In his interview with Paul Weavers, Ravi Menon, Director of the Centre for Functional and Metabolic Mapping (CFMM) at Robarts Research Institute, the first ultra-high field resource in Canada, talks about how MR methods, neuroscience and clinical diagnosis influence each other’s work and the importance of the collaboration of these three worlds in research.
Paul Weavers (PW): The technical parts of MR are extremely interesting by pushing your hardware capabilities and trying to make advancements and publishing papers on MR methods etc. – However, how much of this work is driven by an application in either clinical diagnosis or neuroscience research need?
Ravi Menon (RM): It is a little bit of all three actually. A lot of the neuroscientists come and ask questions for which we generally have sequences well established, such as multiband or MP2RAGE etc., and therefore do not require an enormous amount of innovation for. On the other hand, there are others who want to be able to image the auditory cortex and layers in the motor cortex, which require development and analysis tools and drives the neuroscience innovation.
In clinical research, we are constantly trying to take sequences that were designed, which might take 15 to 20 minutes in a research setting and try to get those down to four to five minutes because there are certainly techniques such as quantitative susceptibility mapping which take too long because people move on the order of the size of the voxels and in that time the voxels are very small at 7T.
The third part of our research program is the combination of hardware and software developments. We particularly innovate on the radiofrequency coil side by constantly developing new coils that will allow new neuroscience or allow a better patient access such as wider fields of view, mirrors, eye tracking, things like that.
PW: We can say that there are three different worlds: the clinical research, the neuroscientists and the technical people working on hardware and software developments. Do you actively have to bridge these different worlds?
RM: Yes, otherwise none of the projects would ever go anywhere! For example, it requires a neuroscientist to say “I need to measure myelin or cortical thickness” or “I would like to see the superior colliculus.” They then have to work with staff members to optimize the pulse sequences or develop new pieces of hardware or whatever it takes to actually make that into a reality. It is a collaboration which you can see in the papers we publish.
PW: Do you feel the neuroscientists at CFMM are learning more MR methods or MR physics?
RM: They do, we have a number of courses here that cater to people with different backgrounds and that is very critical. We teach a course called “fMRI for newbies” which introduces statistical methods and fMRI paradigm design and things like that to people with neuroscience and psychology backgrounds. On the flip side, we teach advanced fMRI methods in our biophysics department for the small number of physics orientated students. We also provide advanced statistical courses of various types because acquisition is only half the battle, analysis is actually where the real abuse occurs with the data. If you do not set your thresholds and you do not apply your statistics correctly – that is when you can really get into trouble!
PW: Sounds like a lot of analysis and statistical didactic construction. What about the image formation process? Do they get into that at all?
RM: To varying extents. Certainly, within the biophysics department they do in great detail. Within psychology and neuroscience not as much, however we do focus on the artifacts a lot. They need to be able to recognize the artifacts and try to figure out where those might be coming from. This is particularly critical these days because fMRI analysis has become very pipeline orientated. It is a black box, images come out of the scanner go through this black box and out comes process data – “garbage in and garbage out”. There is usually no visualization of anything in that pipeline, therefore people need to learn to look at the data, look at the motion characteristics and all of that.
PW: On the flipside of this, putting on some neuroscientist to learn about MR, do you have a strong push for MR methods people to figure out an application for their work?
RM: Absolutely, most of the students and Postdocs have an endgame in sight – and that endgame is either neuroscience or clinical application. Students in particular are typically paired up with a clinician or neuroscientist during the course of their PhD – depending on their project of course – and similarly with Postdocs.
PW: Do you feel the funding model or the funding process is helpful in making a pure MR methods person aware of what they need to do in an application, or a neuroscientist aware of the image formation process?
RM: It varies by agency. We have this very large grant in cognitive neuroscience right now, which is 66 Mio. Dollars. However, we do not do a lot of technical development in that because grants submitted to that program typically say: “this is not cognitive neuroscience”, except that cognitive neuroscience would not be possible without technical development. Therefore, in situations like that it can be a barrier.
In other situations, Canadian Institutes of Health Research for example, they are quite motivated for you to cross disciplines and to have a co-applicant, who might be a MS neurologist, a visual motor neuroscientist or something like that. And that is viewed as a strength. As long it is not window-dressing.
PW: We have been talking about meeting together between the MR methods people and neuroscientists. Does it need to be funding driven? Is that the real incentive to be able to work or is there some other way you can foster?
RM: We foster it through a lot of meetings, colloquia, informal networking groups and things like that from which ideas come. Those ideas eventually find some funding to implement them. I do not think the funding drives the science, the science really motivates going out and getting the funding!
PW: You have participated in the development of fMRI from the outset of its development as a subset MR research. How do you see current improvements in methods of acquisition impacting fMRI as a tool as a scientific test?
RM: I think the major improvements in the last decade have focused around better access to high-fields which make the BOLD effect more pronounced but come with a lot of other artifacts that one has to deal with. The availability of multiplexed sequences, whether there are things like GRAPPA and SENSE or multiband or simultaneous multi-slice techniques have allowed the “greedy” neuroscientists to cover the whole brain as quickly as possible.
PW: To clarify – from the neuroscientist outcome perspective – what do they get from that? Is it an improvement in spatial coverage of the brain and done faster? Is there a downstream endpoint?
RM: Definitely, they want the whole brain and to be able to do it faster than a second because even though the fMRI responses is sluggish there are ways to extract some degree of tiny information.
Perhaps the third piece of that puzzle is that stability is extremely important. Therefore, scanner manufacturers and companies like Skope have provided lots of tools to make the acquisitions as stable as possible. You become limited by physiologic noise – which is where you want to be! You do not want your scanner to determine the noise.
You have to think about how do you eliminate that physiologic noise: Whether it is in real time or retrospectively, whether there are things such as field cameras that measure real time field shifts, real time motion shifts? Things like that will reduce the physiologic noise. This critical to seeing smaller and smaller effects and seeing effects particularly in individuals. fMRI is mostly done in group averages. That is not really interesting if you want to know whether your mild cognitive impairment is going to turn into Alzheimer’s. So, for N=1, which is for diagnostics, you need really good stability.
PW: Is the end goal to say “we want to get to a single subject study for fMRI”?
RM: Yes, such as really robust findings. You have done an experiment and have shown that there is a group difference between A and B, which might be neuro-scientifically interesting, however, if you are going to apply it in the clinic it has to work for one subject. This can be beyond hardware and software, probably goes into various forms of classifiers and AI of some sort, to be able to make those. Therefore, having the least amount of noise and most amount of signal always helps.
PW: We have talked about how methods improvements have changed questions investigators are asking. Do you have any particular examples of your head-only 7T performance system? What are you asking here which cannot be asked somewhere else?
RM: Head gradients fell out of fashion and now they are coming back into fashion because they allow gradients slew-rates that really cannot do with a body gradient coil without getting peripheral nerve stimulation. They also allow extremely strong gradients and for various sorts of diffusion that is very important, for example the Connectome scanner and the next generation of the Berkeley-MGH scanner. All of these are high slew rate, high gradient strength head gradient systems. That allows us to go to resolutions and look at larger animal models that we cannot do in our 9.4T like non-human primates done in the 7T. And they are done in resolutions that are very high compared to the whole-body system.
PW: So, smaller voxels, more signal and less noise?