As described in my previous post, I recently had the once in a lifetime opportunity to visit the Max Planck Institute for Human Cognitive and Brain Sciences to carry out my own fMRI project (read about winning the Young Scientist Award and my experiences at the MPI here). Before arriving to the MPI, I gained some experience with functional magnetic resonance imaging (fMRI) by enrolling in Professor Joe Devlin’s Designing and Analysing fMRI Experiments module during my MSc degree at UCL. I was by no means under the illusion that I managed to fully understand the principles of fMRI through just this once course and the limited (nevertheless super exciting!) hands-on access to conducting an fMRI experiment and analysing the data – in fact, Prof Devlin would often tell us that even a full year of studying about this method might not be enough to get a truly thorough grasp of all the small details that have major impacts on the outcomes. Nevertheless, I was confident in my knowledge, so while I knew I would need to learn more during the process of data collection and analyses, I prepared a project proposal using fMRI for my MBB Young Scientist Award submission.
Designing an fMRI project
Little did I know at that point that my knowledge in truth was barely scratching the surface of all there is to know about fMRI study designs and analyses! For example, only after multiple meetings with the fMRI experts of the MPI and continuously reading books (*for fMRI beginners – these two books in particular were super helpful for me: xxx & xxx) and methodological papers was I able to make a decision about the most basic properties of the planned experiment; properties of the sequence such as the TR (repetition time) and the voxel size to be used. These seemingly small details each have more and less ideal values corresponding to them; a smaller TR means that you will end up with more images and thus perhaps a more accurate idea of the change in blood flow over time, while a smaller voxel size results in a higher quality image and thus a higher accuracy in the spatial properties of the changes in blood flow. And while one would perhaps want to take very small values on both properties to maximise both the spatial and the temporal resolution of the data, this might not be plausible – there is a general trade-off between TR and voxel size, and a small TR may mean that one cannot set a voxel size as small as might be desired. It is thus clear that how one sets up an fMRI study is not simply based on rules of thumb; each experiment must be carefully considered with regards to how one can maximise getting information about the specific research question at hand.
In my experiment, I ended up choosing a smaller TR, which, as the fMRI experts of the MPI explained to me, in my case might be more important than having very small voxel sizes. The reason for this is that fMRI experiments using Cyberball generally include only 3 blocks. Because this is relatively low compared to the number of blocks one would ideally use (nevertheless necessary in this case to avoid participant fatigue, boredom, etc.), a low TR ensures that in the end we still get a maximum number of data points. Of course, this is just one of the numerous details that one must carefully consider, and that are can become very overwhelming for someone just starting their fMRI journey. In addition to ensuring that the sequence is set up correctly, it is important to also carefully think through the actual length of the experiment and how much time the participants will be asked to spend inside the scanner, the order of the various sequences in cases multiple are used (e.g., task fMRI and resting-state fMRI), or whether a block or event-related design is more suitable to answer the research question.
Analysing fMRI data
You are done with the planning of all little details of your experiment. The ethical committee approved your project. You manage to deal with all things going wrong in the last minute despite weeks and months of planning and preparation. You spend weeks in the lab collecting data, including weekends and public holidays. Data collection is done! You survived! Surely, after all of this stress, the analyses will be like a holiday! Right?
I am not sure how other scientists might feel about this, but coming from a background in psychology and dealing with primarily behavioural data most of the time, this was my thought process at the time. Sure, I remembered that my little experience of analysing fMRI data previously was incredibly tedious, but I still managed to do it with not too much trouble! Of course, that was a very small sample, and I had a lot of help during the analyses, and was really only expected to investigate very basic differences between left- and right-finger tapping, something those familiar with fMRI will surely know very reliably produces a strong significant difference between the left and right motor cortex. So my expectations of what fMRI data analyses will look like (a bit of hassle, kind of tedious, it will take a few days, maybe even a week or at most two, but overall kind of straight forward) was nowhere near my experience of it (simultaneously finding not enough information on how to perform the analyses and too much information, with each paper/book chapter/expert you talk to seeming to have a very strong opinion about what is the only way to do it correctly, yet each of them describing little-to-very different methods; feeling completely lost; and ending up spending months looking at the data and running everything multiple times; oh and did I mentione you probably will end up learning both Bash scripting and Matlab scripting (or whatever your choice of torture is) in order to be able to deal with your large sample that you used to be so happy to be able to collect, although now you honestly kind of regret ever wanting to do anything besides a case study (anything over maybe 6-8 participants is just not feasible to analyse by hand)).
I am of course by no means trying to say that doing an fMRI project is not worth it – it definitely is! – but when you begin it is important to be prepared for what you are getting yourself into, so you don’t lose motivation when you are facing difficulties. Before even really beginning the analyses, you will have to pre-process your data, and this may take multiple days. Moreover, there are a lot of ways you can do this, with having many different options to take in multiple steps of the preprocessing (e.g., what mm FMWH smoothing kernel you use, or if you even decide to use one or not), which will in the end all have an effect on the data you are about to analyse.
You should also be prepared for criticism. Different scientists have different ideas about the best ways to pre-process the data, and if you do something slightly differently, they will most likely say so. This is completely okay, but make sure that you understand why you chose certain parameters, so you are able to defend your analyses! Besides the various parameters, there are also multiple software packages or scripts existing to aid pre-processing. Using FSL or SPM might be the most popular, and you are sure to find some online tutorials on these. For my project, I chose to use fMRI prep, published recently in Nature Neuroscience (link to article), developed by Russel Poldrack’s lab. FMRI prep combines various parts of different softwares, creating a very advanced preprocessing pipeline.
Congratulations! You are now able to begin the analyses! You will most probably have two steps here: First-level analyses, conducted individually for each of your participants, and group-level analyses. You may be fine with using one of the user-friendly, popular softwares (e.g., FSL; SPM), but depending on your research question, you might be once again looking at weeks of reading, coding, and running scripts – you may need to use specific toolboxes for finding beta values that you can use in mediation analyses or create unique masks for region-of-interest (ROI) analyses.
In sum – designing and analysing fMRI experiments is a challenging, time-consuming, and often frustrating process. So is it still worth it? YES! Although it undeniably does get tough at times, it is also very exciting to learn this method, it holds the capacity to answer research questions which you would otherwise not be able to investigate, and, to be completely honest, it is a lot of fun to look brain scans!! I hope that over time and with practice and experience, like most things, this also gets easier, and that it will not always be as difficult as the first time!