People have done all kinds of wacky things in the name of science. Thankfully, the events of this story are a bit less drastic than that time a scientist fed birds rice to see if they would explode, or when the discoverer of LSD took a pretty hefty dose to test it out, culminating in a very surreal cycle ride home. But when one researcher entered his local grocery store to purchase a large Atlantic salmon, with the intention of performing complex brain scans on the deceased, who could have known it would become a key part of a very important scientific debate?
A new way of looking at the brain
First, a little background. Functional magnetic resonance imaging, or fMRI, is a technique that opened up a whole new world of possibilities in brain research. Developed in the 1990s, it allows scientists to measure blood oxygenation in different parts of the brain while a person is performing a task or is otherwise exposed to some kind of stimulus.
The theory behind it is that brain areas with higher levels of activity are going to have a higher demand for oxygen, and thus more blood will flow to these areas.
So, you lie inside a big tube for an hour or so, pushing a few buttons and listening to the soothing voice of the researcher through the com system whilst trying desperately not to move your head, and Bob’s your uncle: the scientists now have a 3D map of your brain, colored to show where your neurons were firing most.
Because the technique is noninvasive and safe for most people, it’s become a mainstay of neuroscience research. It’s helped change our view of how the brain works; challenged our misconceptions; and in the future, it could even be combined with AI technology to read our thoughts.
But there was a problem. And we’re sorry to inform you that that problem involves math.
The issue of multiple comparisons
Scientists aren’t out here scanning people’s brains for the fun of it – they’re trying to answer a question, and to do that you need to perform some statistical analysis. If you have enough data, and run it through enough statistical tests, random chance dictates that you will eventually find the result you’re looking for. But how do you limit the risk that what you’re ending up with is a false positive?
Statisticians have come up with a number of different methods for this, which are collectively known as multiple comparisons correction. fMRI experiments are an example of a scientific study that generates lots of data, which can be put through lots of statistical tests to compare all the different variables. Within that you’re bound to get some false positives, which is a problem; but correct too harshly, and you’ll end up with false negatives, which is also a problem.
Something that began to concern lead study author Craig Bennett, and the catalyst for the whole palaver with the salmon which we promise we’re getting to in a minute, was that no one could seem to agree on whether, and how, multiple comparisons correction could best be applied to fMRI studies.
“We should scan a whole fish”
As promised, it’s salmon story time.
Bennett details how the salmon scanning came to be in a blog post, explaining how as a grad student at Dartmouth, he and some colleagues wanted to find some interesting objects to test out the new MRI protocols they were developing.
They started with a pumpkin, and then made the jump straight to dead animals with a Cornish game hen before they really decided to up the ante, with one team member declaring, “We should scan a whole fish.”
“I picked up the salmon from our local supermarket early on [a] Saturday morning in spring of 2005,” Bennett writes. “The clerk behind the counter was a little shocked to be selling a full-length Atlantic salmon at 6:30 AM, especially when I told her what was about it happen to it.” A fair reaction, all things considered.
The unfortunate fish was scanned, whilst being asked to perform a task that involved determining the emotions in a series of images of human faces. Having performed inadequately at this task, owing to the fact that it was a salmon and also dead, it was then reportedly consumed, the data filed away for another day.
That day came in 2008, when a conversation about the problem of multiple comparisons correction led Bennett to take another look at the salmon scans.
“I ran the fish data through my [statistical] processing pipelines and couldn’t believe what I saw,” writes Bennett. “Sure, there were some false positives. [...] Rather, it was where the false positives occurred that really floored me. A cluster of three significant voxels [data points] were arranged together right along the midline of the salmon’s brain.”
Remember that the salmon was being shown images while it was in the scanner? Well, according to the data Bennett was looking at, the salmon (which was dead, we can’t stress this enough) was actually thinking about those images.
Bennett and colleagues didn’t publish their findings right away, but after a lot of back and forth about the potential importance of this story to the wider debate around multiple comparisons correction, they did present the work as a poster at a conference – although, as Bennett notes, “Just about everyone thought it was a joke.”
Once people got to see the poster for themselves, however, its significance became clear. Soon afterward, a full paper was published in the fabulously named Journal of Serendipitous and Unexpected Results, and we can promise you it’s worth a read.
Where are we now?
In the years since the dead salmon paper, the debate around how best to report and interpret fMRI data has, if anything, become even more heated.
As highlighted in one article published in the journal Brain, the point of Bennett and colleagues’ study “was never to invalidate functional MRI.” The technique itself is not at fault, but that doesn’t mean scientists shouldn't have a serious conversation about how to deal with the data it generates.
fMRI studies continue to generate fascinating findings – many of which happily come from living animals – and are a key part of the neuroscientist’s toolbox. Hilarity notwithstanding, the dead salmon debacle is an important episode in the development of the field, even bagging the team an Ig Nobel Prize in 2012.
As Bennett himself summarized, “The more I think about the affair the more I believe that the fish has the chance to impact the field of neuroimaging in a very positive way.”
We’re sure the salmon would be delighted with this legacy.