So, I've been working with fMRI since 1998, and in the intervening time, a lot has changed. The technology has come a long way, a number of new techniques have been developed, and also, most importantly for this group, the translation of this technique from a neuroscience tool directed towards understanding things about groups of brains has really begun to be translated and applied to the neurosurgical problem of understanding organization of a specific brain and of disordered brains. So, why would fMRI be helpful in the context of a program that's devoted largely to awake mapping? And I put up some things here, including questions that we face every day in the clinic with our patients. Is the tumor operable? Should it be operated on with awake mapping or other types of adjunctive neurophysiology? What would we do with a patient who, for some reason, cannot tolerate awake craniotomy? How can we discuss the level of risk with our patients? How can we maybe do more focused mapping, make our craniotomy smaller, feel more confident about negative mapping? And then a fallback for situations in which the mapping is either incomplete or difficult to interpret. So, I think there's a number of reasons why fMRI can be helpful. Other things that you just heard, also, for example, the fact that large portions of the motor representation are actually in the sulcus between the pre- and post-central gyrus, and that those sulcal depths may be difficult to assess with intraoperative mapping. So, there are things that fMRI can do that can be difficult to do with awake mapping. Another thing you just heard is the complexity of these systems. The complexity of the motor system, the complexity of the language system, which means that to most completely understand the individual functional networks, we actually would like to test all of these sub-functions. And that's obviously impractical during operative testing. It's not necessarily possible to test every subsystem with fMRI, but certainly we have the luxury of more time when we are doing things preoperatively. So, with that introduction, I just want to review very briefly how fMRI works, that largely this is a blood flow signal, and that we're contrasting neuronal areas, which are in the basal state, with those areas in the activated state. And when they become activated, they cause a vasodilation that's in the post-capillary venules and the capillaries, that's in response to the increased demand, but it's actually an overshoot. So, while the increased demand happens over a relatively short time frame, the vascular response is delayed in time, and what we're actually measuring is the overshoot, not the actual increased utilization of oxygen. So that means that this signal is actually spatially and temporally smoothed, and it's important to bear that in mind when we think about it. It also means that the way a functional map is generated is by doing statistics on a series of images. So, what we do is we acquire a time series of images during a task or a series of tasks, and we usually call these epochs, where the subject is alternating between a lower-level task and a higher-level task. And an example of that that would be very simple is where the low-level task might be rest, and the higher-level task might be moving your thumb. And then we compare the images acquired during those two different tasks, so you can see the epochs coming and going. So these are task epochs, and these are rest epochs, and we apply a statistic to them, and we apply a statistic to each voxel in the brain. Sometimes we correct for multiple comparisons, and we come up with what we call an activation, or some people call a blob. And the reason I think it's important to go through that is that it hinges on many different aspects of the acquisition and processing of this signal before we get this picture. And in that way, it's quite different than, for example, structural MR, where basically what you see is what you get. If you see it on the scan, it must be there. If you see a blob, it's very important to understand how that blob comes to pass on the study. So the applications of this technology to patients has largely been in these domains. Motor, sensory are the most established, visual mapping also quite established, language, more complex, more difficult, still emerging, and memory, the same, perhaps even less developed for a variety of reasons. This is an example of how we might do motor mapping and how it might look in a patient. So this is a patient with a low-grade glioma in the motor area. During a hand task, you see very clearly the hand area, the SMA. You can look at this in multiple dimensions, which can help with your planning, and also other tasks can be incorporated. And there's often multiple tasks that might be applied in a given patient, including, for example, multiple sensory tasks, multiple motor tasks, combination thereof. With all fMRI, the subject, or in our case, the patient, has to be able to perform the task effectively. And that can be hard because some of the patients who are mapping have preexisting deficits. And so one effort that's been made is to look at how effective passive finger movement might be for mapping. And this is work that we actually did many years ago using a pneumatic finger plunger movement. So this was a kind of a glove that would move the fingers passively. And we found that we were able to get good activation in both M1 and S1 using this paradigm and in both healthy subjects and patients. And also, interestingly, that there is less head motion when this is done. And that is a very important thing to mention, which is that head motion, whether it's correlated with the task or not correlated, is extremely prone to causing artifacts in this type of data. And that gets to the fact that in order to do these studies effectively, we have to have a patient who is adequately coached and adequately able to take part in the studies. There have been a number of studies of fMRI motor mapping. I can't possibly go through all of them. But the point is that, in general, it's really quite sensitive. And I think that most people who use it