So, you want to learn (f)MRI
Brain imaging using MRI is a useful and popular technique, and a PhD or a postdoc is an excellent time to acquire some expertise in this area. There are many skills involved in conducting MRI studies, which makes it difficult to master quickly. This page is intended to give an overview about the various skills and time commitment involved.
What are the basic skills involved?
A short list of required skills:
Experimental design (both psychological and MRI-specific concerns).
Some programming knowledge—in my lab, this is primarily Matlab scripting and debugging. We use aa analysis scripts to analyze data, which rely on Matlab scripts, and you will likely need to learn how to extract data from MRI images so you can use other statistics programs (like R). (Python is also a popular option.)
Specifics of fMRI analysis packages (such as SPM or FSL) and details of aa analysis scripts.
Some degree of familiarity with the terminal (command line).
A program for presenting experimental stimuli (my lab currently uses PsychoPy and EPrime).
Statistics. MRI typically relies on a combination of "standard" statistics (i.e., General Linear Model) and some issues that come up especially frequently in neuroimaging (cluster-based statistics, correcting for multiple comparisons, avoiding circularity in inference).
Software to display the results (such as MRIcron, Connectome Workbench, or other similar programs).
It can be hard to learn these things out of context, so most likely this will occur as you analyze data, from a combination of books, websites, and mailing lists.
How?
I think the most common way of learning neuroimaging analyses is through an iterative process involving self-directed study, research (textbooks and internet, such as software mailing lists), and asking colleagues and mentors. A good place to start is to look at some of the great material available online—lectures and presentations can provide a great overview. You'll probably want to watch them more than once as you gain more experience, because each time you'll learn something additional.
I don't know how to understand MRI design and analysis without reading metods articles. I've put together a reading list on MRI methods. It's incomplete, but a good place to start. Rather than read through articles from beginning to end, one nice approach is to couple reading and analysis, so that as you implement a stage of MRI analysis you are reading related papers. This can be very useful in understanding the analysis options and provides some practical context for what can be dense methods papers.
One of the most important tools in learning analysis is trying lots of different approaches and comparing the results. For example, you may have noticed that papers will use different amonts of spatial smoothing, ranging from no smoothing up through 12 or 16 mm FWHM. To understand why this matters, a good first step is to search for papers that have systematically looked at the effect of smoothing on results. However, it is also informative to try it tourself: take the same dataset, smooth at three different levels, and see how the results differ. (You also may need to spend some time thinking about how to quantify the differences: subtracting images, histograms of t statistics, reliability over two sessions, etc.).
Asking for help
Of course, sometimes after trying all of your other resources you'll be stuck and the best thing to do is ask for advice. Input from others does the most good once other avenues have been exhausted. Imagine these two kinds of questions:
- Scenario 1: "How do I correct for motion in my images?"
- Scenario 2: "I've performed motion correction on sample data and read the chapter in the SPM manual on motion correction, and the chapter in the Statistical Parametric Mapping book on rigid body registration. When I did a literature search I found two papers that suggested simply including the motion parameters as covariates wasn't the best approach. I searched the SPM and FSL email lists and there doesn't seem to be a clear consensus (here are links to replies that I found). What do you think?"
In the second example above, it is much easier for a more knowledgeable colleague to help, and as an added bonus you've learned a lot along the way (although doing so takes time).
In my experience, researchers who are the most successful at being competent MRI scientists have spent many hours trying to figure things out themselves through reading, internet searches, and trying out different ways to analyze the data.
Courses on MRI analysis can also be extremely helpful, but are not a quick fix. In my experience as an attendee, I've found I use about 10% of the information that is presented. As a novice, I picked up some basic ideas but most of the lectures were over my head. As I learned more, the basic parts became less interesting, but advanced tidbits extremely helpful. So, courses are great, but don't count on becoming an expert over the course of one workshop!
How long does it take?
Assuming no prior experience, my best guess is that it usually takes people 1-2 years to become comfortable designing and analyzing fMRI experiments relatively independently. There is an enormous amount of variability in this; the biggest factor is probably someone's motivation (and, relatedly, the amount of time they invest).
Because becoming a competent brain imager is not a quick process, I typically don't plan on people doing 1-2 year research projects in my lab doing their own MRI study or analyses.
The substantial time involved is daunting, but also has some upsides:
- Being a good neuroimager takes time but there is no magic involved. If you are motivated, conscientious, and put in the time you will probably succeed.
- There are so many resources available online, and so many example datasets out there, that if you are really interested in learning (f)MRI analysis there is nothing to stop you. In other words, you don't need to wait to attend some magic course or for your boss to sit down and walk you through. You can just learn it!
fMRI in the Peelle Lab
If you are considering joining my lab as a graduate student or postdoc, here are some specific notes about what you might find:
- I'm a firm believer in open and reproducible neuroimaging. For new studies that we run, most will be preregistered, and data and analysis scripts will be available for all published manuscripts.
- Speaking of analysis scripts, the first time you learn something it's fine to "point and click" your way through an analysis. However, all final analyses must be scripted. The first time you do this it will take a while, but it will be well worth it in terms of improving our science and preparing you for your future. You don't have to enjoy programming to be a good brain imager, but you need to do it.
- You'll need to be able to learn a lot about imaging relatively independently. Of course, I will help you and answer questions, but the motivation and the step-by-step progress will need to come from your internal motivation. There will be some times where you just hit your head against the wall, but that's OK.
Getting started in the Peelle Lab
There is no perfect approach, but you could do worse than the following:
- Get to a basic level of understanding Matlab.
- Download SPM and work through the example datasets. At each stage, read the manual and related papers to understand what is going on (yes, this takes a while).
- Download aa and work through the example. Try a couple of different analysis options and compare them.
- Find a dataset you care about and analyze it. Bonus points for writing down or preregistering your analysis choices ahead of time.
