And obviously, they're offset in time a little bit because the hemodynamic response is somewhat slow, as we've talked about previously. But basically, the concatenation of the hemodynamic response functions results in these high areas of activation for the on condition relative to the off condition. And then the question obviously becomes, of all the voxels in our brain or in the volume that we're scanning, where do we see this pattern of low and high activation that corresponds to the onset times and the stop times of the rest conditions and the finger tapping conditions? Once we find that voxel or more often a cluster of voxels, we can conclude that that area of the brain is responsible for finger tapping. And in this particular case, it's obviously in the area of activation, we see in the motor cortex in this particular case, the right motor cortex. So this is a simplest form of an fMRI experiment where you used this hemodynamic response function in a block design to activate an area of the brain responsible for this finger tapping task. And block design are a very commonly used design in fMRI research. Because there's many advantages to using a block design. It's also sometimes refered to as an epoch design. They're simple to design and implement. You have an on condition and an off condition and you typically just alternate the two. The analysis of this data is very straightforward. You have a block of time in the off condition and a block of time in the on condition and you contrast activity between the two. It results in very robust activation typically because you have a large number of events that are causing a hemodynamic response. So the overall fMRI activity typically is very high in a blocked condition like this. It's also robust to the uncertainty and timing and shape of the hemodynamic response function. Let's say, for example, there's some variation in the length of the hemodynamic response function or the steepness of the hemodynamic response function in a particular area of the brain. A block design is somewhat insensitive to that because it aggregates over a number of hemodynamic response functions and just contrasts high activity conditions with the low activity conditions. So the individual differences in hemodynamic response functions for each finger tap are less consequential in this type of design. But as you can imagine, there's obviously also some downsides to the block design. You have to assume for this to work that there's a single mode of activation of a constant level over time. If I'm asking a person to continuously finger tap, but activity in response to the second and third or fourth finger tap is significantly different from the first, a block design would not necessarily work. You need a constant level of activity to each and every one of the different stimuli in order for a block design to successfully pick up areas of high activity versus low activity. It is also limited in the sense that you cannot infer anything about the individual events within a block. If you do a block design, you typically compare on to off as we've said, but you would not be able to differentiate activity in response to the first finger tap versus the fifth or the sixth within a block. It treats the entire on condition as one single activation. And sometimes that's a limit depending on the type of question that you have for your experiment. It also cannot infer anything as we've just discussed in the positive even though it's robust against variations in the hemodynamic response function. It also does not allow you to look at individual variations in the hemodynamic response function. So if a particular area of the brain shows a slightly different hemodynamic response function or different timing of it, you would no be able to pick that up with a block design like this. An alternative then to a block design is an event related design where some of these things are possible. So at the top I'm showing you a schematic representation of the block design. You have on and the off conditions as we've talked about. You multiply that with the hemodynamic response function that you expect and you get these areas of high activity and low activity temporally offset relative to the on and off task conditions. An event related design works very similar but the events are much shorter and usually consist of one trial. So for example, it would be one finger tap. Each individual event is spaced from the others in variable intervals of time. The expected hemodynamic response function remains the same, but because of the spacing, you would get a different accumulation of the fMRI signal of the time series than you would get from block design. So at the bottom right hand side, I'm showing you an example of what that expected signal would look like in an event related experimental design. Now how then do we differentiate these individual events from that accumulated or that accumulative signal? You can do that by essentially defining a single equation for each time point of your stimulus onset, multiply that by the function that you would expect to see based on the hemodynamic response function. And essentially deconvolve that complex signal into individual events associated with the timing onset of the stimulus that you're interested in. So in the middle shows you that schematic representation of what the individual events look like relative to a block design and an event related design. And one of the biggest advantages in this design is that it allows you to make inferences or draw conclusions about individual events in s series. For example, in the right hand side you can see a series of pictures. You could ask for each individual stimulus what the response to that item is and what the representation or the activation in the brain is to each of those stimuli separately. There's several other advantages to event related designs in fMRI. It allows for inferences about the timing of neural activation, like I said, these individual trials can be considered separately. It allows for very flexible analysis of the data including post-hoc trial sorting. So if you want to combine different trial types together into a single event type, or you want to reorganize things temporally, an event related design allows you to do that whereas a block design does not. And when you space the events of the trials sort of randomly apart from each other, you make sure that you don't confound the next trial or a series of trials after that. There are several disadvantages of event related designs