Full Citation: Meidenbauer, K. L., Choe, K. W., Cardenas-Iniguez, C., Huppert, T. J., & Berman, M.G. (2021). Load-dependent relationships between frontal fNIRS activity and performance: A data-driven PLS approach. NeuroImage, 230, 117795.
Open access link: https://www.sciencedirect.com/science/article/pii/S1053811921000720?via%3Dihub
Meta-Note: One thing we’re planning on using this blog space for is to describe our recently published research in a way that makes sense to those not immersed in our field and want to know “why” this research matters. So here, we’ll be discussing not just what the studies did and found, but also how this work fits into the big picture of the ENL’s research goals and the research programs of the primary author(s).
One of the most-used tools for examining brain activity while humans engage in psychological tasks is that of function magnetic resonance imaging, or fMRI. fMRI can provide a lot of incredibly useful information, and you can see brain activity with very good spatial resolution – at the level of millimeters! But MRI has limitations. Most notably for the purpose of looking the effects of the environment – you can’t conduct fMRI studies outside of the scanner, so examining brain activity in real environments is off the table. This is where fNIRS, or functional near-infrared spectroscopy, becomes really helpful. fNIRS measures the same biological signal as fMRI but uses a different approach that allows activity on the outermost surface of the brain (cortex) to be measured using a cap equipped with LEDs and light detectors. fNIRS is used a lot in studies which require brain activity to be monitored in unconventional settings (such as outdoors), while people engage in physical activity, and in populations where getting someone to sit still in the MRI would be a challenge, like kids.
But fNIRS has been used much less than fMRI, so there’s still some work to be done on ensuring that the data collection, analysis, and results are robust and reliable. There have been rapid improvements in both the devices themselves (i.e. the hardware) and in analysis (i.e., the software) in recent years, so our first goal upon getting this fNIRS device was to make sure that in using this improved hardware and more rigorous methods, we could replicate what’s been shown in previous fNIRS and fMRI studies, and feel confident in this new-to-us neuroimaging technique.
One of the most well-established effects in cognitive neuroscience has to do with increased activity in the prefrontal cortex and the parietal cortex of the brain as a task gets more demanding. A typical task that elicits this type of response is the N-back task, where people are asked to compare a word (or image or other stimulus) to another word presented “N” trials back. And by manipulating “N”, we can make the task more or less difficult. If you have to remember a word 1 trial back (where N = 1) it’s easier than if you have to remember a word 3 trials back (where N = 3). So, our initial goal of this study was to have a large sample of individuals (70 adults) come in and do this N-back task (where N was 1, 2, or 3) while we recorded fNIRS activity.
We found that, consistent with a lot of previous research in fNIRS and fMRI, performing the 2-back task versus the 1-back task led to more activity across the entire area of the brain we covered with our fNIRS cap – in bilateral (both sides of) frontal cortex and right parietal cortex. These areas are heavily implicated in working memory and sustained attention, so having more activity in these frontal and parietal cortical areas when people are doing a task that requires more memory and concentration is exactly what we’d expect.
Unexpectedly, participants didn’t show reliably increased neural activity in these regions when they completed a 3-back task (the hardest task). We thought that this might have to do with the fact that not everyone was able to do this (admittedly VERY difficult) task accurately. That is, if people are having a hard time with the task, they might simply be guessing or not as invested in trying to focus all their attention on doing the task. So, if people aren’t trying as hard or aren’t able to exert any more neural “output” to perform well on the 3-back task, then you might not expect to see more activity.
So, we decided to test that hypothesis. We wondered if how well people performed on the task (particularly the 3-back task) was influencing the brain activation patterns in our data. There are a lot of ways you could test this, but one approach we take a lot in the Environmental Neuroscience Lab, is that of data-driven, multivariate analyses. In this case, we looked at an analysis called behavioral partial least squares analysis (behavioral PLS). Without getting too technical, there are a few key strengths of this type of approach in neuroimaging work. Here, we want to examine whether the relationship between task performance (accuracy) and brain activity (across over 40 locations) is the same or different when a task is relatively easy versus when it’s difficult. But with most standard (univariate) analyses, we would need to simplify things first before running any kind of statistics. For example, we might first approach this by grouping some participants into “high” and “low” performers. We’d also need to decide how we want to deal with the 40+ brain locations we are dealing with. We could look at all of the locations individually or we could group some together based on where they are on the surface of the brain. And by doing these simplifications (which could be done in a LOT of different ways), we might miss some interesting effects that can only be reliably detected by looking at all of the data together. Multivariate techniques like PLS allow us to do just that – by examining whether there is interesting and meaningful structure in how task performance, task difficulty, and brain activity relate to one another.
And by doing this approach, we found that indeed, the link between brain activity and task performance was different depending on how hard the task was. The participants in our study were the most accurate on the 1-back task (the easiest one) if the amount of neural activity in the prefrontal cortex was relatively low, and they were the most accurate on the 3-back task (the hardest one) when activity in this area was high. In other words, these results suggest that if your brain is more efficient (or doing the task is more automatic) when the task is easy, you’ll do better on it. But on a hard task, you need to be able to recruit these prefrontal brain regions to perform well.
We were the first researchers to use these analyses in fNIRS and we think that this work shows how useful multivariate approaches can be in this context. So, we shared all of our data, our analysis code, our experiment code, and our results publicly in the hopes that other scientists can use this method in their own fNIRS work. (For all you fellow fNIRS nerds, here’s that link: https://osf.io/sh2bf/)
Circling back to the *original* goal of this work, we’re feeling pretty confident about using the technique in the lab, and once we can do in-person research again, we hope to be able to take our fNIRS device out of the lab and into some real environments. For example, we could record fNIRS activity while people go for a walk in a park versus along a city street, or have people complete cognitive tasks outdoors in areas that vary in environmental factors we care about, such as naturalness or temperature. In the meantime, we’re looking at examining whether we can use scale-invariance of fNIRS data as an index of task difficulty. This is a measure the lab has used in fMRI and EEG previously (check out this paper from 2016, this paper from 2020, and this preprint for a few examples), so we’re interested in seeing whether we can also extract this scale-invariant signal in our fNIRS data. Stay tuned 😊
Kim Lewis Meidenbauer