regularly find that they can rely on this strongly for motor mapping. This is another example of what the motor mapping might look like in a given patient, and how it can be helpful to guide the clinician to some ideas of where they're going to locate these functions intraoperatively. So then language mapping. As you just heard, language is immensely complex. And we frequently have patients who come to us who have various deficits of different parts of their language function. And it's very important to assess them and find out what they can and cannot do prior to trying to acquire fMRI. In much the same way that a mildly apparently aphasic patient, when you get them in the operating room and sedate them and put their heads in pins and make them fatigued over a few hours, may not do as well as they did before. We can find that a patient who might seem to have a fairly mild aphasia, when we put them in the magnet and ask them to do something like verb generation every two seconds, they fall apart, and they can't do the task either. And so we have a pretty rigorous system where we evaluate the patients to see what tasks they can and cannot do, and try to tailor the tasks to what they can do, to coach them heavily on not moving during these language tasks so that we don't get artifact from it. And we have tasks that we can substitute for patients who are aphasic. So for example, instead of doing antonym generation, we might just do word repetition. Or instead of sentence completion, sentence listening. And we also, as incidentally, have mapped at least 25 different languages at the Brigham because of who our population is. So being able to tailor this towards speakers of different languages is also something we found to be helpful. This is an example of a patient of mine who had clearly right lateralized language across all of these tasks, and who had become aphasic prior to her surgery. This is an example of some robust language maps that we found. This is an example of some not-so-successful language maps that we found. And you can deconstruct these to any number of reasons. For example, in this patient, you can see large areas of edematous, dysfunctional brain, problems with attention, problems with holding still, problems with understanding the task instructions. So whenever these maps, or a series of maps, come up, and we're looking at them to either create a report or to plan surgery, a huge amount of interpretation and clinical context has to be brought to bear when looking at these, and not just kind of cursory inspection. Again, there are a number of studies that are looking at the correlation between fMRI and direct cortical stimulation mapping. This is one example. This is a relatively large study of 40 patients. One point that I'd like to make, this was a really quite sophisticated study, and they used multiple tasks. But one thing that was used was a fixed threshold, and a fairly stringent fixed threshold. And that means that that statistical map that you saw is going to show far fewer areas of activation. And not surprisingly, that led to a very low sensitivity, 37 percent. But they actually redid this analysis with several different thresholds, including a more globally lenient threshold, which, as you might imagine, increased the sensitivity but decreased the specificity. And then they also had a subject-specific sort of individualized threshold done by an expert. And in that case, you can see that the sensitivity and specificity increase significantly. So one thing I'm going to leave you with, in addition to the very careful selection and administration, selection of the patient, selection of the task, and administration, is also, at the analysis stage, bringing to bear an informed thresholding. And the way that actually I think this works best is with dynamic thresholding. So that you can have a cursor and have the clinician move the cursor up and down to change the threshold and look at where the activations are. And there are some softwares that allow us to do that now. This is another study, again, with the same point, which is that thresholds really need to be individualized. And also, different tasks need to be looked at. So when looking at a patient's language map, we don't look at just one task. We look at all three of them. And we compare whether they look similar to each other, where they look different, and how robust we think the findings are. So another point, of course, is that there are patients who are going to have trouble with task-based fMRI. And we would like to have something to offer those patients, especially because those are the same ones who are likely to have trouble with awake surgery. And that kind of mapping. So this is work from Ralph Suarez, who used to be in my lab. This is work done at Children's now, looking at a passive language task. So this is a story-listening task, and something that you might imagine could be much easier for patients who have a mild aphasia, but not a complete aphasia. And to take that to the next level is resting state functional mapping. And this is a large, large body of work in the neurosciences that looks at autocorrelations across brain regions. You heard our previous speaker mention it briefly. And this group, Eric Luthardt's group, wrote a nice review on this, and numerous groups are now applying resting state to these different problems. So there's been work applying this to motor function, and this is really quite robust. The motor network has these autocorrelations within it. You place a seed within a known language area, and you can bring out the whole language network. And actually, in this case, the resting state was actually both more sensitive and more specific than the task-based fMRI. So this is, I think, a really interesting proof of principle that we can move away from task-based fMRI in that simple, straightforward paradigm. But it can't be done sort of willy-nilly. It requires a really great team to put together the analysis strategies, because you're extracting signal out of what's a pretty noisy background. And this has even actually been used for motor mapping in the intraoperative