as well. It does have a reduced ability to detect changes that you would get from a block design. Because it's only an individual event, activation and response to these individual events are typically a lot smaller than you would get from a series of events using a block design. It's very sensitive to errors in the hemodynamic response function, so if there's variability in the hemodynamic response function, an event related design is much more sensitive to those changes. And there can be refractory effects that carry over from one trial into the next. So the response to the first trial could affect the response to the second trial in certain conditions, and if that's important, it could confound the results that you're dealing with as well. So what we've now seen from both the block design and the event related design is that there are areas of on activity or high activity that we contrast with time periods of low activity or no activity. The tapping versus rest, and in the event related design, certain stimuli versus not having any stimuli. This is referred to as the cognitive subtraction method, which is a critical component of fMRI research. If you look at a time series, you see that there's always activation in the brain. So it's not that we have to look for areas in the brain where there's activity and in other areas there's no activity at all. We have to do a comparison between two contrasting conditions to see what the difference in activation is, that we can then assign to a cognitive function that we're interested in. This is also known as a categorical design, it assumes that different cognitive components are independent in space. For example, that memory occurs predominately in the hippocampus, that motor control predominately occurs in the motor strip. That these are spatially separate, and that we can dissociate them from each other. It also assumes, based on a principle proposed by Donders in 1868, the Pure Insertion Principle, which processes in complex conditions are simply added on top of simpler or baseline conditions. So, for example, if I do a complex executive function type task, and I have to tap a button to give my answer, that the simple motor task is just laid on top of the executive function task, which presumably happens somewhere else in the brain. And that that's not any different from the button press that occurs with a simple arithmetic task or a simple language task. So, it assumes that behavior and cognition is somewhat modularly organized, and that complexer tasks are just a combination of simpler modules that we can discern on this fMRI signal. The key, then is to subtract activation during a control task from activation during the on or experimental task. And the assumption there is that the difference only shows activity related to the cognitive construct of interest. So on the right hand side, I'm showing you an example that is very frequently used, of a flickering checkered pattern, which is used to activate the occipital cortex, important for vision. The checkered board pattern comes on at a certain frequency and they turn it off for a period of rest. They turn it back on for a period of stimulation, and you compare the two activity levels for these two conditions. And in the top right, you can see the off state, and if you look carefully in the bottom brain image there, you will see some intenser signal in the occipital lobe on the very bottom side of that brain, which presumably corresponds to the on condition of this flicker pattern. But we know from experimental design that subtraction of two conditions is only valid if the condition is only varied in one property. If the difference between the two experimental conditions is more than one property, we're going to have difficulty concluding which one of those two was responsible for the experimental manipulation that we enforced. So any factor that covaries with the independent variable is going to be a confounding factor and that must be controlled for. So experiments that typically use rest or not doing anything at all as a comparison condition or a control condition are actually a fairly poor approach. At the bottom, I'm showing you an example of a paper where that was explicitly tested, that using a baseline of not doing anything or a baseline of rest is actually a very poor control condition. It turns out that during rest, there's quite a number of things that are going on. We're uncoding our current experience, our current location, but there's often also the case of mind wandering. What's going to come next, how did I do on this task that I just finished? I need to remember to pick up a gallon of milk on the way home. There's quite a bit of cognitive activity that goes on when we're supposedly not doing anything, that using such a rest period as a controlled condition is actually a very poor design. It is a much stronger approach to have active contrast between condition A and condition B, and not use rest at all. So, for example, in a study where you're interested in an area of the brain that is responsible for processing faces, you could imagine a situation where the on condition would consist of pictures of faces that you show to people. And the control condition would be a similar image, merely a scrambled version of that face. So, that all the visual information, the contrast, the light gray, the light dark shading, everything remains the same between this two stimuli. The only difference is that the one constitutes a picture of a face, and the other one does not, allowing you to look for areas of the brain that are particularly sensitive to faces, pictures of faces, or faces in general. And there is such an area, and it's called, the fusiform face area. Alternatively, you could expand a little bit on this experiment and make a combination of faces, objects, and scrambled versions of those two. Allowing you to contrast the area of the brain responsible for faces, or for objects, and comparing both of those against the condition on which the same information is there, merely scrambled in a non-sensical stimulus. And at the bottom, you can see some activation that results from this type of approach. Important to remember from these slides is that the experimental design and the design of your fMRI test is critically important for the conclusions that you're going to be able to draw from the resulting activation. So in the next module, I'm going to step through a number of other factors, that should be critically considered when you're designing an fMRI experiment.