setting, although I think that might be somewhat controversial, the value of intraoperative fMRI if you have intraoperative mapping. Our group has worked on two approaches, which we call task-free fMRI for language mapping. One is a complete resting state, and the other is a movie-watching condition. In order to do this, we employ some pretty sophisticated analysis strategies. We do independent component analysis. We also create some very complex linear regressors, which we can create from healthy subjects and then apply to the tumor patients, where we don't make any presuppositions about what the structural organization of the brain is. But we do rely on what we think the temporal responses are. And this is just an example looking at the patient with a GBM, looking at the comparison between the sentence-based task and the resting state task. And again, I'll make the point, this is easier for the patients. They're more likely to be able to complete it, and they're also less likely to move during it. And a task-based fMRI, where the patient gets off-task, is going to be really essentially worthless. This is another example where we can see correlations with electrocortical mapping. How much more time do I have? Two minutes. Okay. There are studies that show that fMRI is helpful in predicting risk of neurologic deficits. This is kind of intuitive, that the further away the fMRI activations are from the lesion, the less risk that is posed by the surgery. And as I said, I think this can be very helpful for counseling patients and their families prior to surgery. There's now a meta-analysis of correlations. I think the biggest point here is there's a huge range. And I think the reason there's a huge range of either correlation or discrepancy between this modality and mapping is because there's a huge variation in how these are applied. And I've been talking about fMRI and the variety of ways that that can be implemented and analyzed and that it's absolutely technique-dependent. And of course, the same is true of cortical mapping at surgery. That is absolutely technique-dependent. And so it's not surprising that there's going to be quite a considerable variation in terms of what the sensitivity and specificity is. And even more so if it's possible that not every site that's identified at surgery is actually truly eloquent. And that's a big problem because we don't have any real ground truth. We have assumed that intraoperative mapping is the ground truth, but really the best ground truth would be to be able to do a randomized controlled study where patients get either preoperative fMRI or intraoperative mapping and we look only at their final outcomes. And that study has not been done yet. So I'm going to fast forward a little bit and just say that it's very helpful to then be able to take these images, integrate them into the neuro-navigation system, use them to plan the surgery, plan the approach, plan the resection. Here we've injected the outlines of the fMRI for the motor area into the microscope heads-up view and used it to guide the resection. We also have found in a number of cases, so this is some data from a decade ending in 2013, these are the patients who had fMRI. Of them, 129 were done asleep. 63 were done awake with mapping. And we found that in a number of cases, the patient was unable to tolerate the mapping for a variety of reasons. They had seizures. They had problems with their airways. They just got tired and weren't unwilling to cooperate. And I would posit that when you're in a situation where you're partway through a surgery and the patient becomes unable to give you helpful information, it's nice to have this kind of data to fall back on when doing this type of surgery. And this just shows the clinical adoption over the years at the Brigham of this type of technique. So I'm just going to close by saying there are some benefits. It's obviously not invasive. It can be done before surgery. It can be used for counseling. It can be repeated. You can study numerous functions, so you can start to break down these sub-functions of, for example, the language system. You can see into the sulcal depths, which I think is a particularly important thing as we've heard. And it provides the functional and structural data integrated into a single framework and can be viewed in different planes and can be brought into the operating room. Some of the disadvantages, it does not distinguish essential from non-essential cortex. And how we actually differentiate essential from non-essential cortex is a very big question. We partially do that when we block stimulation. We block with stimulation at surgery, but this is not that type of technique. It's an observational technique, so you're seeing which parts of the brain are participating. Again, it's very dependent on patient task performance. Different types of pathology can degrade the signal reliability, so bringing that knowledge to bear when looking at the imaging is very important. Head motion is just a terrible artifact and basically makes the study worthless and needs to be prevented. It cannot really be regressed out. We make some efforts to measure it and regress it out, but it really is such a big contaminant as to make the study worthless. And then this very important issue, which is how do you choose the threshold and then how do we, as a group, come up with some standards that makes this more homogenous across study centers. So, I'm going to leave you with these last messages. In terms of adopting and getting the most out of this technology, patient selection and coaching, paradigm selection and deployment, a rigorous analysis workflow, and QA, including individualized thresholding, and then informed interpretation are really the pieces that have to come together to do that. So, since I'm out of time, I'm just going to close there and say thank you. I really look forward to bringing all of these streams of thoughts together over the course of the meeting today and tomorrow because I think they have a lot to inform each other about. Thank you.

fMRI

neurosurgery

awake mapping

decision-making

mapping accuracy

patient selection